Tag Archives: Amazon Simple Queue Service (SQS)

Managing backend requests and frontend notifications in serverless web apps

Post Syndicated from James Beswick original https://aws.amazon.com/blogs/compute/managing-backend-requests-and-frontend-notifications-in-serverless-web-apps/

Web and mobile applications usually interact with a backend service, often via an API. Many front-end applications pass requests for processing, wait for a result, and then display this to the user. This synchronous approach is only one way to handle messages, but modern applications have alternatives to provide a better user experience.

There are three common ways to make and manage requests from your frontend. This blog post explains the benefits and use-cases of each approach. This post references the Ask Around Me example application, which allows users to ask and answer questions in their local geographic area in real time. To learn more, refer to part 1 of the blog application series.

The synchronous model

The synchronous call is the most common API request pattern, where the caller makes a request to an API and then waits for the response:

Synchronous API example

This type of request is easy to implement and understand because it mirrors the functional call-response pattern many developers are familiar with. The requestor is blocked until the calls completes, so it is well suited to simple requests with short execution times. Use cases include: retrieving the contents of a shopping cart, looking up a value in a database, or submitting an email address from a web form.

However, when your API interacts with other services or external workflows, the synchronous model can have limitations. In the following ecommerce example, any slow performance in the downstream services delays the entire roundtrip performance. Additionally, any outages in one of those services may result in an apparent failure of the entire service.

Tight coupling of services

For services with lengthy workflows, you may reach API Gateway’s 29-second integration timeout. In the following ride sharing example, the service responsible for finding available drivers may have a highly variable response time. This request may time out. It also provides a poor user experience, as there is no feedback to the user for a considerable period.

Lengthy request example

Synchronous requests also have others limitations. You cannot receive more than one response per request, nor can you subscribe to future changes in data. In this example, the API request can only inform callers about drivers at the end of the length process request.

The asynchronous model

Asynchronous tasks are common in serverless applications and distributed applications. They allow separate parts of an application to communicate without needing to wait for a synchronous response. Asynchronous workloads often use queues between services to help manage throughout and assist with retry logic.

With asynchronous tasks, the caller hands off the event and continues on to the next task after receiving the acknowledgment response. The caller does not wait for the entire task to complete. The downstream service works on this event while the caller continues servicing other requests. The ecommerce example, converted to an asynchronous flow, looks like this:

Asynchronous request example

In this example, a caller submits an order and receives a response from API Gateway almost immediately. With service integrations, API Gateway then stores the request directly in a durable store such as Amazon SQS or DynamoDB, before any processing has occurred. This results in a relatively consistent caller response time, regardless of downstream service processing time.

The downstream services fetch messages from the SQS queue or the DynamoDB table for processing. If there is a downstream outage, messages are persisted in the queue and may be retried later. From the user’s perspective, the request has been successfully submitted.

The Ask Around Me application handles the publishing of both questions and answers asynchronously. The API passes the user data to a Lambda function that stores the message in an SQS queue. SQS responds immediately to indicate that the message has been stored successfully, ending the API response. Another Lambda function then takes the messages from the SQS queue and processes these independently.

Ask Around Me save question example

Both synchronous and asynchronous requests are useful for different functions in web applications, so it can be helpful to compare their features and behaviors:

Synchronous requestsAsynchronous requests
The caller waits until the end of processing for a response.The caller receives an acknowledgment quickly while processing continues.
Waiting may incur cost.Minimizes the cost of waiting.
Downstream slowness or outages affects the overall request.Queuing separates ingestion of the request from the processing of the request.
Passes payloads between steps.More often passes transaction identifiers.
Failure affects entire request.Failure only affects segment of request.
Easy to implement.Moderate complexity in implementation

Handling response values and state for asynchronous requests

With asynchronous processes, you cannot pass a return value back to the caller in the same way as you can for synchronous processes. Beyond the initial acknowledgment that the request has been received, there is no return path to provide further information. There are a couple of options available to web and mobile developers to track the state of inflight requests:

  • Polling: the initial request returns a tracking identifier. You create a second API endpoint for the frontend to check the status of the request, referencing the tracking ID. Use DynamoDB or another data store to track the state of the request.
  • WebSocket: this is a bidirectional connection between the frontend client and the backend service. It allows you to send additional information after the initial request is completed. Your backend services can continue to send data back to the client by using a WebSocket connection.

Polling is a simple mechanism to implement for many systems but can result in many empty calls. There is also a delay between data availability and the client being notified. WebSockets provide notifications that are closer to real time and reduce the number of messages between the client and backend system. However, implementing WebSockets is often more complex.

Using AWS IoT Core for real-time messaging

In both the synchronous and asynchronous models, it’s assumed that the caller makes a request and is only interested in the final result of that request. This doesn’t allow for partial information, such as the percentage of a task complete, or being notified continuously as data changes.

Modern web applications commonly use the publish-subscribe pattern to receive notifications as data changes. From receiving alerts when new email arrives to providing dashboard analytics, this method allows for much richer streams of event from backend systems.

In Ask Around Me, the application uses this pattern when listening for new questions from the user’s local area. The frontend subscribes to the geohash value of the user’s location via the AWS SDK. It then waits for messages published by the backend to this topic.

AWS IoT Core between frontend and backend

The SDK automatically manages the WebSocket connection and also handles many common connectivity issues in web apps. The messages are categorized using topics, which are strings defining channels of messages.

The AWS IoT Core service manages broadcasts between backend publishers and frontend subscribers. This enables fan-out functionality, which occurs when multiple subscribers are listening to the same topic. You can broadcast messages to thousands of frontend devices using this mechanism. For web application integration, this is the preferable way to implement publish-subscribe than using Amazon SNS.

The IotData class in the AWS SDK returns a client that uses the MQTT protocol. Once the frontend application establishes the connection, it returns messages, errors, and the connection status via callbacks:

        mqttClient.on('connect', function () {
          console.log('mqttClient connected')
        })

        mqttClient.on('error', function (err) {
          console.log('mqttClient error: ', err)
        })

        mqttClient.on('message', function (topic, payload) {
          const msg = JSON.parse(payload.toString())
          console.log('IoT msg: ', topic, msg)
        })

For more details on how to implement MQTT WebSocket connectivity for your application, see the Ask Around Me sample application code.

Combining multiple approaches for your frontend application

Many frontend applications can combine these models depending on the request type. The Ask Around Me application uses multiple approaches in managing the state of user questions:

Combining multiple models in one application

  1. When the application starts, it retrieves an initial set of questions from the synchronous API endpoint. This returns the available list of questions up to this point in time.
  2. Simultaneously, the frontend subscribes to the geohash topic via AWS IoT Core. Any new questions for this geohash location are sent from the backend processing service to the frontend via this topic. This allows the frontend to receive new questions without subsequent API calls.
  3. When a new question is posted, it is saved to the relevant SQS queue and acknowledged. The question is processed asynchronously by a backend process, which sends updates to the topic.

There are several benefits to combining synchronous, asynchronous, and real-time messaging approaches like this. Most importantly, the user experience remains consistent. The user receives immediate feedback to posting new questions and answers, while longer-running processes are managed asynchronously.

When new information becomes available, the frontend is notified in near-real time. This happens without needing to poll an API endpoint or have the user refresh the user interface. This also reduces the number of unnecessary API calls on the backend service, reducing the cost of running this application. Finally, this uses scalable managed services so the frontend application can support large numbers of users without impacting performance.

Conclusion

Web applications commonly use synchronous APIs when communicating with backend services. For longer-running processes, asynchronous workflows can offer an improved user experience and help manage scaling. By using durable message stores like SQS or DynamoDB, you can separate the request ingestion and response from the request processing.

In this post, I show how modern web applications use real-time messaging via WebSockets to improve the user experience. This provides a transport mechanism for pushing state updates from the backend to the frontend client. The AWS IoT Core service can fan out messages using topics, broadcasting messages to large numbers of frontend subscribers.

To see these three methods in an example frontend application, read more about the Ask Around Me example application.

Building a location-based, scalable, serverless web app – part 3

Post Syndicated from James Beswick original https://aws.amazon.com/blogs/compute/building-a-location-based-scalable-serverless-web-app-part-3/

In part 2, I cover the API configuration, geohashing algorithm, and real-time messaging architecture used in the Ask Around Me web application. These are needed for receiving and processing questions and answers, and sending results back to users in real time.

In this post, I explain the backend processing architecture, how data is aggregated, and how to deploy the final application to production. The code and instructions for this application are available in the GitHub repo.

Processing questions

The frontend sends new user questions to the backend via the POST questions API. While the predicted volume of questions is only 1,000 per hour, it’s possible for usage to spike unexpectedly. To help handle this load, the PostQuestions Lambda function puts incoming questions onto an Amazon SQS queue. The ProcessQuestions function takes messages from the Questions queue in batches of 10, and loads these into the Questions table in Amazon DynamoDB.

Questions processing architecture

This asynchronous process smooths out traffic spikes, ensuring that the application is not throttled by DynamoDB. It also provides consistent response times to the front-end POST request, since the API call returns as soon as the message is durably persisted to the queue.

Currently, the ProcessQuestions function does not parse or validate user questions. It would be easy to add message filtering at this stage, using Amazon Comprehend to detect sentiment or inappropriate language. These changes would increase the processing time per question, but by handling this asynchronously, the initial POST API latency is not adversely affected.

The ProcessQuestions function uses the Geo Library for Amazon DynamoDB that converts the question’s latitude and longitude into a geohash. This geohash attribute is one of the indexes in the underlying DynamoDB table. The GetQuestions function using the same library for efficiently querying questions based on proximity to the user.

There are a couple of different mechanisms used to pass information between the frontend and backend applications. When the frontend first initializes, it retrieves the current location of the user from the browser. It then calls the questions API to get a list of active questions within 5 miles of the current location. This retrieves the state up to this point in time. To receive notifications of new messages posted in the user’s area, the frontend also subscribes to the geohash topic in AWS IoT Core.

Processing answers

Answers processing architecture

The application allows two types of question that have different answer types. First, the rating questions accept an answer with a 0–5 score range. Second, the geography questions accept a geo-point, which is a latitude and longitude representing a location.

Similar to the way questions are handled, answers are also queued before processing. However, the PostAnswers Lambda function sends answers to different queues, depending on question type. Ratings messages are sent to the StarAnswers queue, while geography messages are routed to the GeoAnswers queue. Star ratings are saved as raw data in the Answers table by the ProcessAnswerStar function. Geography answers are first converted to a geohash before they are stored.

It’s possible for users to submit updates to their answers. For a star rating, the processing function simply saves the new score. For geography answers, if the updated answer contains a latitude and longitude close enough to the original answer, it results in the same geohash. This is due to the different aggregation processes used for these types of answers.

Aggregating data

In this application, the users asking questions are seeking aggregated answers instead of raw data. For example, “How do you rate the park?” shows an average score from users instead of thousands of individual ratings. To maintain performance, this aggregation occurs when new answers are saved to the database, not when the application fetches the question list.

The Answers table emits updates to a DynamoDB stream whenever new items are inserted or updated. The StreamSpecification parameter in the table definition is set to NEW_AND_OLD_IMAGES, meaning the stream record contains both the new and old item record.

New answers to questions are new items in the table, so the stream record only contains the new image. If users update their answers, this creates an updated item in the table, and the stream record contains both the new and old images of the item.

For star ratings, when receiving an updated rating, the Aggregation function uses both images to calculate the delta in the score. For example, if the old rating was 2 and the user changes this to 5, then the delta is 3. The summary score related to the answer is updated in the Questions table, using a DynamoDB update expression:

    const result = await myGeoTableManager.updatePoint({
      RangeKeyValue: { S: update }, 
      GeoPoint: {
        latitude: item.lat,
        longitude: item.lng
      },
      UpdateItemInput: {
        UpdateExpression: 'ADD answers :deltaAnswers, totalScore :deltaTotalScore',
        ExpressionAttributeValues: {
          ':deltaAnswers': { N: item.deltaAnswers.toString()},
          ':deltaTotalScore': { N: item.deltaValue.toString()}
        }
      }
    }).promise()

For geo-point ratings, the same approach is used but if the geohash changes, then the delta is -1 for the geohash in the old image, and +1 for the geohash in the new image. The update expression automatically creates a new geohash attribute on the DynamoDB item if it is not already present:

    const result = await myGeoTableManager.updatePoint({
      RangeKeyValue: { S: item.ID }, 
      GeoPoint: {
        latitude: item.lat,
        longitude: item.lng
      },
      UpdateItemInput: {
        UpdateExpression: `ADD ${item.geohash} :deltaAnswers, answers :deltaAnswers`,
        ExpressionAttributeValues: {
          ':deltaAnswers': { N: item.deltaAnswers.toString() }          
        }
      }
    }).promise()

By using a Lambda function as a DynamoDB stream processor, you can aggregate large amounts of data in near real time. The Questions and Answers tables have a one-to-many relationship – many answers belong to one question. As answers are saved, the aggregation process updates the summaries in the Questions table.

The Questions table also publishes updates to another DynamoDB stream. These are consumed by a Lambda function that sends the aggregated update to topics in AWS IoT Core. This is how updated scores are sent back to the frontend client application.

Publishing to production with Amplify Console

At this point, you can run the application on your local development machine and view the application via the localhost Vue.js server. Once you are ready to launch the application to users, you must deploy to production.

Single-page applications are easy to deploy publicly. The build process creates static HTML, JS, and CSS files. These can be served via Amazon S3 and Amazon CloudFront, together with any image and media assets used. The process of running the build process and managing the deployment can be automated using AWS Amplify Console.

In this walk through, I use GitHub as the repo provider. You can also use AWS CodeCommit, Bitbucket, GitLab, or upload the build directory from your machine.

To deploy the front end via Amplify Console:

  1. From the AWS Management Console, select the Services dropdown and choose AWS Amplify. From the initial splash screen, choose Get Started under Deploy.Amplify Console getting started
  2. Select GitHub as the repository provider, then choose Continue:Select GitHub as your code repo
  3. Follow the prompts to enable GitHub access, then select the repository dropdown and choose the repo. In the Branch dropdown, choose master. Choose Next.Add repository branch
  4. In the App build and test settings page, choose Next.
  5. In the Review page, choose Save and deploy.
  6. The final screen shows the deployment pipeline for the connected repo, starting at the Provision phase:Amplify Console deployment pipeline

After a few minutes, the Build, Deploy, and Verify steps show green checkmarks. Open the URL in a browser, and you see that the application is now served by the public URL:

Ask Around Me - Deployed application

Finally, before logging in, you must add the URL to the list of allowed URLs in the Auth0 settings:

  1. Log into Auth0 and navigate to the dashboard.
  2. Choose Applications in the menu, then select Ask Around Me from the list of applications.
  3. On the Settings tab, add the application’s URL to Allowed Callback URLs, Allowed Logout URLs, and Allowed Web Origins. Separate from the existing values using a comma.Updating the Auth0 configuration
  4. Choose Save changes. This allows the new published domain name to interact with Auth0 for authentication your application’s users.

Anytime you push changes to the code repository, Amplify Console detects the commit and redeploys the application. If errors are detected, the existing version is presented to users. If there are no errors, the new version is served to visitors.

Conclusion

In the last part of this series, I show how the application queues posted questions and answers. I explain how this asynchronous approach smooths traffic spikes and helps maintain responsive APIs.

I cover how answers are collected from thousands of users and are aggregated using DynamoDB streams. These totals are saved as summaries in the Questions table, and live updates are pushed via AWS IoT Core back to the frontend.

Finally, I show how you can automate deployment using Amplify Console. By connecting the service directly with your code repository, it publishes and serves your application with no need to manually copy files.

To learn more about this application, see the accompanying GitHub repo.

Building scalable serverless applications with Amazon S3 and AWS Lambda

Post Syndicated from James Beswick original https://aws.amazon.com/blogs/compute/building-scalable-serverless-applications-with-amazon-s3-and-aws-lambda/

Well-designed serverless applications are typically a combination of managed services connected by custom business logic. One of the most powerful combinations for enterprise application development is Amazon S3 and AWS Lambda. S3 is a highly durable, highly available object store that scales to meet your storage needs. Lambda runs custom code in response to events, automatically scaling with the size of the workload. When you use the two services together, they can provide a scalable core for serverless solutions.

This blog post shows how to design and deploy serverless applications designed around S3 events. The solutions presented use AWS services to create scalable serverless architectures, using minimal custom code. This is the conclusion of a series showing how the S3-to-Lambda pattern can implement the following business solutions:

Bringing the compute layer to the data

Much traditional software operates by bringing data to the compute layer. This means that processes run on batches of data in files, databases, and other sources. This is inherently harder to scale as data volumes grow, often needing a fleet of servers to scale out at peak times. For the developer, this creates operational overhead to ensure that the compute capacity is keeping pace with the data volume.

The S3-to-Lambda serverless pattern instead brings the compute layer to the data. As data arrives, the compute process scales up and down automatically to meet the demand. This allows developers to focus on building business logic for a single item of data, and the execution at scale is handled by the Lambda service.

The image optimization application is a good example for comparing the traditional and serverless approaches. For a busy media site, capturing hundreds of images per minute in an S3 bucket, the operations overhead becomes clearer. A script running on a server must scale up across multiple instances to keep pace with this level of traffic. Compare this to the Lambda-based approach, which scales on-demand. The code itself does not change, whether it is used for a single image or thousands.

Receiving and processing events from S3 in custom code

S3 raises events when objects are put, copied, or deleted in a bucket. It also raises a broad number of other notifications, such as when lifecycle events occur. You can configure S3 to invoke Lambda from these events by using the S3 console, Lambda console, AWS CLI, or AWS Serverless Application Model (SAM) templates.

S3 passes details of the event, not the object itself, to the Lambda function in a JSON object. This object contains an array of records, so it’s possible to receive more than one S3 event per invocation:

S3 passes event details to Lambda

As the Lambda handler may receive more than one record, it should iterate through the records collection. It’s best practice to keep the handler small and generic, calling out to the business logic in a separate function or file:

const processEvent = require('my-custom-logic’)

// A Node.js Lambda handler
exports.handler = async (event) => {

  // Capture event – can be used to create mock events
  console.log (JSON.stringify(event, null, 2))  

  // Handle each incoming S3 object in the event
  await Promise.all(
    event.Records.map(async (event) => {
      try {
	  // Pass each event to the business logic handler
        await processEvent(event)
      } catch (err) {
        console.error('Handler error: ', err)
      }
    })
  )
}

This code example takes advantage of concurrent asynchronous executions available in Node.js but similar constructs are available in many other languages. This means that multiple objects are processed in parallel to minimize the overall function execution time.

Instead of handling and logging any errors within the function’s code, it’s also possible to use destinations for asynchronous invocations. You use an On failure condition to route the error to various potential targets, including another Lambda function or other AWS services. For complex applications or those handling large volumes, this provides greater control for managing events that fail processing.

During the development process, you can debug and test the S3-to-Lambda integration locally. First, capture a sample event during development to create a mock event for local testing. The sample applications in this series each use a test harness so the developer can test the handler on a local machine. The test harness invokes the handler locally, providing mock environment variables:

// Mock event
const event = require('./localTestEvent')

// Mock environment variables
process.env.AWS_REGION = 'us-east-1'
process.env.localTest = true
process.env.language = 'en'

// Lambda handler
const { handler } = require('./app')

const main = async () => {
  console.time('localTest')
  await handler(event)
  console.timeEnd('localTest')
}

main().catch(error => console.error(error))

Scaling up when more data arrives

The Lambda service scales up if S3 sends multiple events simultaneously. How this works depends on several factors. If the target Lambda function has sufficient concurrency available, and if any active instances of the function are already processing events, the Lambda service scales up.

Lambda scaling up as events queue grows

The function does not scale up if the reserved concurrency is set to 1 or the scaling capacity is fully consumed for a Region in your account. In this case, the events from S3 are queued internally until a Lambda instance is available for processing. You can request to increase the regional concurrency limit by submitting a request in the Support Center console. You may also intend to perform one-at-a-time processing by setting the reserved concurrency to 1.

One-at-a-time processing with Lambda

Generally, multiple instances of a function are invoked simultaneously when S3 receives multiple objects, to process the events as quickly as possible. It’s this rapid scaling and parallelization in both S3 and Lambda that make this pattern such a powerful core architecture for many applications.

Amazon SNS and Amazon SQS integrations

The native S3 to Lambda integration provides a reliable way to invoke one function per prefix or suffix-pattern per bucket. In example, invoking a function when object keys end in .pdf in a single bucket. This works well for the vast majority of use-cases but you may want to invoke multiple Lambda functions per S3 event.

In this case, S3 can publish notifications to SNS, where events are delivered to a range of targets. These include Lambda functions, SQS queues, HTTP endpoints, email, text messages and push notifications. SNS provides fan-out capability, enabling one event to be delivered to multiple destinations, such as Lambda functions or web hooks, for example.

In busy applications, the volume of S3 events may be too large for a downstream system, such as a non-serverless service. In this case, you can also use an SQS queue as a notification target. After events are published to a queue, they can be consumed by Lambda functions and other services. The queue acts as a buffer and can help smooth out traffic for systems consuming these events. See the DynamoDB importer repository for an example.

Uploading data to S3 in upstream applications

You may have upstream services in your architecture that generate the data stored in S3. Some upstream workloads have spiky usage patterns and large numbers of users, like web or mobile applications. You may increase the performance and throughput of these workloads by uploading directly to S3. This avoids proxying binary data through an API Gateway endpoint or web server.

For example, for a mobile application uploading user photos, S3 and Lambda can handle the upload process for large of numbers of users:

  1. The upstream process, in this case a mobile client, requests a presigned URL from an API Gateway endpoint.
  2. This invokes a Lambda function that requests a presigned URL for the S3 bucket, and returns this back via the API call.
  3. The mobile client sends the data directly to the presigned S3 URL using HTTPS POST. The upload is managed directly by S3.

This simple pattern can be a scalable and cost-effective way to upload large binary data into your applications. After the object successfully uploads, the S3 put event can then asynchronously invoke downstream workflows.

Visit this repository to see an example of a serverless S3 uploader application. You can also see a walkthrough of this process in this YouTube video.

Developing larger applications

As you develop larger serverless applications, it often becomes more practical to split applications into multiple services and repositories for separate teams. Often, individual services must integrate with existing S3 buckets, not create these in the application templates. You may also have to integrate a single service with multiple S3 buckets.

In decoupling larger applications with Amazon EventBridge, I show how you can decouple services within an application using an event bus. This pattern helps separate the producers and consumers of events in your workload. This can make each service become more independent and more resilient to changes with the overall application.

This example demonstrates how the document repository solution can be refactored into several smaller applications that communicate using events. This uses Amazon EventBridge as the event router coordinating the flow. Each application contains a SAM template that defines the EventBridge rule to filter for events, and publishes data back to the event bus after processing is complete.

One of the major benefits to using an event-based architecture is that development teams retain flexibility even as the application grows. It allows developers to separate AWS resources like S3 buckets and DynamoDB tables, from the compute resources, like Lambda functions. This decoupling can simplify the deployment process, help avoid building monoliths, and reduce the cognitive load of developing in large applications.

Conclusion

S3 and Lambda are two highly scalable AWS services that can be powerful when combined in serverless applications. In this post, I summarize many of the patterns shown across this series. I explain the integration pattern and the scaling behavior, and how you can use mock events for local testing and development. You can also use SNS and SQS in some applications for fan-out and buffering of events.

Upstream applications can upload data directly to S3 to achieve greater scalability by avoiding proxies. For larger applications, I show how using an event-based architecture modeled around EventBridge can help decouple application services. This can promote service independence, and help maintain flexibility as applications grow.

To learn more about the S3-to-Lambda architecture pattern, watch the YouTube video series, or explore the articles listed at top of this post.

Building a Scalable Document Pre-Processing Pipeline

Post Syndicated from Joel Knight original https://aws.amazon.com/blogs/architecture/building-a-scalable-document-pre-processing-pipeline/

In a recent customer engagement, Quantiphi, Inc., a member of the Amazon Web Services Partner Network, built a solution capable of pre-processing tens of millions of PDF documents before sending them for inference by a machine learning (ML) model. While the customer’s use case—and hence the ML model—was very specific to their needs, the pipeline that does the pre-processing of documents is reusable for a wide array of document processing workloads. This post will walk you through the pre-processing pipeline architecture.

Pre-processing pipeline architecture-SM

Architectural goals

Quantiphi established the following goals prior to starting:

  • Loose coupling to enable independent scaling of compute components, flexible selection of compute services, and agility as the customer’s requirements evolved.
  • Work backwards from business requirements when making decisions affecting scale and throughput and not simply because “fastest is best.” Scale components only where it makes sense and for maximum impact.
  •  Log everything at every stage to enable troubleshooting when something goes wrong, provide a detailed audit trail, and facilitate cost optimization exercises by identifying usage and load of every compute component in the architecture.

Document ingestion

The documents are initially stored in a staging bucket in Amazon Simple Storage Service (Amazon S3). The processing pipeline is kicked off when the “trigger” Amazon Lambda function is called. This Lambda function passes parameters such as the name of the staging S3 bucket and the path(s) within the bucket which are to be processed to the “ingestion app.”

The ingestion app is a simple application that runs a web service to enable triggering a batch and lists documents from the S3 bucket path(s) received via the web service. As the app processes the list of documents, it feeds the document path, S3 bucket name, and some additional metadata to the “ingest” Amazon Simple Queue Service (Amazon SQS) queue. The ingestion app also starts the audit trail for the document by writing a record to the Amazon Aurora database. As the document moves downstream, additional records are added to the database. Records are joined together by a unique ID and assigned to each document by the ingestion app and passed along throughout the pipeline.

Chunking the documents

In order to maximize grip and control, the architecture is built to submit single-page files to the ML model. This enables correlating an inference failure to a specific page instead of a whole document (which may be many pages long). It also makes identifying the location of features within the inference results an easier task. Since the documents being processed can have varied sizes, resolutions, and page count, a big part of the pre-processing pipeline is to chunk a document up into its component pages prior to sending it for inference.

The “chunking orchestrator” app repeatedly pulls a message from the ingest queue and retrieves the document named therein from the S3 bucket. The PDF document is then classified along two metrics:

  • File size
  • Number of pages

We use these metrics to determine which chunking queue the document is sent to:

  • Large: Greater than 10MB in size or greater than 10 pages
  • Small: Less than or equal to 10MB and less than or equal to 10 pages
  • Single page: Less than or equal to 10MB and exactly one page

Each of these queues is serviced by an appropriately sized compute service that breaks the document down into smaller pieces, and ultimately, into individual pages.

  • Amazon Elastic Cloud Compute (EC2) processes large documents primarily because of the high memory footprint needed to read large, multi-gigabyte PDF files into memory. The output from these workers are smaller PDF documents that are stored in Amazon S3. The name and location of these smaller documents is submitted to the “small documents” queue.
  • Small documents are processed by a Lambda function that decomposes the document into single pages that are stored in Amazon S3. The name and location of these single page files is sent to the “single page” queue.

The Dead Letter Queues (DLQs) are used to hold messages from their respective size queue which are not successfully processed. If messages start landing in the DLQs, it’s an indication that there is a problem in the pipeline. For example, if messages start landing in the “small” or “single page” DLQ, it could indicate that the Lambda function processing those respective queues has reached its maximum run time.

An Amazon CloudWatch Alarm monitors the depth of each DLQ. Upon seeing DLQ activity, a notification is sent via Amazon Simple Notification Service (Amazon SNS) so an administrator can then investigate and make adjustments such as tuning the sizing thresholds to ensure the Lambda functions can finish before reaching their maximum run time.

In order to ensure no documents are left behind in the active run, there is a failsafe in the form of an Amazon EC2 worker that retrieves and processes messages from the DLQs. This failsafe app breaks a PDF all the way down into individual pages and then does image conversion.

For documents that don’t fall into a DLQ, they make it to the “single page” queue. This queue drives each page through the “image conversion” Lambda function which converts the single page file from PDF to PNG format. These PNG files are stored in Amazon S3.

Sending for inference

At this point, the documents have been chunked up and are ready for inference.

When the single-page image files land in Amazon S3, an S3 Event Notification is fired which places a message in a “converted image” SQS queue which in turn triggers the “model endpoint” Lambda function. This function calls an API endpoint on an Amazon API Gateway that is fronting the Amazon SageMaker inference endpoint. Using API Gateway with SageMaker endpoints avoided throttling during Lambda function execution due to high volumes of concurrent calls to the Amazon SageMaker API. This pattern also resulted in a 2x inference throughput speedup. The Lambda function passes the document’s S3 bucket name and path to the API which in turn passes it to the auto scaling SageMaker endpoint. The function reads the inference results that are passed back from API Gateway and stores them in Amazon Aurora.

The inference results as well as all the telemetry collected as the document was processed can be queried from the Amazon Aurora database to build reports showing number of documents processed, number of documents with failures, and number of documents with or without whatever feature(s) the ML model is trained to look for.

Summary

This architecture is able to take PDF documents that range in size from single page up to thousands of pages or gigabytes in size, pre-process them into single page image files, and then send them for inference by a machine learning model. Once triggered, the pipeline is completely automated and is able to scale to tens of millions of pages per batch.

In keeping with the architectural goals of the project, Amazon SQS is used throughout in order to build a loosely coupled system which promotes agility, scalability, and resiliency. Loose coupling also enables a high degree of grip and control over the system making it easier to respond to changes in business needs as well as focusing tuning efforts for maximum impact. And with every compute component logging everything it does, the system provides a high degree of auditability and introspection which facilitates performance monitoring, and detailed cost optimization.

Creating a scalable serverless import process for Amazon DynamoDB

Post Syndicated from James Beswick original https://aws.amazon.com/blogs/compute/creating-a-scalable-serverless-import-process-for-amazon-dynamodb/

Amazon DynamoDB is a web-scale NoSQL database designed to provide low latency access to data. It’s well suited to many serverless applications as a primary data store, and fits into many common enterprise architectures. In this post, I show how you can import large amounts of data to DynamoDB using a serverless approach. This uses Amazon S3 as a staging area and AWS Lambda for the custom business logic.

This pattern is useful as a general import mechanism into DynamoDB because it separates the challenge of scaling from the data transformation logic. The incoming data is stored in S3 objects, formatted as JSON, CSV, or any custom format your applications produce. The process works whether you import only a few large files or many small files. It takes advantage of parallelization to import data quickly into a DynamoDB table.

Using S3-to-Lambda to import at scale to DynamoDB.

This is useful for applications where upstream services produce transaction information, and can be effective for handling data generated by spiky workloads. Alternatively, it’s also a simple way to migrate from another data source to DynamoDB, especially for large datasets.

In this blog post, I show two different import applications. The first is a direct import into the DynamoDB table. The second explores a more advanced method for smoothing out volume in the import process. The code uses the AWS Serverless Application Model (SAM), enabling you to deploy the application in your own AWS Account. This walkthrough creates resources covered in the AWS Free Tier but you may incur cost for large data imports.

To set up both example applications, visit the GitHub repo and follow the instructions in the README.md file.

Directly importing data from S3 to DynamoDB

The first example application loads data directly from S3 to DynamoDB via a Lambda function. This uses the following architecture:

Architecture for the first example application.

  1. A downstream process creates source import data in JSON format and writes to an S3 bucket.
  2. When the objects are saved, S3 invokes the main Lambda function.
  3. The function reads the S3 object and converts the JSON into the correct format for the DynamoDB table. It uploads this data in batches to the table.

The repo’s SAM template creates a DynamoDB table with a partition key, configured to use on-demand capacity. This mode enables the DynamoDB service to scale appropriately to match the number of writes required by the import process. This means you do not need to manage DynamoDB table capacity, as you would in the standard provisioned mode.

  DDBtable:
    Type: AWS::DynamoDB::Table
    Properties:
      AttributeDefinitions:
      - AttributeName: ID
        AttributeType: S
      KeySchema:
      - AttributeName: ID
        KeyType: HASH
      BillingMode: PAY_PER_REQUEST

The template defines the Lambda function to import the data:

  ImportFunction:
    Type: AWS::Serverless::Function 
    Properties:
      CodeUri: importFunction/
      Handler: app.handler
      Runtime: nodejs12.x
      MemorySize: 512
      Environment:
        Variables:
          DDBtable: !Ref DDBtable      
      Policies:
        - DynamoDBCrudPolicy:
            TableName: !Ref DDBtable        
        - S3ReadPolicy:
            BucketName: !Ref InputBucketName
      Events:
        FileUpload:
          Type: S3
          Properties:
            Bucket: !Ref InputS3Bucket
            Events: s3:ObjectCreated:*
            Filter: 
              S3Key:
                Rules:
                  - Name: suffix
                    Value: '.json'            

This uses SAM policy templates to provide write access to the DynamoDB table and read access to S3 bucket. It also defines the event that causes the function invocation from S3, filtering only for new objects with a .json suffix.

Testing the application

  1. Deploy the first application by following the README.md in the GitHub repo, and note the application’s S3 bucket name and DynamoDB table name.
  2. Change into the dataGenerator directory:
    cd ./dataGenerator
  3. Create sample data for testing. The following command creates 10 files of 100 records each:
    node ./app.js 100 10
  4. Upload the sample data into your application’s S3 bucket, replacing your-bucket below with your bucket name:
    aws s3 cp ./data/ s3://your-bucket --recursiveYour console output shows the following, confirming that the sample data is uploaded to S3.Generating and uploading sample data for testing.
  5. After a few seconds, enter this command to show the number of items in the application’s DynamoDB table. Replace your-table with your deployed table name:aws dynamodb scan --table-name your-table --select "COUNT"Your console output shows that 1,000 items are now stored in the DynamoDB and the files are successfully imported.Checking the number of items stored in the DynamoDB.

With on-demand provisioning, the per-table limit of 40,000 write request units still applies. For high volumes or sudden, spiky workloads, DynamoDB may throttle the import when using this approach. Any throttling events appears in the Metrics tab of the table in the DynamoDB service console. Throttling is intended to protect your infrastructure but there are times when you want to process these high volumes. The second application in the repo shows how to address this.

Handling extreme loads and variability in the import process

In this next example, the goal is to smooth out the traffic, so that the load process into DynamoDB is much more consistent. The key service used to achieve this is Amazon SQS, which holds all the items until a loader process stores the data in DynamoDB. The architecture looks like this:

Architecture for the second example application.

  1. A downstream process creates source import data in JSON format and writes to an S3 bucket.
  2. When the objects are saved, S3 invokes a Lambda function that transforms the input and adds these as messages in an Amazon SQS queue.
  3. The Lambda polls the SQS queue and invokes a function to process the messages batches.
  4. The function converts the JSON messages into the correct format for the DynamoDB table. It uploads this data in batches to the table.

Testing the application

In this test, you generate a much larger amount of data using a greater number of S3 objects. The instructions below creates 100,000 sample records, so running this code may incur cost on your AWS bill.

  1. Deploy the second application by following the README.md in the GitHub repo, and note the application’s S3 bucket name and DynamoDB table name.
  2. Change into the dataGenerator directory:
    cd ./dataGenerator
  3. Create sample data for testing. The following command creates 100 files of 1,000 records each:
    node ./app.js 1000 100
  4. Upload the sample data into your application’s S3 bucket, replacing your-bucket below with your deployed bucket name:aws s3 cp ./data/ s3://your-bucket --recursiveThis process takes around 10 minutes to complete with the default configuration in the repo.
  5. From the DynamoDB console, select the application’s table and then choose the Metrics tab. Select the Write capacity graph to zoom into the chart:Using CloudWatch to view WCUs consumed in the uploading process.

The default configuration deliberately slows down the load process to illustrate how it works. Using this approach, the load into the database is much more consistent, consuming between 125-150 write capacity units (WCUs) per minute. This design makes it possible to vary how quickly you load data into the DynamoDB table, depending upon the needs of your use-case.

How this works

In this second application, there are multiple points where the application uses a configuration setting to throttle the flow of data to the next step.

Throttling points in the architecture.

  1. AddToQueue function: this loads data from the source S3 object into SQS in batches of 25 messages. Depending on the size of your source records, you may add more records into a single SQS message, which has a size limit of 256 Kb. You can also compress this message with gzip to further add more records.
  2. Function concurrency: the SAM template sets the Loader function’s concurrency to 1, using the ReservedConcurrentExecutions attribute. In effect, this stops Lambda from scaling this function, which means it keeps fetching the next batch from SQS as soon as processing finishes. The concurrency is a multiplier – as this value is increased, the loading into the DynamoDB table increases proportionately, if there are messages available in SQS. Select a value greater than 1 to use parallelization in the load process.
  3. Loader function: this consumes messages from the SQS queue. The BatchSize configured in the SAM template is set to four messages per invocation. Since each message contains 25 records, this represents 100 records per invocation when the queue has enough messages. You can set a BatchSize value from 1 to 10, so could increase this from the application’s default.

When you combine these settings, it’s possible to dramatically increase the throughput for loading data into DynamoDB. Increasing the load also increases WCUs consumed, which increases cost. Your use case can inform you about the optimal balance between speed and cost. You can make changes, too – it’s simple to make adjustments to meet your requirements.

Additionally, each of the services used has its own service limits. For high production loads, it’s important to understand the quotas set, and whether these are soft or hard limits. If your application requires higher throughput, you can request raising soft limits via an AWS Support Center ticket.

Conclusion

DynamoDB does not offer a native import process and existing solutions may not meet your needs for unplanned, large-scale imports. The AWS Database Migration Service is not serverless, and the AWS Data Pipeline is schedule-based rather than event-based. This solution is designed to provide a fully serverless alternative that responds to incoming data to S3 on-demand.

In this post, I show how you can create a simple import process directly to the DynamoDB table, triggered by objects put into an S3 bucket. This provides a near-real time import process. I also show a more advanced approach to smooth out traffic for high-volume or spiky workloads. This helps create a resilient and consistent data import for DynamoDB.

To learn more, watch this video to see how to deploy and test the DynamoDB importer application.

Serving Billions of Ads in Just 100 ms Using Amazon Elasticache for Redis

Post Syndicated from Rodrigo Asensio original https://aws.amazon.com/blogs/architecture/serving-billions-of-ads-with-amazon-elasticache-for-redis/

This post was co-written with Lucas Ceballos, CTO of Smadex

Introduction

Showing ads may seem to be a simple task, but it’s not. Showing the right ad to the right user is an incredibly complex challenge that involves multiple disciplines such as artificial intelligence, data science, and software engineering. Doing it one million times per second with a 100-ms constraint is even harder.

In the ad-tech business, speed and infrastructure costs are the keys to success. The less the final user waits for an ad, the higher the probability of that user clicking on the ad. Doing that while keeping infrastructure costs under control is crucial for business profitability.

About Smadex

Smadex is the leading mobile-first programmatic advertising platform specifically built to deliver best user acquisition performance and complete transparency.

Its state-of-the-art digital signal processing (DSP) technology provides advertisers with the tools they need to achieve their goals and ROI, with measurable results from web forms, post-app install events, store visits, and sales.

Smadex advertising architecture

What does showing ads look like under the hood? At Smadex, our technology works based on the OpenRTB (Real-Time Bidding) protocol.

RTB is a means by which advertising inventory is bought and sold on a per-impression basis, via programmatic instantaneous auction, which is similar to financial markets.

To show ads, we participate in auctions deciding in real time which ad to show and how much to bid trying to optimize the cost of every impression.

High level diagram

  1. The final user browses the publisher’s website or app.
  2. Ad-exchange is called to start a new auction.
  3. Smadex receives the bid request and has to decide which ad to show and how much to offer in just 100 ms (and this is happening one million times per second).
  4. If Smadex won the auction, the ad must be sent and rendered on the publisher’s website or app.
  5. In the end, the user interacts with the ad sending new requests and events to Smadex platform.

Flow of data

As you can see in the previous diagram, showing ads is just one part of the challenge. After the ad is shown, the final user interacts with it in multiple ways, such as clicking it, installing an application, subscribing to a service, etc. This happens during a determined period that we call the “attribution window.” All of those interactions must be tracked and linked to the original bid transaction (using the request_id parameter).

Doing this is complicated: billions of bid transactions must be stored and available so that they can be quickly accessed every time the user interacts with the ad. The longer we store the transactions, the longer we can “wait” for an interaction to take place, and the better for our business and our clients, too.

Detailed diagram

Challenge #1: Cost

The challenge is: What kind of database can store billions of records per day, with at least a 30-day retention capacity (attribution window), be accessed by key-value, and all by spending as little as possible?

The answer is…none! Based on our research, all the available options that met the technical requirements were way out of our budget.

So…how to solve it? Here is when creativity and the combination of different AWS services comes into place.

We started to analyze the time dispersion of the events trying to find some clues. The interesting thing we spotted was that 90% of what we call “post-bid events” (impression, click, install, etc.) happened within one hour after the auction took place.

That means that we can process 90% of post-bid events by storing just one hour of bids.

Under our current workload, in one hour we participate in approximately 3.7 billion auctions generating 100 million bid records of an average 600 bytes each. This adds up to 55 gigabytes per hour, an easier amount of data to process.

Instead of thinking about one single database to store all the bid requests, we decided to split bids into two different categories:

  • Hot Bid: A request that took place within the last hour (small amount and frequently accessed)
  • Cold Bid: A request that took place more than our hour ago (huge amount and infrequently accessed)

Amazon ElastiCache for Redis is the best option to store 55 GB of data in memory, which gives us the ability to query in a key-value way with the lowest possible latency.

Hot Bids flow

Hot Bids flow diagram

  1. Every new bid is a hot bid by definition so it’s going to be stored in the hot bids Redis cluster.
  2. At the moment of the user interaction with the ad, the Smadex tracker component receives an HTTPS notification, including the bid request UUID that originated it.
  3. Based on the date of occurrence extracted from the received UUID, the tracker component can determine if it’s looking for a hot bid or not. If it’s a hot bid, the tracker reads it directly from Redis performing a key-value lookup query.

It’s been easy so far but what to do with the other 29 days and 23 hours we need to store?

Challenge #2: Performance

As we previously mentioned, cold bids are a huge infrequently accessed number of records with only 10% of post-bid events pointing to them. That sounds like a good use case for an inexpensive and slower data store like Amazon S3.

Thanks to the S3 low-cost storage prices combined with the ability to query S3 objects directly using Amazon Athena, we were able to optimize our costs by storing and querying cold bids by implementing a serverless architecture.

Cold Bids Flow

Cold Bids flow diagram

  1. Incoming bids are buffered by Fluentd and flushed to S3 every one minute in JSON format. Every single file flushed to S3 contains all the bids processed by a specific EC2 instance for one minute.
  2. An AWS Lambda function is automatically triggered on every new PutObject event from S3. This function transforms the JSON records to Parquet format and will save it back the S3 bucket, but this time into a specific partition folder based on file creation timestamp.
  3. As seen on the hot bids flow, the tracker component will determine if it’s looking for a hot or a cold bid based on the extracted timestamp of the request UUID. In this case, the cold bid will be retrieved by running an Amazon Athena look-up query leveraging the use of partitions and Parquet format to reduce as much as possible the latency and data that needs to be scanned.

Conclusion

Thanks to this combined approach using different technologies and a variety of AWS services we were able to extend our attribution window from 30 to 90 days while reducing the infrastructure costs by 45%.

 

 

Translating documents at enterprise scale with serverless

Post Syndicated from James Beswick original https://aws.amazon.com/blogs/compute/translating-documents-at-enterprise-scale-with-serverless/

For organizations operating in multiple countries, helping customers in different languages is an everyday reality. But in many IT systems, data remains static in a single language, making it difficult or impossible for international customers to use. In this blog post, I show how you can automate language translation at scale to solve a number of common enterprise problems.

Many types of data are good targets for translation. For example, product catalog information, for sharing with a geographically broad customer base. Or customer emails and interactions in multiple languages, translating back to a single language for analytics. Or even resource files for mobile applications, to automate string processing into different languages during the build process.

Building the machine learning models for language translation is extraordinarily complex, so fortunately you have a pay-as-you-go service available in Amazon Translate. This can accurately translate text between 54 languages, and automatically detects the source language.

Developing a scalable translation solution for thousands of documents can be challenging using traditional, server-based architecture. Using a serverless approach, this becomes much easier since you can use storage and compute services that scale for you – Amazon S3 and AWS Lambda:

Integrating S3 with Translate via Lambda.

In this post, I show two solutions that provide an event-based architecture for automated translation. In the first, S3 invokes a Lambda function when objects are stored. It immediately reads the file and requests a translation from Amazon Translate. The second solution explores a more advanced method for scaling to large numbers of documents, queuing the requests and tracking their state. The walkthrough creates resources covered in the AWS Free Tier but you may incur costs for usage.

To set up both example applications, visit the GitHub repo and follow the instructions in the README.md file. Both applications use the AWS Serverless Application Model (SAM) to make it easy to deploy in your AWS account.

Translating in near real time

In the first application, the workflow is straightforward. The source text is sent immediately to Translate for processing, and the result is saved back into the S3 bucket. It provides near real-time translation whenever an object is saved. This uses the following architecture:

Architecture for the first example application.

  1. The source text is saved in the Batching S3 bucket.
  2. The S3 put event invokes the Batching Lambda function. Since Translate has a limit of 5,000 characters per request, it slices the contents of the input into parts small enough for processing.
  3. The resulting parts are saved in the Translation S3 bucket.
  4. The S3 put events invoke the Translation function, which scales up concurrently depending on the number of parts.
  5. Amazon Translate returns the translations back to the Lambda function, which saves the results in the Translation bucket.

The repo’s SAM template allows you to specify a list of target languages, as a space-delimited list of supported language codes. In this case, any text uploaded to S3 is translated into French, Spanish, and Italian:

Parameters:
  TargetLanguage:
    Type: String
    Default: 'fr es it'

Testing the application

  1. Deploy the first application by following the README.md in the GitHub repo. Note the application’s S3 Translation and Batching bucket names shown in the output:Output values after SAM deployment.
  2. The testdata directory contains several sample text files. Change into this directory, then upload coffee.txt to the S3 bucket, replacing your-bucket below with your Translation bucket name:
    cd ./testdata/
    aws s3 cp ./coffee.txt s3://your-bucket
  3. The application invokes the translation workflow, and within a couple of seconds you can list the output files in the translations folder:aws s3 ls s3://your-bucket/translations/

    Translations output.

  4. Create an output directory, then download the translations to your local machine to view the contents:
    mkdir output
    aws s3 cp s3://your-bucket/translations/ ./output/ --recursive
    more ./output/coffee-fr.txt
    more ./output/coffee-es.txt
    more ./output/coffee-it.txt
  5. For the next step, translate several text files containing test data. Copy these to the Translation bucket, replacing your-bucket below with your bucket name:aws s3 cp ./ s3://your-bucket --include "*.txt" --exclude "*/*" --recursive
  6. After a few seconds, list the files in the translations folder to see your translated files:aws s3 ls s3://your-bucket/translations/

    Listing the translated files.

  7. Finally, translate a larger file using the batching process. Copy this file to the Batching S3 bucket (replacing your-bucket with this bucket name):
    cd ../testdata-batching/
    aws s3 cp ./your-filename.txt s3://your-bucket
  8. Since this is a larger file, the batching Lambda function breaks it apart into smaller text files in the Translation bucket. List these files in the terminal, together with their translations:
    aws s3 ls s3://your-bucket
    aws s3 ls s3://your-bucket/translations/

    Listing the output files.

In this example, you can translate a reasonable number of text files for a trivial use-case. However, in an enterprise environment where there could be thousands of files in a single bucket, you need a more robust architecture. The second application introduces a more resilient approach.

Scaling up the translation solution

In an enterprise environment, the application must handle long documents and large quantities of documents. Amazon Translate has service limits in place per account – you can request an increase via an AWS Support Center ticket if needed. However, S3 can ingest a large number of objects quickly, so the application should decouple these two services.

The next example uses Amazon SQS for decoupling, and also introduces Amazon DynamoDB for tracking the status of each translation. The new architecture looks like this:

Decoupled translation architecture.

  1. A downstream process saves text objects in the Batching S3 bucket.
  2. The Batching function breaks these files into smaller parts, saving these in the Translation S3 bucket.
  3. When an object is saved in this bucket, this invokes the Add to Queue function. This writes a message to an SQS queue, and logs the item in a DynamoDB table.
  4. The Translation function receives messages from the SQS queue, and requests translations from the Amazon Translate service.
  5. The function updates the item as completed in the DynamoDB table, and stores the output translation in the Results S3 bucket.

Testing the application

This test uses a much larger text document – the text version of the novel War and Peace, which is over 3 million characters long. It’s recommended that you use a shorter piece of text for the walkthrough, between 20-50 kilobytes, to minimize cost on your AWS bill.

  1. Deploy the second application by following the README.md in the GitHub repo, and note the application’s S3 bucket name and DynamoDB table name.
    The output values from the SAM deployment.
  2. Download your text sample and then upload it the Batching bucket. Replace your-bucket with your bucket name and your-text.txt with your text file name:aws s3 cp ./your-text.txt s3://your-bucket/ 
  3. The batching process creates smaller files in the Translation bucket. After a few seconds, list the files in the Translation bucket (replacing your-bucket with your bucket name):aws s3 ls s3://patterns-translations-v2/ --recursive --summarize

    Listing the translated files.

  4. To see the status of the translations, navigate to the DynamoDB console. Select Tables in the left-side menu and then choose the application’s DynamoDB table. Select the Items tab:Listing items in the DynamoDB table.

    This shows each translation file and a status of Queue or Translated.

  5. As translations complete, these appear in the Results bucket:aws s3 ls s3://patterns-results-v2/ --summarize

    Listing the output translations.

How this works

In the second application, the SQS queue acts as a buffer between the Batching process and the Translation process. The Translation Lambda function fetches messages from the SQS queue when they are available, and submits the source file to Amazon Translate. This throttles the overall speed of processing.

There are configuration settings you can change in the SAM template to vary the speed of throughput:

  • Translator function: this consumes messages from the SQS queue. The BatchSize configured in the SAM template is set to one message per invocation. This is equivalent to processing one source file at a time. You can set a BatchSize value from 1 to 10, so could increase this from the application’s default.
  • Function concurrency: the SAM template sets the Loader function’s concurrency to 1, using the ReservedConcurrentExecutions attribute. In effect, this means Lambda can only invoke 1 function at the same time. As a result, it keeps fetching the next batch from SQS as soon as processing finishes. The concurrency is a multiplier – as this value is increased, the translation throughput increases proportionately, if there are messages available in SQS.
  • Amazon Translate limits: the service limits in place are designed to protect you from higher-than-intended usage. If you need higher soft limits, open an AWS Support Center ticket.

Combining these settings, you have considerable control over the speed of processing. The defaults in the sample application are set at the lowest values possible so you can observe the queueing mechanism.

Conclusion

Automated translation using deep learning enables you to make documents available at scale to an international audience. For organizations operating globally, this can improve your user experience and increase customer access to your company’s products and services.

In this post, I show how you can create a serverless application to process large numbers of files stored in S3. The write operation in the S3 bucket triggers the process, and you use SQS to buffer the workload between S3 and the Amazon Translate service. This solution also uses DynamoDB to help track the state of the translated files.

Creating a searchable enterprise document repository

Post Syndicated from James Beswick original https://aws.amazon.com/blogs/compute/creating-a-searchable-enterprise-document-repository/

Enterprise customers frequently have repositories with thousands of documents, images and other media. While these contain valuable information for users, it’s often hard to index and search this content. One challenge is interpreting the data intelligently, and another issue is processing these files at scale.

In this blog post, I show how you can deploy a serverless application that uses machine learning to interpret your documents. The architecture includes a queueing mechanism for handling large volumes, and posting the indexing metadata to an Amazon Elasticsearch domain. This solution is scalable and cost effective, and you can modify the functionality to meet your organization’s requirements.

The overall architecture for a searchable document repository solution.

The application takes unstructured data and applies machine learning to extract context for a search service, in this case Elasticsearch. There are many business uses for this design. For example, this could help Human Resources departments in searching for skills in PDF resumes. It can also be used by Marketing departments or creative groups with a large number of images, providing a simple way to search for items within the images.

As documents and images are added to the S3 bucket, these events invoke AWS Lambda functions. This runs custom code to extract data from the files, and also calls Amazon ML services for interpretation. For example, when adding a resume, the Lambda function extracts text from the PDF file while Amazon Comprehend determines the key phrases and topics in the document. For images, it uses Amazon Rekognition to determine the contents. In both cases, once it identifies the indexing attributes, it saves the data to Elasticsearch.

The code uses the AWS Serverless Application Model (SAM), enabling you to deploy the application easily in your own AWS Account. This walkthrough creates resources covered in the AWS Free Tier but you may incur cost for large data imports. Additionally, it requires an Elasticsearch domain, which may incur cost on your AWS bill.

To set up the example application, visit the GitHub repo and follow the instructions in the README.md file.

Creating an Elasticsearch domain

This application requires an Elasticsearch development domain for testing purposes. To learn more about production configurations, see the Elasticsearch documentation.

To create the test domain:

  1. Navigate to the Amazon Elasticsearch console. Choose Create a new domain.
  2. For Deployment type, choose Development and testing. Choose Next.
  3. In the Configure Domain page:
    1. For Elasticsearch domain name, enter serverless-docrepo.
    2. Change Instance Type to t2.small.elasticsearch.
    3. Leave all the other defaults. Choose Next at the end of the page.
  4. In Network Configuration, choose Public access. This is adequate for a tutorial but in a production use-case, it’s recommended to use VPC access.
  5. Under Access Policy, in the Domain access policy dropdown, choose Custom access policy.
  6. Select IAM ARN and in the Enter principal field, enter your AWS account ID (learn how to find your AWS account ID). In the Select Action dropdown, select Allow.
  7. Under Encryption, leave HTTPS checked. Choose Next.
  8. On the Review page, review your domain configuration, and then choose Confirm.

Your domain is now being configured, and the Domain status shows Loading in the Overview tab.

After creating the Elasticsearch domain, the domain status shows ‘Loading’.

It takes 10-15 minutes to fully configure the domain. Wait until the Domain status shows Active before continuing. When the domain is ready, note the Endpoint address since you need this in the application deployment.

The Endpoint for an Elasticsearch domain.

Deploying the application

After cloning the repo to your local development machine, open a terminal window and change to the cloned directory.

  1. Run the SAM build process to create an installation package, and then deploy the application:
    sam build
    sam deploy --guided
  2. In the guided deployment process, enter unique names for the S3 buckets when prompted. For the ESdomain parameter, enter the Elasticsearch domain Endpoint from the previous section.
  3. After the deployment completes, note the Lambda function ARN shown in the output:
    Lambda function ARN output.
  4. Back in the Elasticsearch console, select the Actions dropdown and choose Modify Access Policy. Paste the Lambda function ARN as an AWS Principal in the JSON, in addition to the root user, as follows:
    Modify access policy for Elasticsearch.
  5. Choose Submit. This grants the Lambda function access to the Elasticsearch domain.

Testing the application

To test the application, you need a few test documents and images with the file types DOCX (Microsoft Word), PDF, and JPG. This walkthrough uses multiple files to illustrate the queuing process.

  1. Navigate to the S3 console and select the Documents bucket from your deployment.
  2. Choose Upload and select your sample PDF or DOCX files:
    Upload files to S3.
  3. Choose Next on the following three pages to complete the upload process. The application now analyzes these documents and adds the indexing information to Elasticsearch.

To query Elasticsearch, first you must generate an Access Key ID and Secret Access Key. For detailed steps, see this documentation on creating security credentials. Next, use Postman to create an HTTP request for the Elasticsearch domain:

  1. Download and install Postman.
  2. From Postman, enter the Elasticsearch endpoint, adding /_search?q=keyword. Replace keyword with a search term.
  3. On the Authorization tab, complete the Access Key ID and Secret Access Key fields with the credentials you created. For Service Name, enter es.Postman query with AWS authorization.
  4. Choose Send. Elasticsearch responds with document results matching your search term.REST response from Elasticsearch.

How this works

This application creates a processing pipeline between the originating Documents bucket and the Elasticsearch domain. Each document type has a custom parser, preparing the content in a Queuing bucket. It uses an Amazon SQS queue to buffer work, which is fetched by a Lambda function and analyzed with Amazon Comprehend. Finally, the indexing metadata is saved in Elasticsearch.

Serverless architecture for text extraction and classification.

  1. Documents and images are saved in the Documents bucket.
  2. Depending upon the file type, this triggers a custom parser Lambda function.
    1. For PDFs and DOCX files, the extracted text is stored in a Staging bucket. If the content is longer than 5,000 characters, it is broken into smaller files by a Lambda function, then saved in the Queued bucket.
    2. For JPG files, the parser uses Amazon Rekognition to detect the contents of the image. The labeling metadata is stored in the Queued bucket.
  3. When items are stored in the Queued bucket, this triggers a Lambda function to add the job to an SQS queue.
  4. The Analyze function is invoked when there are messages in the SQS queue. It uses Amazon Comprehend to find entities in the text. This function then stores the metadata in Elasticsearch.

S3 and Lambda both scale to handle the traffic. The Elasticsearch domain is not serverless, however, so it’s possible to overwhelm this instance with requests. There may be a large number of objects stored in the Documents bucket triggering the workflow, so the application uses SQS couple to smooth out the traffic. When large numbers of objects are processed, you see the Messages Available increase in the SQS queue:

SQS queue buffers messages to smooth out traffic.
For the Lambda function consuming messages from the SQS queue, the BatchSize configured in the SAM template controls the rate of processing. The function continues to fetch messages from the SQS queue while Messages Available is greater than zero. This can be a useful mechanism for protecting downstream services that are not serverless, or simply cannot scale to match the processing rates of the upstream application. In this case, it provides a more consistent flow of indexing data to the Elasticsearch domain.

In a production use-case, you would scale the Elasticsearch domain depending upon the load of search requests, not just the indexing traffic from this application. This tutorial uses a minimal Elasticsearch configuration, but this service is capable of supporting enterprise-scale use-cases.

Conclusion

Enterprise document repositories can be a rich source of information but can be difficult to search and index. In this blog post, I show how you can use a serverless approach to build scalable solution easily. With minimum code, we can use Amazon ML services to create the indexing metadata. By using powerful image recognition and language comprehension capabilities, this makes the metadata more useful and the search solution more accurate.

This also shows how serverless solutions can be used with existing non-serverless infrastructure, like Elasticsearch. By decoupling scalable serverless applications, you can protect downstream services from heavy traffic loads, even as Lambda scales up. Elasticsearch provides a fast, easy way to query your document repository once the serverless application has completed the indexing process.

To learn more about how to use Elasticsearch for production workloads, see the documentation on managing domains.

Application Integration Using Queues and Messages

Post Syndicated from Mithun Mallick original https://aws.amazon.com/blogs/architecture/application-integration-using-queues-and-messages/

In previous blog posts in this messaging series, we provided an overview of messaging and we also explained the common characteristics to consider when  evaluating messaging channel technologies. In this post, we will explain some of the semantics of queue-based processing, its use in designing flexible systems, and how to apply it to your use cases. AWS offers two queue-based services: Amazon Simple Queue Service (SQS) and Amazon MQ. We will focus on SQS in this blog.

Building Blocks

In the digital world, even the most basic design of web based systems requires the use of queues to integrate applications. SQS is a secure, serverless, durable, and highly available message queue service. It provides a simple REST API to create a queue as well as send, receive, and delete messages.

Producing Messages

Message producers are processes that call the SendMessage APIs of SQS. SQS supports two types of queues: standard and first-in-first-out (FIFO). Standard queues provide best effort ordering while FIFO provides first-in-first-out ordering. Messages can be sent either as single messages or in batches. Standard queues can support unlimited throughput by adding as many concurrent producers as needed, whereas FIFO queues can support up to 300 TPS without batching and 3000 TPS with batching.

app integration- producing messages

In terms of message delivery, SQS standard supports “at-least-once” delivery, meaning that messages will be delivered at least once but occasionally, more than one copy of the message will be delivered. SQS FIFO provides “exactly once” delivery, meaning it can detect duplicate messages.

Consuming Messages

Message consumers are processes that make the ReceiveMessage API call on SQS. Messages from queues can be processed either in batch or as a single message at a time. Each approach has its advantages and disadvantages.

  • Batch processing: This is where each message can be processed independently, and an error on a single message is not required to disrupt the entire batch. Batch processing provides the most throughput for processing, and also provides the most optimizations for resources involved in reading messages.
  • Single message processing: Single message processing is commonly used in scenarios where each message may trigger multiple processes within the consumer. In case of errors, the retry is confined to the single message.

To maintain the order of processing, FIFO queues are typically consumed by a single process. Messages across message groups can still be processed in parallel. However, the overall throughput limit will still apply for FIFO queues.

SQS supports long and short polling on queues. Short polling will sample a subset of servers and will return the messages or provide an empty response if there are no messages in those servers. Long polling will query all servers and return immediately if there are messages. Long polling can reduce cost by reducing the number of calls in cases of empty responses. Get more details and sample code for sending and receiving messages.

It’s simple to poll for messages but it does have an overhead of running a process continuously. In some situations it may be hard to monitor such processes and troubleshoot in case of errors. SQS integration with AWS Lambda takes the undifferentiated heavy lifting of polling logic into a Lambda agent.

app integration- consuming messages

Get more details for SQS as an event source for Lambda in this blog post: AWS Lambda Adds Amazon Simple Queue Service to Supported Event Sources. This is supported for both standard as well as FIFO queues.

Benefits

Queue-based processing can protect backend systems like relational databases from unexpected surge in front-end traffic. You can decouple the processing logic by using an intermediate queue.

Consider a scenario that involves a web application to order products related to a TV show. It’s possible that the ordering and payment systems rely on a traditional relational database. During peak show seasons, queues can be used to control the traffic to the ordering and payment system. In the following diagram we show how the web application requests are staged in the queue, while the consumption of the users on the backend is based on the capacity of the database.

app integration - benefits

Error Handling

Asynchronous message processing presents unique challenges with error handling since they just manifest as log entries on the producer or consumer and create backlogs in processing. In order to handle errors elegantly, the first requirement is to address the nature of the error. In case the error is transient, it may help to retry after a brief delay and eventually move to a dead letter queue if repeated attempts fail. The other error scenario can be cases where the data itself is bad and the error won’t fix, even after repeated attempts. In such cases, consumer process needs to make this determination and needs to move the message to an error queue.

In some cases, due to its highly distributed architecture, Standard SQS can deliver a duplicate message, which it can result in duplicate key errors for the consumer. The best way to mitigate it is to make the consumers idempotent so that processing the same message multiple times produces the same result.

Conclusion

In this blog, we explained the importance of messaging in building distributed applications, various aspects of queue based processing using SQS like sending and receiving messages, FIFO versus Standard SQS, and common error handling scenarios. We also covered using SQS as an event source for AWS Lambda. Please refer to following blog posts for using messaging in different integration patterns:

Halodoc: Building the Future of Tele-Health One Microservice at a Time

Post Syndicated from Annik Stahl original https://aws.amazon.com/blogs/architecture/halodoc-building-the-future-of-tele-health-one-microservice-at-a-time/

Halodoc, a Jakarta-based healthtech platform, uses tele-health and artificial intelligence to connect patients, doctors, and pharmacies. Join builder Adrian De Luca for this special edition of This is My Architecture as he dives deep into the solutions architecture of this Indonesian healthtech platform that provides healthcare services in one of the most challenging traffic environments in the world.

Explore how the company evolved its monolithic backend into decoupled microservices with Amazon EC2 and Amazon Simple Queue Service (SQS), adopted serverless to cost effectively support new user functionality with AWS Lambda, and manages the high volume and velocity of data with Amazon DynamoDB, Amazon Relational Database Service (RDS), and Amazon Redshift.

For more content like this, subscribe to our YouTube channels This is My Architecture, This is My Code, and This is My Model, or visit the This is My Architecture AWS website, which has search functionality and the ability to filter by industry, language, and service.

ICYMI: Serverless Q4 2019

Post Syndicated from Rob Sutter original https://aws.amazon.com/blogs/compute/icymi-serverless-q4-2019/

Welcome to the eighth edition of the AWS Serverless ICYMI (in case you missed it) quarterly recap. Every quarter, we share the most recent product launches, feature enhancements, blog posts, webinars, Twitch live streams, and other interesting things that you might have missed!

In case you missed our last ICYMI, checkout what happened last quarter here.

The three months comprising the fourth quarter of 2019

AWS re:Invent

AWS re:Invent 2019

re:Invent 2019 dominated the fourth quarter at AWS. The serverless team presented a number of talks, workshops, and builder sessions to help customers increase their skills and deliver value more rapidly to their own customers.

Serverless talks from re:Invent 2019

Chris Munns presenting 'Building microservices with AWS Lambda' at re:Invent 2019

We presented dozens of sessions showing how customers can improve their architecture and agility with serverless. Here are some of the most popular.

Videos

Decks

You can also find decks for many of the serverless presentations and other re:Invent presentations on our AWS Events Content.

AWS Lambda

For developers needing greater control over performance of their serverless applications at any scale, AWS Lambda announced Provisioned Concurrency at re:Invent. This feature enables Lambda functions to execute with consistent start-up latency making them ideal for building latency sensitive applications.

As shown in the below graph, provisioned concurrency reduces tail latency, directly impacting response times and providing a more responsive end user experience.

Graph showing performance enhancements with AWS Lambda Provisioned Concurrency

Lambda rolled out enhanced VPC networking to 14 additional Regions around the world. This change brings dramatic improvements to startup performance for Lambda functions running in VPCs due to more efficient usage of elastic network interfaces.

Illustration of AWS Lambda VPC to VPC NAT

New VPC to VPC NAT for Lambda functions

Lambda now supports three additional runtimes: Node.js 12, Java 11, and Python 3.8. Each of these new runtimes has new version-specific features and benefits, which are covered in the linked release posts. Like the Node.js 10 runtime, these new runtimes are all based on an Amazon Linux 2 execution environment.

Lambda released a number of controls for both stream and async-based invocations:

  • You can now configure error handling for Lambda functions consuming events from Amazon Kinesis Data Streams or Amazon DynamoDB Streams. It’s now possible to limit the retry count, limit the age of records being retried, configure a failure destination, or split a batch to isolate a problem record. These capabilities help you deal with potential “poison pill” records that would previously cause streams to pause in processing.
  • For asynchronous Lambda invocations, you can now set the maximum event age and retry attempts on the event. If either configured condition is met, the event can be routed to a dead letter queue (DLQ), Lambda destination, or it can be discarded.

AWS Lambda Destinations is a new feature that allows developers to designate an asynchronous target for Lambda function invocation results. You can set separate destinations for success and failure. This unlocks new patterns for distributed event-based applications and can replace custom code previously used to manage routing results.

Illustration depicting AWS Lambda Destinations with success and failure configurations

Lambda Destinations

Lambda also now supports setting a Parallelization Factor, which allows you to set multiple Lambda invocations per shard for Kinesis Data Streams and DynamoDB Streams. This enables faster processing without the need to increase your shard count, while still guaranteeing the order of records processed.

Illustration of multiple AWS Lambda invocations per Kinesis Data Streams shard

Lambda Parallelization Factor diagram

Lambda introduced Amazon SQS FIFO queues as an event source. “First in, first out” (FIFO) queues guarantee the order of record processing, unlike standard queues. FIFO queues support messaging batching via a MessageGroupID attribute that supports parallel Lambda consumers of a single FIFO queue, enabling high throughput of record processing by Lambda.

Lambda now supports Environment Variables in the AWS China (Beijing) Region and the AWS China (Ningxia) Region.

You can now view percentile statistics for the duration metric of your Lambda functions. Percentile statistics show the relative standing of a value in a dataset, and are useful when applied to metrics that exhibit large variances. They can help you understand the distribution of a metric, discover outliers, and find hard-to-spot situations that affect customer experience for a subset of your users.

Amazon API Gateway

Screen capture of creating an Amazon API Gateway HTTP API in the AWS Management Console

Amazon API Gateway announced the preview of HTTP APIs. In addition to significant performance improvements, most customers see an average cost savings of 70% when compared with API Gateway REST APIs. With HTTP APIs, you can create an API in four simple steps. Once the API is created, additional configuration for CORS and JWT authorizers can be added.

AWS SAM CLI

Screen capture of the new 'sam deploy' process in a terminal window

The AWS SAM CLI team simplified the bucket management and deployment process in the SAM CLI. You no longer need to manage a bucket for deployment artifacts – SAM CLI handles this for you. The deployment process has also been streamlined from multiple flagged commands to a single command, sam deploy.

AWS Step Functions

One powerful feature of AWS Step Functions is its ability to integrate directly with AWS services without you needing to write complicated application code. In Q4, Step Functions expanded its integration with Amazon SageMaker to simplify machine learning workflows. Step Functions also added a new integration with Amazon EMR, making EMR big data processing workflows faster to build and easier to monitor.

Screen capture of an AWS Step Functions step with Amazon EMR

Step Functions step with EMR

Step Functions now provides the ability to track state transition usage by integrating with AWS Budgets, allowing you to monitor trends and react to usage on your AWS account.

You can now view CloudWatch Metrics for Step Functions at a one-minute frequency. This makes it easier to set up detailed monitoring for your workflows. You can use one-minute metrics to set up CloudWatch Alarms based on your Step Functions API usage, Lambda functions, service integrations, and execution details.

Step Functions now supports higher throughput workflows, making it easier to coordinate applications with high event rates. This increases the limits to 1,500 state transitions per second and a default start rate of 300 state machine executions per second in US East (N. Virginia), US West (Oregon), and Europe (Ireland). Click the above link to learn more about the limit increases in other Regions.

Screen capture of choosing Express Workflows in the AWS Management Console

Step Functions released AWS Step Functions Express Workflows. With the ability to support event rates greater than 100,000 per second, this feature is designed for high-performance workloads at a reduced cost.

Amazon EventBridge

Illustration of the Amazon EventBridge schema registry and discovery service

Amazon EventBridge announced the preview of the Amazon EventBridge schema registry and discovery service. This service allows developers to automate discovery and cataloging event schemas for use in their applications. Additionally, once a schema is stored in the registry, you can generate and download a code binding that represents the schema as an object in your code.

Amazon SNS

Amazon SNS now supports the use of dead letter queues (DLQ) to help capture unhandled events. By enabling a DLQ, you can catch events that are not processed and re-submit them or analyze to locate processing issues.

Amazon CloudWatch

Amazon CloudWatch announced Amazon CloudWatch ServiceLens to provide a “single pane of glass” to observe health, performance, and availability of your application.

Screenshot of Amazon CloudWatch ServiceLens in the AWS Management Console

CloudWatch ServiceLens

CloudWatch also announced a preview of a capability called Synthetics. CloudWatch Synthetics allows you to test your application endpoints and URLs using configurable scripts that mimic what a real customer would do. This enables the outside-in view of your customers’ experiences, and your service’s availability from their point of view.

CloudWatch introduced Embedded Metric Format, which helps you ingest complex high-cardinality application data as logs and easily generate actionable metrics. You can publish these metrics from your Lambda function by using the PutLogEvents API or using an open source library for Node.js or Python applications.

Finally, CloudWatch announced a preview of Contributor Insights, a capability to identify who or what is impacting your system or application performance by identifying outliers or patterns in log data.

AWS X-Ray

AWS X-Ray announced trace maps, which enable you to map the end-to-end path of a single request. Identifiers show issues and how they affect other services in the request’s path. These can help you to identify and isolate service points that are causing degradation or failures.

X-Ray also announced support for Amazon CloudWatch Synthetics, currently in preview. CloudWatch Synthetics on X-Ray support tracing canary scripts throughout the application, providing metrics on performance or application issues.

Screen capture of AWS X-Ray Service map in the AWS Management Console

X-Ray Service map with CloudWatch Synthetics

Amazon DynamoDB

Amazon DynamoDB announced support for customer-managed customer master keys (CMKs) to encrypt data in DynamoDB. This allows customers to bring your own key (BYOK) giving you full control over how you encrypt and manage the security of your DynamoDB data.

It is now possible to add global replicas to existing DynamoDB tables to provide enhanced availability across the globe.

Another new DynamoDB capability to identify frequently accessed keys and database traffic trends is currently in preview. With this, you can now more easily identify “hot keys” and understand usage of your DynamoDB tables.

Screen capture of Amazon CloudWatch Contributor Insights for DynamoDB in the AWS Management Console

CloudWatch Contributor Insights for DynamoDB

DynamoDB also released adaptive capacity. Adaptive capacity helps you handle imbalanced workloads by automatically isolating frequently accessed items and shifting data across partitions to rebalance them. This helps reduce cost by enabling you to provision throughput for a more balanced workload instead of over provisioning for uneven data access patterns.

Amazon RDS

Amazon Relational Database Services (RDS) announced a preview of Amazon RDS Proxy to help developers manage RDS connection strings for serverless applications.

Illustration of Amazon RDS Proxy

The RDS Proxy maintains a pool of established connections to your RDS database instances. This pool enables you to support a large number of application connections so your application can scale without compromising performance. It also increases security by enabling IAM authentication for database access and enabling you to centrally manage database credentials using AWS Secrets Manager.

AWS Serverless Application Repository

The AWS Serverless Application Repository (SAR) now offers Verified Author badges. These badges enable consumers to quickly and reliably know who you are. The badge appears next to your name in the SAR and links to your GitHub profile.

Screen capture of SAR Verifiedl developer badge in the AWS Management Console

SAR Verified developer badges

AWS Developer Tools

AWS CodeCommit launched the ability for you to enforce rule workflows for pull requests, making it easier to ensure that code has pass through specific rule requirements. You can now create an approval rule specifically for a pull request, or create approval rule templates to be applied to all future pull requests in a repository.

AWS CodeBuild added beta support for test reporting. With test reporting, you can now view the detailed results, trends, and history for tests executed on CodeBuild for any framework that supports the JUnit XML or Cucumber JSON test format.

Screen capture of AWS CodeBuild

CodeBuild test trends in the AWS Management Console

Amazon CodeGuru

AWS announced a preview of Amazon CodeGuru at re:Invent 2019. CodeGuru is a machine learning based service that makes code reviews more effective and aids developers in writing code that is more secure, performant, and consistent.

AWS Amplify and AWS AppSync

AWS Amplify added iOS and Android as supported platforms. Now developers can build iOS and Android applications using the Amplify Framework with the same category-based programming model that they use for JavaScript apps.

Screen capture of 'amplify init' for an iOS application in a terminal window

The Amplify team has also improved offline data access and synchronization by announcing Amplify DataStore. Developers can now create applications that allow users to continue to access and modify data, without an internet connection. Upon connection, the data synchronizes transparently with the cloud.

For a summary of Amplify and AppSync announcements before re:Invent, read: “A round up of the recent pre-re:Invent 2019 AWS Amplify Launches”.

Illustration of AWS AppSync integrations with other AWS services

Q4 serverless content

Blog posts

October

November

December

Tech talks

We hold several AWS Online Tech Talks covering serverless tech talks throughout the year. These are listed in the Serverless section of the AWS Online Tech Talks page.

Here are the ones from Q4:

Twitch

October

There are also a number of other helpful video series covering Serverless available on the AWS Twitch Channel.

AWS Serverless Heroes

We are excited to welcome some new AWS Serverless Heroes to help grow the serverless community. We look forward to some amazing content to help you with your serverless journey.

AWS Serverless Application Repository (SAR) Apps

In this edition of ICYMI, we are introducing a section devoted to SAR apps written by the AWS Serverless Developer Advocacy team. You can run these applications and review their source code to learn more about serverless and to see examples of suggested practices.

Still looking for more?

The Serverless landing page has much more information. The Lambda resources page contains case studies, webinars, whitepapers, customer stories, reference architectures, and even more Getting Started tutorials. We’re also kicking off a fresh series of Tech Talks in 2020 with new content providing greater detail on everything new coming out of AWS for serverless application developers.

Throughout 2020, the AWS Serverless Developer Advocates are crossing the globe to tell you more about serverless, and to hear more about what you need. Follow this blog to keep up on new launches and announcements, best practices, and examples of serverless applications in action.

You can also follow all of us on Twitter to see latest news, follow conversations, and interact with the team.

Chris Munns: @chrismunns
Eric Johnson: @edjgeek
James Beswick: @jbesw
Moheeb Zara: @virgilvox
Ben Smith: @benjamin_l_s
Rob Sutter: @rts_rob
Julian Wood: @julian_wood

Happy coding!

re:Invent 2019: Introducing the Amazon Builders’ Library (Part I)

Post Syndicated from Annik Stahl original https://aws.amazon.com/blogs/architecture/reinvent-2019-introducing-the-amazon-builders-library-part-i/

Today, I’m going to tell you about a new site we launched at re:Invent, the Amazon Builders’ Library, a collection of living articles covering topics across architecture, software delivery, and operations. You get to peek under the hood of how Amazon architects, releases, and operates the software underpinning Amazon.com and AWS.

Want to know how Amazon.com does what it does? This is for you. In this two-part series (the next one coming December 23), I’ll highlight some of the best architecture articles written by Amazon’s senior technical leaders and engineers.

Avoiding insurmountable queue backlogs

Avoiding insurmountable queue backlogs

In queueing theory, the behavior of queues when they are short is relatively uninteresting. After all, when a queue is short, everyone is happy. It’s only when the queue is backlogged, when the line to an event goes out the door and around the corner, that people start thinking about throughput and prioritization.

In this article, I discuss strategies we use at Amazon to deal with queue backlog scenarios – design approaches we take to drain queues quickly and to prioritize workloads. Most importantly, I describe how to prevent queue backlogs from building up in the first place. In the first half, I describe scenarios that lead to backlogs, and in the second half, I describe many approaches used at Amazon to avoid backlogs or deal with them gracefully.

Read the full article by David Yanacek – Principal Engineer

Timeouts, retries, and backoff with jitter

Timeouts, retries and backoff with jitter

Whenever one service or system calls another, failures can happen. These failures can come from a variety of factors. They include servers, networks, load balancers, software, operating systems, or even mistakes from system operators. We design our systems to reduce the probability of failure, but impossible to build systems that never fail. So in Amazon, we design our systems to tolerate and reduce the probability of failure, and avoid magnifying a small percentage of failures into a complete outage. To build resilient systems, we employ three essential tools: timeouts, retries, and backoff.

Read the full article by Marc Brooker, Senior Principal Engineer

Challenges with distributed systems

Challenges with distributed systems

The moment we added our second server, distributed systems became the way of life at Amazon. When I started at Amazon in 1999, we had so few servers that we could give some of them recognizable names like “fishy” or “online-01”. However, even in 1999, distributed computing was not easy. Then as now, challenges with distributed systems involved latency, scaling, understanding networking APIs, marshalling and unmarshalling data, and the complexity of algorithms such as Paxos. As the systems quickly grew larger and more distributed, what had been theoretical edge cases turned into regular occurrences.

Developing distributed utility computing services, such as reliable long-distance telephone networks, or Amazon Web Services (AWS) services, is hard. Distributed computing is also weirder and less intuitive than other forms of computing because of two interrelated problems. Independent failures and nondeterminism cause the most impactful issues in distributed systems. In addition to the typical computing failures most engineers are used to, failures in distributed systems can occur in many other ways. What’s worse, it’s impossible always to know whether something failed.

Read the full article by Jacob Gabrielson, Senior Principal Engineer

Static stability using Availability Zones

Static stability using availability zones

At Amazon, the services we build must meet extremely high availability targets. This means that we need to think carefully about the dependencies that our systems take. We design our systems to stay resilient even when those dependencies are impaired. In this article, we’ll define a pattern that we use called static stability to achieve this level of resilience. We’ll show you how we apply this concept to Availability Zones, a key infrastructure building block in AWS and therefore a bedrock dependency on which all of our services are built.

Read the full article by Becky Weiss, Senior Principal Engineer, and Mike Furr, Principal Engineer

Check back in two weeks to read about some other architecture-based expert articles that let you in on how Amazon does what it does.

ICYMI: Serverless pre:Invent 2019

Post Syndicated from Eric Johnson original https://aws.amazon.com/blogs/compute/icymi-serverless-preinvent-2019/

With Contributions from Chris Munns – Sr Manager – Developer Advocacy – AWS Serverless

The last two weeks have been a frenzy of AWS service and feature launches, building up to AWS re:Invent 2019. As there has been a lot announced we thought we’d ship an ICYMI post summarizing the serverless service specific features that have been announced. We’ve also dropped in some service announcements from services that are commonly used in serverless application architectures or development.

AWS re:Invent

AWS re:Invent 2019

We also want you to know that we’ll be talking about many of these features (as well as those coming) in sessions at re:Invent.

Here’s what’s new!

AWS Lambda

On September 3, AWS Lambda started rolling out a major improvement to how AWS Lambda functions work with your Amazon VPC networks. This change brings both scale and performance improvements, and addresses several of the limitations of the previous networking model with VPCs.

On November 25, Lambda announced that the rollout of this new capability has completed in 6 additional regions including US East (Virginia) and US West (Oregon).

New VPC to VPC NAT for Lambda functions

New VPC to VPC NAT for Lambda functions

On November 18, Lambda announced three new runtime updates. Lambda now supports Node.js 12, Java 11, and Python 3.8. Each of these new runtimes has new language features and benefits so be sure to check out the linked release posts. These new runtimes are all based on an Amazon Linux 2 execution environment.

Lambda has released a number of controls for both stream and async based invocations:

  • For Lambda functions consuming events from Amazon Kinesis Data Streams or Amazon DynamoDB Streams, it’s now possible to limit the retry count, limit the age of records being retried, configure a failure destination, or split a batch to isolate a problem record. These capabilities will help you deal with potential “poison pill” records that would previously cause streams to pause in processing.
  • For asynchronous Lambda invocations, you can now set the maximum event age and retry attempts on the event. If either configured condition is met, the event can be routed to a dead letter queue (DLQ), Lambda destination, or it can be discarded.

In addition to the above controls, Lambda Destinations is a new feature that allows developers to designate an asynchronous target for Lambda function invocation results. You can set one destination for a success, and another for a failure. This unlocks really useful patterns for distributed event-based applications and can reduce code to send function results to a destination manually.

Lambda Destinations

Lambda Destinations

Lambda also now supports setting a Parallelization Factor, which allows you to set multiple Lambda invocations per shard for Amazon Kinesis Data Streams and Amazon DynamoDB Streams. This allows for faster processing without the need to increase your shard count, while still guaranteeing the order of records processed.

Lambda Parallelization Factor diagram

Lambda Parallelization Factor diagram

Lambda now supports Amazon SQS FIFO queues as an event source. FIFO queues guarantee the order of record processing compared to standard queues which are unordered. FIFO queues support messaging batching via a MessageGroupID attribute which allows for parallel Lambda consumers of a single FIFO queue. This allows for high throughput of record processing by Lambda.

Lambda now supports Environment Variables in AWS China (Beijing) Region, operated by Sinnet and the AWS China (Ningxia) Region, operated by NWCD.

Lastly, you can now view percentile statistics for the duration metric of your Lambda functions. Percentile statistics tell you the relative standing of a value in a dataset, and are useful when applied to metrics that exhibit large variances. They can help you understand the distribution of a metric, spot outliers, and find hard-to-spot situations that create a poor customer experience for a subset of your users.

AWS SAM CLI

AWS SAM CLI deploy command

AWS SAM CLI deploy command

The SAM CLI team simplified the bucket management and deployment process in the SAM CLI. You no longer need to manage a bucket for deployment artifacts – SAM CLI handles this for you. The deployment process has also been streamlined from multiple flagged commands to a single command, sam deploy.

AWS Step Functions

One of the powerful features of Step Functions is its ability to integrate directly with AWS services without you needing to write complicated application code. Step Functions has expanded its integration with Amazon SageMaker to simplify machine learning workflows, and added a new integration with Amazon EMR, making it faster to build and easier to monitor EMR big data processing workflows.

Step Functions step with EMR

Step Functions step with EMR

Step Functions now provides the ability to track state transition usage by integrating with AWS Budgets, allowing you to monitor and react to usage and spending trends on your AWS accounts.

You can now view CloudWatch Metrics for Step Functions at a one-minute frequency. This makes it easier to set up detailed monitoring for your workflows. You can use one-minute metrics to set up CloudWatch Alarms based on your Step Functions API usage, Lambda functions, service integrations, and execution details.

AWS Step Functions now supports higher throughput workflows, making it easier to coordinate applications with high event rates.

In US East (N. Virginia), US West (Oregon), and EU (Ireland), throughput has increased from 1,000 state transitions per second to 1,500 state transitions per second with bucket capacity of 5,000 state transitions. The default start rate for state machine executions has also increased from 200 per second to 300 per second, with bucket capacity of up to 1,300 starts in these regions.

In all other regions, throughput has increased from 400 state transitions per second to 500 state transitions per second with bucket capacity of 800 state transitions. The default start rate for AWS Step Functions state machine executions has also increased from 25 per second to 150 per second, with bucket capacity of up to 800 state machine executions.

Amazon SNS

Amazon SNS now supports the use of dead letter queues (DLQ) to help capture unhandled events. By enabling a DLQ, you can catch events that are not processed and re-submit them or analyze to locate processing issues.

Amazon CloudWatch

CloudWatch announced Amazon CloudWatch ServiceLens to provide a “single pane of glass” to observe health, performance, and availability of your application.

CloudWatch ServiceLens

CloudWatch ServiceLens

CloudWatch also announced a preview of a capability called Synthetics. CloudWatch Synthetics allows you to test your application endpoints and URLs using configurable scripts that mimic what a real customer would do. This enables the outside-in view of your customers’ experiences, and your service’s availability from their point of view.

On November 18, CloudWatch launched Embedded Metric Format to help you ingest complex high-cardinality application data in the form of logs and easily generate actionable metrics from them. You can publish these metrics from your Lambda function by using the PutLogEvents API or for Node.js or Python based applications using an open source library.

Lastly, CloudWatch announced a preview of Contributor Insights, a capability to identify who or what is impacting your system or application performance by identifying outliers or patterns in log data.

AWS X-Ray

X-Ray announced trace maps, which enable you to map the end to end path of a single request. Identifiers will show issues and how they affect other services in the request’s path. These can help you to identify and isolate service points that are causing degradation or failures.

X-Ray also announced support for Amazon CloudWatch Synthetics, currently in preview. X-Ray supports tracing canary scripts throughout the application providing metrics on performance or application issues.

X-Ray Service map with CloudWatch Synthetics

X-Ray Service map with CloudWatch Synthetics

Amazon DynamoDB

DynamoDB announced support for customer managed customer master keys (CMKs) to encrypt data in DynamoDB. This allows customers to bring your own key (BYOK) giving you full control over how you encrypt and manage the security of your DynamoDB data.

It is now possible to add global replicas to existing DynamoDB tables to provide enhanced availability across the globe.

Currently under preview, is another new DynamoDB capability to identify frequently accessed keys and database traffic trends. With this you can now more easily identify “hot keys” and understand usage of your DynamoDB tables.

CloudWatch Contributor Insights for DynamoDB

CloudWatch Contributor Insights for DynamoDB

Last but far from least for DynamoDB, is adaptive capacity, a feature which helps you handle imbalanced workloads by isolating frequently accessed items automatically and shifting data across partitions to rebalance them. This helps both reduce cost by enabling you to provision throughput for a more balanced out workload vs. over provisioning for uneven data access patterns.

AWS Serverless Application Repository

The AWS Serverless Application Repository (SAR) now offers Verified Author badges. These badges enable consumers to quickly and reliably know who you are. The badge will appear next to your name in the SAR and will deep-link to your GitHub profile.

SAR Verified developer badges

SAR Verified developer badges

AWS Code Services

AWS CodeCommit launched the ability for you to enforce rule workflows for pull requests, making it easier to ensure that code has pass through specific rule requirements. You can now create an approval rule specifically for a pull request, or create approval rule templates to be applied to all future pull requests in a repository.

AWS CodeBuild added beta support for test reporting. With test reporting, you can now view the detailed results, trends, and history for tests executed on CodeBuild for any framework that supports the JUnit XML or Cucumber JSON test format.

CodeBuild test trends

CodeBuild test trends

AWS Amplify and AWS AppSync

Instead of trying to summarize all the awesome things that our peers over in the Amplify and AppSync teams have done recently we’ll instead link you to their own recent summary: “A round up of the recent pre-re:Invent 2019 AWS Amplify Launches”.

AWS AppSync

AWS AppSync

Still looking for more?

We only covered a small bit of all the awesome new things that were recently announced. Keep your eyes peeled for more exciting announcements next week during re:Invent and for a future ICYMI Serverless Q4 roundup. We’ll also be kicking off a fresh series of Tech Talks in 2020 with new content helping to dive deeper on everything new coming out of AWS for serverless application developers.

Running Cost-effective queue workers with Amazon SQS and Amazon EC2 Spot Instances

Post Syndicated from peven original https://aws.amazon.com/blogs/compute/running-cost-effective-queue-workers-with-amazon-sqs-and-amazon-ec2-spot-instances/

This post is contributed by Ran Sheinberg | Sr. Solutions Architect, EC2 Spot & Chad Schmutzer | Principal Developer Advocate, EC2 Spot | Twitter: @schmutze

Introduction

Amazon Simple Queue Service (SQS) is used by customers to run decoupled workloads in the AWS Cloud as a best practice, in order to increase their applications’ resilience. You can use a worker tier to do background processing of images, audio, documents and so on, as well as offload long-running processes from the web tier. This blog post covers the benefits of pairing Amazon SQS and Spot Instances to maximize cost savings in the worker tier, and a customer success story.

Solution Overview

Amazon SQS is a fully managed message queuing service that enables customers to decouple and scale microservices, distributed systems, and serverless applications. It is a common best practice to use Amazon SQS with decoupled applications. Amazon SQS increases applications resilience by decoupling the direct communication between the frontend application and the worker tier that does data processing. If a worker node fails, the jobs that were running on that node return to the Amazon SQS queue for a different node to pick up.

Both the frontend and worker tier can run on Spot Instances, which offer spare compute capacity at steep discounts compared to On-Demand Instances. Spot Instances optimize your costs on the AWS Cloud and scale your application’s throughput up to 10 times for the same budget. Spot Instances can be interrupted with two minutes of notification when EC2 needs the capacity back. You can use Spot Instances for various fault-tolerant and flexible applications. These can include analytics, containerized workloads, high performance computing (HPC), stateless web servers, rendering, CI/CD, and queue worker nodes—which is the focus of this post.

Worker tiers of a decoupled application are typically fault-tolerant. So, it is a prime candidate for running on interruptible capacity. Amazon SQS running on Spot Instances allows for more robust, cost-optimized applications.

By using EC2 Auto Scaling groups with multiple instance types that you configured as suitable for your application (for example, m4.xlarge, m5.xlarge, c5.xlarge, and c4.xlarge, in multiple Availability Zones), you can spread the worker tier’s compute capacity across many Spot capacity pools (a combination of instance type and Availability Zone). This increases the chance of achieving the scale that’s required for the worker tier to ingest messages from the queue, and of keeping that scale when Spot Instance interruptions occur, while selecting the lowest-priced Spot Instances in each availability zone.

You can also choose the capacity-optimized allocation strategy for the Spot Instances in your Auto Scaling group. This strategy automatically selects instances that have a lower chance of interruption, which decreases the chances of restarting jobs due to Spot interruptions. When Spot Instances are interrupted, your Auto Scaling group automatically replenishes the capacity from a different Spot capacity pool in order to achieve your desired capacity. Read the blog post “Introducing the capacity-optimized allocation strategy for Amazon EC2 Spot Instances” for more details on how to choose the suitable allocation strategy.

We focus on three main points in this blog:

  1. Best practices for using Spot Instances with Amazon SQS
  2. A customer example that uses these components
  3. Example solution that can help you get you started quickly

Application of Amazon SQS with Spot Instances

Amazon SQS eliminates the complexity of managing and operating message-oriented middleware. Using Amazon SQS, you can send, store, and receive messages between software components at any volume, without losing messages or requiring other services to be available. Amazon SQS is a fully managed service which allows you to set up a queue in seconds. It also allows you to use your preferred SDK to start writing and reading to and from the queue within minutes.

In the following example, we describe an AWS architecture that brings together the Amazon SQS queue and an EC2 Auto Scaling group running Spot Instances. The architecture is used for decoupling the worker tier from the web tier by using Amazon SQS. The example uses the Protect feature (which we will explain later in this post) to ensure that an instance currently processing a job does not get terminated by the Auto Scaling group when it detects that a scale-in activity is required due to a Dynamic Scaling Policy.Architecture diagram for using Amazon SQS with Spot Instances and Auto Scaling groups

AWS reference architecture used for decoupling the worker tier from the web tier by using Amazon SQS

Customer Example: How Trax Retail uses Auto Scaling groups with Spot Instances in their Amazon SQS application

Trax decided to run its queue worker tier exclusively on Spot Instances due to the fault-tolerant nature of its architecture and for cost-optimization purposes. The company digitizes the physical world of retail using Computer Vision. Their ‘Trax Factory’ transforms individual shelf into data and insights about retail store conditions.

Built using asynchronous event-driven architecture, Trax Factory is a cluster of microservices in which the completion of one service triggers the activation of another service. The worker tier uses Auto Scaling groups with dynamic scaling policies to increase and decrease the number of worker nodes in the worker tier.

You can create a Dynamic Scaling Policy by doing the following:

  1. Observe a Amazon CloudWatch metric. Watch the metric for the current number of messages in the Amazon SQS queue (ApproximateNumberOfMessagesVisible).
  2. Create a CloudWatch alarm. This alarm should be based on that metric you created in the prior step.
  3. Use your CloudWatch alarm in a Dynamic Scaling Policy. Use this policy increase and decrease the number of EC2 Instances in the Auto Scaling group.

In Trax’s case, due to the high variability of the number of messages in the queue, they opted to enhance this approach in order to minimize the time it takes to scale, by building a service that would call the SQS API and find the current number of messages in the queue more frequently, instead of waiting for the 5 minute metric refresh interval in CloudWatch.

Trax ensures that its applications are always scaled to meet the demand by leveraging the inherent elasticity of Amazon EC2 instances. This elasticity ensures that end users are never affected and/or service-level agreements (SLA) are never violated.

With a Dynamic Scaling Policy, the Auto Scaling group can detect when the number of messages in the queue has decreased, so that it can initiate a scale-in activity. The Auto Scaling group uses its configured termination policy for selecting the instances to be terminated. However, this policy poses the risk that the Auto Scaling group might select an instance for termination while that instance is currently processing an image. That instance’s work would be lost (although the image would eventually be processed by reappearing in the queue and getting picked up by another worker node).

To decrease this risk, you can use Auto Scaling groups instance protection. This means that every time an instance fetches a job from the queue, it also sends an API call to EC2 to protect itself from scale-in. The Auto Scaling group does not select the protected, working instance for termination until the instance finishes processing the job and calls the API to remove the protection.

Handling Spot Instance interruptions

This instance-protection solution ensures that no work is lost during scale-in activities. However, protecting from scale-in does not work when an instance is marked for termination due to Spot Instance interruptions. These interruptions occur when there’s increased demand for On-Demand Instances in the same capacity pool (a combination of an instance type in an Availability Zone).

Applications can minimize the impact of a Spot Instance interruption. To do so, an application catches the two-minute interruption notification (available in the instance’s metadata), and instructs itself to stop fetching jobs from the queue. If there’s an image still being processed when the two minutes expire and the instance is terminated, the application does not delete the message from the queue after finishing the process. Instead, the message simply becomes visible again for another instance to pick up and process after the Amazon SQS visibility timeout expires.

Alternatively, you can release any ongoing job back to the queue upon receiving a Spot Instance interruption notification by setting the visibility timeout of the specific message to 0. This timeout potentially decreases the total time it takes to process the message.

Testing the solution

If you’re not currently using Spot Instances in your queue worker tier, we suggest testing the approach described in this post.

For that purpose, we built a simple solution to demonstrate the capabilities mentioned in this post, using an AWS CloudFormation template. The stack includes an Amazon Simple Storage Service (S3) bucket with a CloudWatch trigger to push notifications to an SQS queue after an image is uploaded to the Amazon S3 bucket. Once the message is in the queue, it is picked up by the application running on the EC2 instances in the Auto Scaling group. Then, the image is converted to PDF, and the instance is protected from scale-in for as long as it has an active processing job.

To see the solution in action, deploy the CloudFormation template. Then upload an image to the Amazon S3 bucket. In the Auto Scaling Groups console, check the instance protection status on the Instances tab. The protection status is shown in the following screenshot.

instance protection status in console

You can also see the application logs using CloudWatch Logs:

/usr/local/bin/convert-worker.sh: Found 1 messages in https://sqs.us-east-1.amazonaws.com/123456789012/qtest-sqsQueue-1CL0NYLMX64OB

/usr/local/bin/convert-worker.sh: Found work to convert. Details: INPUT=Capture1.PNG, FNAME=capture1, FEXT=png

/usr/local/bin/convert-worker.sh: Running: aws autoscaling set-instance-protection --instance-ids i-0a184c5ae289b2990 --auto-scaling-group-name qtest-autoScalingGroup-QTGZX5N70POL --protected-from-scale-in

/usr/local/bin/convert-worker.sh: Convert done. Copying to S3 and cleaning up

/usr/local/bin/convert-worker.sh: Running: aws s3 cp /tmp/capture1.pdf s3://qtest-s3bucket-18fdpm2j17wxx

/usr/local/bin/convert-worker.sh: Running: aws sqs --output=json delete-message --queue-url https://sqs.us-east-1.amazonaws.com/123456789012/qtest-sqsQueue-1CL0NYLMX64OB --receipt-handle

/usr/local/bin/convert-worker.sh: Running: aws autoscaling set-instance-protection --instance-ids i-0a184c5ae289b2990 --auto-scaling-group-name qtest-autoScalingGroup-QTGZX5N70POL --no-protected-from-scale-in

Conclusion

This post helps you architect fault tolerant worker tiers in a cost optimized way. If your queue worker tiers are fault tolerant and use the built-in Amazon SQS features, you can increase your application’s resilience and take advantage of Spot Instances to save up to 90% on compute costs.

In this post, we emphasized several best practices to help get you started saving money using Amazon SQS and Spot Instances. The main best practices are:

  • Diversifying your Spot Instances using Auto Scaling groups, and selecting the right Spot allocation strategy
  • Protecting instances from scale-in activities while they process jobs
  • Using the Spot interruption notification so that the application stop polling the queue for new jobs before the instance is terminated

We hope you found this post useful. If you’re not using Spot Instances in your queue worker tier, we suggest testing the approach described here. Finally, we would like to thank the Trax team for sharing its architecture and best practices. If you want to learn more, watch the “This is my architecture” video featuring Trax and their solution.

We’d love your feedback—please comment and let me know what you think.


About the authors

 

Ran Sheinberg is a specialist solutions architect for EC2 Spot Instances with Amazon Web Services. He works with AWS customers on cost optimizing their compute spend by utilizing Spot Instances across different types of workloads: stateless web applications, queue workers, containerized workloads, analytics, HPC and others.

 

 

 

 

As a Principal Developer Advocate for EC2 Spot at AWS, Chad’s job is to make sure our customers are saving at scale by using EC2 Spot Instances to take advantage of the most cost-effective way to purchase compute capacity. Follow him on Twitter here! @schmutze

 

New AWS Lambda controls for stream processing and asynchronous invocations

Post Syndicated from Benjamin Smith original https://aws.amazon.com/blogs/compute/new-aws-lambda-controls-for-stream-processing-and-asynchronous-invocations/

Today AWS Lambda is introducing new controls for asynchronous and stream processing invocations. These new features allow you to customize responses to Lambda function errors and build more resilient event-driven and stream-processing applications.

Stream processing function invocations

When processing data from event sources such as Amazon Kinesis Data Streams, and Amazon DynamoDB Streams, Lambda reads records in batches via shards. A shard is a uniquely identified sequence of data records. Your function is then invoked to process records from the batch “in order.” If an error is returned, Lambda retries the batch until processing succeeds or the data expires. This retry behavior is desirable in many cases, but not all:

  1. Until the issue is resolved, no data in the shard is processed. A single malformed record can prevent processing on an entire shard since “in order” guarantee ensures that failed batches are retried until the data record expires. This is often referred to as a “poison pill” event in that it prevents the rest of the system from processing data.
  2. In some cases, it may not be helpful to retry if subsequent invocations will also fail.
  3. If the function is not idempotent, retrying might produce unexpected side effects, and may result in duplicate outputs (for example, multiple database entries or business transactions).
Stream processing function invocationsFigure 1: Default stream invocation retry logs.

The new Lambda controls for stream processing invocations

With the new customizable controls, users are able to control how function errors and retries impact stream processing function invocations. A new event source-mapping configuration subresource (shown below), allows developers to clearly specify which controls will apply specifically to invocations.

DestinationConfig: {
    OnFailure: {
       Destination: SNS/SQS arn (String)
    }
}
Figure 2: Stream processing destination configuration.
{
    "MaximumRetryAttempts": integer,
    "BisectBatchOnFunctionError" : boolean,
    "MaximumRecordAgeInSeconds" : integer
}
Figure 3: Stream processing failure configuration.

MaximumRetryAttempts

Set a maximum number of retry attempts for batches before they can be skipped to unblock processing and prevent duplicate outputs.

Minimum: 0 | maximum: 10,000 | default: 10,000

MaximumRecordAgeInSeconds

Define a record maximum age in seconds, with expired records skipped to allow processing to continue. Data records that do not get successfully processed within the defined age of submission are skipped

Minimum: 60 | default/maximum: 604,800

BisectBatchOnFunctionError

This gives the option to recursively split the failed batch and retry on a smaller subset of records, eventually isolating the metadata causing the error.

Default: false

On-failure Destination

When either MaximumRetryAttempts or MaximumRecordAgeInSeconds reaches the specified value, a record will be skipped. If ‘Destination’ is set, metadata about the skipped records can be sent to a target ARN (i.e. Amazon SQS or Amazon SNS). If no target is configured, then the record will be dropped.

Getting started with error handling for stream processing invocations

Here you will see how to use the new controls to create a customized error handling configuration for stream processing invocations. You will set the maximum retry attempts to 1, the maximum record age to 60s, and send the metadata of skipped record to an Amazon Simple Queue Service (SQS) queue. First create a Kinesis Stream to put records into, and create an SQS queue to hold the metadata of retry exhausted or expired data records.

  1. Go to the Lambda console and choose Create function.
  2. Ensure that Author from scratch is selected, and enter a function a name such as “customStreamErrorExample.”
  3. Select Runtime as Node.js.12.x.
  4. Choose Create function.
    To enable Kinesis as an event trigger and to send metadata of a failed invocation to the SQS standard queue, you must give the Lambda execution role the required permissions.
  5. Scroll down to the Execution Role section and select the view link below the Existing role drop-down.
    Existing role selection
  6. Choose Add inline policy > JSON, then paste the following into the text box replacing {yourSQSarn} with the ARN of the SQS queue you created earlier and replacing {yourKinesisarn} with the ARN of the stream you created earlier. Choose Review policy.
    {
        "Version": "2012-10-17",
        "Statement": [
            {
                "Sid": "VisualEditor0",
                "Effect": "Allow",
                "Action": [
                    "sqs:SendMessage",
                    "kinesis:GetShardIterator",
                    "kinesis:GetRecords",
                    "kinesis:DescribeStream"
                ],
                "Resource": [
                    "{yourSQSarn}",
                    "{yourKinesisarn}"
                ]
            },
            {
                "Sid": "VisualEditor1",
                "Effect": "Allow",
                "Action": "kinesis:ListStreams",
                "Resource": "*"
            }
        ]
    }
    
  7. Give the policy a name such as CustomSQSKinesis and choose Create policy.
  8. Navigate back to your Lambda function and choose Add trigger.
  9. Select Kinesis from the drop-down. Select your stream in the Kinesis stream dropdown.
  10. To see the new control options, choose the drop-down arrow on the Additional settings section.
  11. Paste the ARN of the previously created SQS queue (refer to this link to find the ARN).Additional settings
  12. Set the Maximum retry attempts to 1. Set the Maximum record age to 60. Choose Add.
    The Kinesis trigger has now been configured and the Lambda function has the required permissions to write to SQS and CloudWatch Logs.Success messageKinesis trigger added
  13. Choose the Lambda icon in the Designer section. Scroll down to the Function code section and paste the following code:
    exports.handler = async (event) => {
        // TODO implement
        console.log(event)
        const response = {
            statusCode: 200,
            body: event,dfh
        };
        return response;
    };

    This code has a syntax error in order to force a failure response.

    Next, you will trigger the Lambda function by adding a record to the Kinesis Stream via the AWS CLI. See how to install the AWS CLI if you have not previously done so.

  14. In your preferred terminal run the following command, replacing {YourStreamName} with the name of the stream you have created.
    aws kinesis put-record --stream-name {YourStreamName} --partition-key 123 --data testdata

    You should see a response similar to this (your sequence number will be different):
    Terminal response

  15. In the AWS Lambda console, choose Monitoring > View logs in CloudWatch and select the most recent log group.CloudWatch Logs
  16. As you can see the Lambda function failed as expected, but this time with only a single retry attempt. This is because the max retry attempt value was set to 1 and the record is younger than the max record age that was set to 60.
  17. In the AWS Management Console navigate to your SQS queue, Services > SQS.
  18. Select your queue and choose Queue Actions > View/Delete Messages > Start Polling for Messages.View/delete messages

Here you will see the metadata of your failed invocation. It has successfully been sent to the SQS destination for further investigation or processing.

Asynchronous function invocations

When a function is asynchronously invoked, Lambda sends the event to a queue before it is processed by your function. Invocations that result in an exception in the function code are retried twice with a delay of one minute before the first retry and two minutes before the second. Some invocations may never run due to a throttle or service fault: these are retried with exponential backoff until the invocation is six hours old.

Default asynchronous invocation retry logs

This retry behaviour shown above, works well in situations where every invocation must run at least once. However, in situations with a large volume of errors subsequent invocations are increasingly delayed as the backlog increases, as each new error places another retry event back into the event queue. If it were possible to skip the event without retrying, it would eliminate this delay and continue processing new events.

In order to avoid this default auto-retry policy, there are several widely used approaches:

  • Function error handling with AWS Step Functions
  • Using the event handler for error routing logic.
  • Third party monitoring services for exception alerts

These workarounds require extra resources, code, or custom logic and add latency to your applications. Retries caused by system failures due to timeouts, lack of memory, early exits, etc. could still occur.

New Lambda controls for asynchronous event invocations

New asynchronous event invoke controls mean that functions can now skip the processing of certain events and discard unwanted or backlogged requests from the asynchronous event queue without the need for “workarounds.” The controls will be accessible from the console and from a new Event Invoke Config subresource, listed in detail below:

{
    "MaximumEventAgeInSeconds": integer,
    "MaximumRetryAttempts": integer
}
Figure 4: Asynchronous event invoke configuration.

MaximumRetryAttempts

Set a maximum number of retry attempts for events before they can be skipped to unblock processing and prevent duplicate outputs.

Minimum: 0 | maximum: 2 | default: 2

MaximumEventAgeInSeconds

Define an event maximum age in seconds, with expired events skipped to allow processing to continue. Events that do not get successfully processed within defined age of submission are written to the function’s configured Dead Letter Queue and/or On-failure Destinations for asynchronous invocations. If none is configured, the event is discarded.

Minimum: 60 | maximum: 21600 (6 hours)

These controls can also be configured from the AWS Lambda console, in the “Failure handling configuration” section shown below:

Edit failure handling configurations

When the same Lambda function is run with maximum retry attempts set to 0, the Amazon CloudWatch Logs show that the function is only invoked a single time, and not retried after the initial error.

CloudWatch Logs

Conclusion

The addition of these new Lambda controls allows you to handle function failures and retries appropriately for both asynchronous and stream processing. This eliminates the need for custom-built logic, alerts, and added overhead.

Developers understand the needs of their own applications. These new features allow them to decide how to react to errors on a per-function basis. Whether that means real-time stream processing, non-blocking event queues, or more retries, it depends on the application’s requirements. With the new controls in place, the power is in your hands.

You can get started with the new controls for stream processing and asynchronous invocations via the AWS Management Console, AWS CLI, AWS SAM, AWS CloudFormation, or AWS SDK for Lambda.

Introducing AWS Lambda Destinations

Post Syndicated from Julian Wood original https://aws.amazon.com/blogs/compute/introducing-aws-lambda-destinations/

Today we’re announcing AWS Lambda Destinations for asynchronous invocations. This is a feature that provides visibility into Lambda function invocations and routes the execution results to AWS services, simplifying event-driven applications and reducing code complexity.

Asynchronous invocations

When a function is invoked asynchronously, Lambda sends the event to an internal queue. A separate process reads events from the queue and executes your Lambda function. When the event is added to the queue, Lambda previously only returned a 2xx status code to confirm that the queue has received this event. There was no additional information to confirm whether the event had been processed successfully.

A common event-driven microservices architectural pattern is to use a queue or message bus for communication. This helps with resilience and scalability. Lambda asynchronous invocations can put an event or message on Amazon Simple Notification Service (SNS), Amazon Simple Queue Service (SQS), or Amazon EventBridge for further processing. Previously, you needed to write the SQS/SNS/EventBridge handling code within your Lambda function and manage retries and failures yourself.

With Destinations, you can route asynchronous function results as an execution record to a destination resource without writing additional code. An execution record contains details about the request and response in JSON format including version, timestamp, request context, request payload, response context, and response payload. For each execution status such as Success or Failure you can choose one of four destinations: another Lambda function, SNS, SQS, or EventBridge. Lambda can also be configured to route different execution results to different destinations.

Asynchronous Function Execution Result

Success

When a function is invoked successfully, Lambda routes the record to the destination resource for every successful invocation. You can use this to monitor the health of your serverless applications via execution status or build workflows based on the invocation result.

You no longer need to chain long-running Lambda functions together synchronously. Previously you needed to complete the entire workflow within the Lambda 15-minute function timeout, pay for idle time, and wait for a response. Destinations allows you to return a Success response to the calling function and then handle the remaining chaining functions asynchronously.

Failure

Alongside today’s announcement of Maximum Event Age and Maximum Retry Attempt for asynchronous invocations, Destinations gives you the ability to handle the Failure of function invocations along with their Success. When a function invocation fails, such as when retries are exhausted or the event age has been exceeded (hitting its TTL), Destinations routes the record to the destination resource for every failed invocation for further investigation or processing.

Dead Letter Queues (DLQ) have been available since 2016 and are a great way to handle asynchronous failure situations. Destinations provide more useful capabilities by passing additional function execution information, including code exception stack traces, to more destination services.

Destinations and DLQs can be used together and at the same time although Destinations should be considered a more preferred solution. If you already have DLQs set up, existing functionality does not change and Destinations does not replace existing DLQ configurations. If both Destinations and DLQ are used for Failure notifications, function invoke errors are sent to both DLQ and Destinations targets.

How to configure Destinations

Adding Destinations is a straightforward process. This walkthrough uses the AWS Management Console but you can also use the AWS CLI, AWS SAM, AWS CloudFormation, or language-specific SDKs for Lambda.

  1. Open the Lambda console Functions page. Choose an existing Lambda function, or create a new one. In this example, I create a new Lambda function. Choose Create Function.
  2. Enter a Function name, select Node.js 12.x for Runtime, and Choose or create an execution role. Ensure that your Lambda function execution role includes access to the destination resource.
    Basic information
  3. Choose Create function.
  4. Within the Function code pane, paste the following Lambda function code. The code generates a function execution result of either Success or Failure depending on a JSON input ("Success": true or "Success": false).
    // Lambda Destinations tester, Success returns a json blob, Failure throws an error
    
    exports.handler = function(event, context, callback) {
        var event_received_at = new Date().toISOString();
        console.log('Event received at: ' + event_received_at);
        console.log('Received event:', JSON.stringify(event, null, 2));
    
        if (event.Success) {
            console.log("Success");
            context.callbackWaitsForEmptyEventLoop = false;
            callback(null);
        } else {
            console.log("Failure");
            context.callbackWaitsForEmptyEventLoop = false;
            callback(new Error("Failure from event, Success = false, I am failing!"), 'Destination Function Error Thrown');
        }
    };
    
  5. Choose Save.
  6. To configure Destinations, within the Designer pane, choose Add destination.
    Designer pane
  7. Select the Source as Asynchronous invocation. Select the Condition as On failure or On success, depending on your use case. In this example, I select On Success.
  8. Enter the Amazon Resource Name (ARN) for the Destination SQS queue, SNS topic, Lambda function, or EventBridge event bus. In this example, I use the ARN of an SNS topic I have already configured.
    Add destination
  9. Choose Save. The Destination is added to SNS for On Success.
    Designer
  10. Add another Destination for Failure to Lambda. Within the Designer pane, choose Add destination.
    Add destination
  11. Select the Source as Asynchronous invocation, the Condition as On failure and Enter a Destination Lambda function ARN, then choose Save.
    Enter a Destination Lambda function ARN, and choose Save
  12. The Destination is added to Lambda for On Failure.
    7. The Destination has been added to Lambda for On Failure.

Success testing

To test invoking the asynchronous Lambda function to generate a Success result, use the AWS CLI:

aws lambda invoke --function-name event-destinations --invocation-type Event --payload '{ "Success": true }' response.json

The Lambda function is invoked successfully with a response "StatusCode": 202.

And an SNS notification email is received, showing the invocation details with "condition":"Success" and the requestPayload.

{
	"version": "1.0",
	"timestamp": "2019-11-24T23:08:25.651Z",
	"requestContext": {
		"requestId": "c2a6f2ae-7dbb-4d22-8782-d0485c9877e2",
		"functionArn": "arn:aws:lambda:sa-east-1:123456789123:function:event-destinations:$LATEST",
		"condition": "Success",
		"approximateInvokeCount": 1
	},
	"requestPayload": {
		"Success": true
	},
	"responseContext": {
		"statusCode": 200,
		"executedVersion": "$LATEST"
	},
	"responsePayload": null
}

Failure testing

The Lambda function can be set to Failure by throwing an exception within the code. To test invoking the asynchronous Lambda function to generate a Failure result, use the AWS CLI:

aws lambda invoke --function-name event-destinations --invocation-type Event --payload '{ "Success": false }' response.json

The Lambda function is executed and reports a successful invoke on the Lambda processing queue. If Lambda is not able to add the event to the queue, the error message appears in the command output.

However, due to the exception error within the code, the function invocation will fail. Destinations then routes the invoke failure to the configured destination Lambda function. You can see the failed function invocation information in the Amazon CloudWatch Logs for the Destination function including "condition": "RetriesExhausted", along with the requestPayload, errorMessage, and stackTrace.

2019-11-24T21:52:47.855Z	d123456-c0dd-4871-a123-a356cb1b3ba6	EVENT
{
    "version": "1.0",
    "timestamp": "2019-11-24T21:52:47.333Z",
    "requestContext": {
        "requestId": "8ea123e4-1db7-4aca-ad10-d9ca1234c1fd",
        "functionArn": "arn:aws:lambda:sa-east-1:123456678912:function:event-destinations:$LATEST",
        "condition": "RetriesExhausted",
        "approximateInvokeCount": 3
    },
    "requestPayload": {
        "Success": false
    },
    "responseContext": {
        "statusCode": 200,
        "executedVersion": "$LATEST",
        "functionError": "Handled"
    },
    "responsePayload": {
        "errorMessage": "Failure from event, Success = false, I am failing!",
        "errorType": "Error",
        "stackTrace": [ "exports.handler (/var/task/index.js:18:18)" ]
    }
}

Destination-specific JSON format

  • For SNS/SQS, the JSON object is passed as the Message to the destination.
  • For Lambda, the JSON is passed as the payload to the function. The destination function cannot be the same as the source function. For example, if LambdaA has a Destination configuration attached for Success, LambdaA is not a valid destination ARN. This prevents recursive functions.
  • For EventBridge, the JSON is passed as the Detail in the PutEvents call. The source is lambda, and detail type is either Lambda Function Invocation Result - Success or Lambda Function Invocation Result – Failure. The resource fields contain the function and destination ARNs.

AWS CloudFormation configuration

Destinations CloudFormation configuration is created via the following YAML.

Resources: 
  EventInvokeConfig:
    Type: AWS::Lambda::EventInvokeConfig
    Properties:
        FunctionName: “YourLambdaFunctionWithEventInvokeConfig”
        Qualifier: "$LATEST"
        MaximumEventAgeInSeconds: 600
        MaximumRetryAttempts: 0
        DestinationConfig:
            OnSuccess:
                Destination: “arn:aws:sns:us-east-1:123456789012:YourSNSTopicOnSuccess”
            OnFailure:
                Destination: “arn:aws:lambda:us-east-1:123456789012:function:YourLambdaFunctionOnFailure”

Conclusion

AWS Lambda Destinations gives you more visibility and control of function execution results. This helps you build better event-driven applications, reducing code, and using Lambda’s native failure handling controls.

There are no additional costs for enabling Lambda Destinations. However, calls made to destination target services may be charged.

To learn more, see Lambda Destinations in the AWS Lambda Developer Guide.

Application integration patterns for microservices: Fan-out strategies

Post Syndicated from Rachel Richardson original https://aws.amazon.com/blogs/compute/application-integration-patterns-for-microservices-fan-out-strategies/

This post is courtesy of Dirk Fröhner

The first blog in this series introduced asynchronous messaging for building loosely coupled systems that can scale, operate, and evolve individually. It considered messaging as a communications model for microservices architectures. This post covers concrete architectural considerations, focusing on the messaging architecture.

Wild Rydes

Wild Rydes is a fictional technology start-up. You may have heard of it – it disrupts individual transportation by replacing traditional taxis with unicorns. We use the Wild Rydes storyline in several hands-on AWS workshops. It illustrates concepts such as serverless development, event-driven design, API management, and messaging in microservices.

This blog post explores the decision-making process in building the Wild Rydes workshop, with a goal of helping you apply these concepts to your applications.

In the workshop, a customer requests a ‘unicorn’ ride using the Wild Rydes customer application. Registered unicorn drivers can use the application to manage their rides. Unicorn drivers submit a ride completion message after they have successfully delivered a customer to their destination.

Wild Rydes app image

Submit a ride completion

API exposed by the unicorn management service

At Wild Rydes, end-user clients are implemented as mobile applications and communicate via REST APIs (also known as hypermedia APIs) with the backend services.

For this use case, the application interacts with the API exposed by the unicorn management service. It uses the submit-ride-completion resource that it discovered from the API’s home document to send the relevant details of a ride to the backend. In response, the backend persists these details, creates a new completed-ride resource. This returns the respective status code, the location, and a representation of the new resource to the client. The API details are shown below.

Request from client to submit the details of a completed ride:

POST /<submit-ride-completion-resource-path> HTTP/1.1
Content-Type: application/json;charset=UTF-8
...

{
    "from": "...",
    "to": "...",
    "duration": "...",
    "distance": "...",
    "customer": "...",
    "fare": "..."
}

Response from the unicorn management service:

HTTP/1.1 201 Created
Date: Sat, 31 Aug 2019 12:00:00 GMT
Location: <url-of-newly-created-completed-ride-resource>
Content-Location: <url-of-newly-created-completed-ride-resource>
Content-Type: application/json;charset=UTF-8
...

{
    "links": {
        "self": {
            "href": "https://..."
        }
    },
    <completed-ride-resource-representation-properties>
}

Schematic architecture for the use case

The schematic architecture for the use case is shown in diagram 1 below:

Diagram 1: Mutliple microservices need information about ride completion

Diagram 1: Multiple microservices need information about ride completion

There are other microservices in Wild Rydes that are also interested in a new completed ride. The examples from the diagram are:

  • Customer notification service: customers should receive a notification in the app about their latest completed ride.
  • Customer accounting service: After all, Wild Rydes is a business, so this service is responsible for collecting the fare from the customer.
  • Customer loyalty service: Everybody wants to collect miles and would like to receive benefits for being a loyal customer.
  • Data lake ingestion service: Wild Rydes is a data-driven company and they want to ingest all data generated from any process into their data lake for arbitrary analytics.
  • Extraordinary rides service: This special service is interested in rides with fares or distances above certain thresholds for preparing insights for business managers.

Based on this scenario, let’s review the integration options.

Integration options

Integration via database

The unicorn management service stores the details of a completed ride in a database. It could share the database with the other services directly, but that creates tight coupling. Sharing the database also restricts your flexibility to scale and evolve your services.

Integration via REST APIs

What about using REST APIs for the integration? The HTTP-based implementation of the REST architectural style uses the distributed architecture concepts of the web. However, what does this mean for the implementation?

Diagram 2: Using REST APIs to communicate to microservices

Diagram 2: Using REST APIs to communicate to microservices

As shown in diagram 2 above:

  • Effectively, all interested services on the right-hand side would have to expose an API resource. These would be called by the unicorn management service for each newly completed ride.
  • To enable elasticity behind a single resource URL, you may need a load balancer in front of each interested service.
  • The unicorn management service would have to know about all these interested services and their respective APIs. Hopefully, each service uses a streamlined API resource.
  • Lastly, the unicorn management service must store, retry, and track all request attempts in case an interested service is not available. This ensures durability so we don’t lose any of these notifications.

One approach is to manage a recipient list in the unicorn management service. This adds additional complexity to the unicorn management service and coupling on both sides. Although there are self-registration and discovery approaches, managing a recipient list is not the core use case of the unicorn management service.

Diagram 3: Using a separate service to manage the fan-out to other services

Diagram 3: Using a separate service to manage the fan-out to other services

A better approach would be to externalize the recipient list into a separate Request Distribution Service, as diagram 3 shows. This decouples both sides, but binds each side to the new service. Still, the unicorn management service is still responsible for the delivery of the ride data to all the recipients. Again, this heavy lifting is not the main task of this service.

Diagram 4: Filtering information for extraordinary rides

Diagram 4: Filtering information for extraordinary rides

In diagram 4, the information filtering for the Extraordinary Rides Service is self-managed. This means that there is code on one side to either not send or to discard irrelevant ride data.

For this use case, integration via REST APIs potentially adds coupling to the services. And it adds heavy lifting to the services that is beyond their actual domain.

Integration via messaging

A third option could use messaging for the integration.

Publish-subscribe pattern

Both Amazon SNS and Amazon EventBridge can be used to implement the publish-subscribe pattern.  In this use case, we recommend Amazon SNS, which scales to support high throughput and fan-out applications. Amazon EventBridge includes direct integrations with software as a service (SaaS) applications and other AWS services. It’s ideal for publish-subscribe use cases involving these types of integrations.

Diagram 5: Using Amazon SNS to implement a publish-subscribe pattern

Diagram 5: Using Amazon SNS to implement a publish-subscribe pattern

Diagram 5 shows an SNS topic called Ride Completion Topic. The unicorn management service can now send the details about a completed ride into that topic. All interested services on the right-hand side can subscribe to this topic.

Using a message topic to publish the details of a completed ride frees us from managing the recipient list, as well as making ensuring reliable delivery of the messages. It also decouples both sides as much as possible. Services on the right-hand side can autonomously subscribe to the topic. The Unicorn Management Service does not know anything about the topic’s subscribers.

Message filter pattern

Looking at the Extraordinary Rides Service, the message filter functionality of Amazon SNS can autonomously and individually discard irrelevant messages. The Extraordinary Rides Service can specify the threshold values for the fare and distance.

Diagram 6: Filtering extraordinary rides using Amazon SNS

Diagram 6: Filtering extraordinary rides using Amazon SNS

Topic-queue-chaining pattern

Consider the publish-subscribe channel between the Unicorn Management Service, and the subscribing services on the right-hand side.

One of the consuming services may go offline for maintenance. Or the code that processes messages from the ride completion topic could run into an exception. These are two examples where a subscriber service could potentially miss topic messages.

A good pattern to apply here is topic-queue-chaining. That means that you add a queue, in our case an SQS queue, between the ride completion topic and each of the subscriber services. As messages are buffered persistently in an SQS queue, it prevents lost messages if a subscriber process run into problems for many hours or days.

Diagram 7: Chaining topics and queues to buffer messages persistently

Diagram 7: Chaining topics and queues to buffer messages persistently

Queues as buffering load balancers

An SQS queue in front of each subscriber service also acts as a buffering load balancer.

Since every message is delivered to one of potentially many consumer processes, you can scale out the subscriber services, and the message load is distributed over the available consumer processes.

As messages are buffered in the queue, they are preserved during a scaling event, such as when you must wait until an additional consumer process becomes operational.

Lastly, these queue characteristics help flatten peak loads for your consumer processes, buffering messages until consumers are available. This allows you to process messages at a pace decoupled from the message source.

Conclusion

The Wild Rydes example shows how messaging can provide decoupling and greater flexibility for your microservices landscape.

In contrast to REST APIs, a messaging system takes care of message delivery outside of your service code. Using a publish-subscribe channel provides simple fan-out capability. And message filters allow for selective message reception without the effort of implementing that logic into your code.

With topic-queue-chaining pattern, you can add queue characteristics to a fan-out scenario so that you can easily scale out on the consumer side, and flatten peak loads.

For a deeper dive into queues and topics and how to use them in your microservices architecture, please use the following resources:

  1. AWS whitepaper: Implementing Microservices on AWS
  2. AWS blog: Implementing enterprise integration patterns with AWS messaging services: point-to-point channels
  3. AWS blog: Implementing enterprise integration patterns with AWS messaging services: publish-subscribe channels
  4. AWS blog: Building Scalable Applications and Microservices: Adding Messaging to Your Toolbox

Understanding asynchronous messaging for microservices

Post Syndicated from Rachel Richardson original https://aws.amazon.com/blogs/compute/understanding-asynchronous-messaging-for-microservices/

This post is courtesy of Dirk Fröhner

One of the implications of applying the microservices architectural style is that much communication between components happens over the network. After all, your microservices landscape is a distributed system. To achieve the promises of microservices, such as being able to individually scale, operate, and evolve each service, this communication must happen in a loosely coupled and reliable manner.

A common way to loosely couple services is to expose an API following the REST architectural style. REST APIs are based on the architecture of the web and provide loose coupling between communicating parties. REST APIs offer a great way to decouple interfaces from concrete implementations, and to advise clients about what they can do next, by the use of links and link relations.

While REST APIs are common and useful in microservices design, REST APIs tend to be designed with synchronous communications, where a response is required. A request coming from an end-user client can trigger a complex communications path within your services landscape, which can effectively add coupling between the services at runtime. After all, this is why there are mitigation patterns like circuit-breaker in the first place. REST APIs can also add some heavy lifting to your infrastructure that we will discuss further below.

Asynchronous messaging

If loose-coupling is important, especially in a system that requires high resilience and has unpredictable scale, another option is asynchronous messaging.

Asynchronous messaging is a fundamental approach for integrating independent systems, or building up a set of loosely coupled systems that can operate, scale, and evolve independently and flexibly. As our colleague Tim Bray said, “If your application is cloud-native, or large-scale, or distributed, and doesn’t include a messaging component, that’s probably a bug.” In this blog post, we will outline some fundamental benefits of asynchronous messaging for the communications between microservices.

For a refresher on the fundamental messaging patterns and their implementations with Amazon SQS, Amazon SNS, and Amazon MQ, please read our previous blog posts

For a summary of the semantics of queues and topics:

  • A queue is like a buffer. You can put messages into a queue, and you can retrieve messages from a queue. Message queues operate so that any given message is only consumed by one receiver, although multiple receivers can be connected to the queue.
  • A topic is like a broadcasting station. You can publish messages to a topic, and anyone interested in these messages can subscribe to the topic. In this model, any message published to a topic is immediately received by all of the subscribers of the topic (unless you have applied the message filter pattern).

Use-case

Consider a typical scenario illustrated in the diagram below. An end-user client (EUC) addresses an API resource of one of our services, through Amazon API Gateway in this example. From there, the request can potentially follow a path across the microservices landscape to get completely processed.

To provide the final result, there will be potentially cascading subsequent requests sent between other microservices. This example illustrates the complexity involved in processing a single end user request.

End User Client accessing a service using an API

Diagram 1: End-User Client accessing a service using an API

End-user clients (EUCs) often communicate with services via REST APIs in a synchronous manner. However, the communication can also be designed using an asynchronous approach. For instance, if an EUC submits a request that takes some time to process, the respective API resource can respond with HTTP status 202 Accepted, and a link to a resource that provides the current processing status. Downstream, the communication between the service that receives that request, and other services that are involved in processing the request, can happen asynchronously using messaging services.

There are situations where a communications model using asynchronous messaging can make your life easier than using REST APIs.

Infrastructure complexity

Start with looking at the infrastructure complexity for the backends of your services. Depending on your implementation paradigm, you have to include different components in your infrastructure that you don’t have to deal with when using messaging.

Imagine your services each expose a REST API. Typically, this means you add a load balancer in front of your compute layer, and your backend implementation includes an HTTP server. It is usually a good idea to decouple your services APIs from their concrete implementations, so you could also consider adding Amazon API Gateway in front of your load balancer.

For a serverless approach, you don’t need to worry about load balancing and scaling out infrastructure. Amazon API Gateway with AWS Lambda integration provides a fully managed solution for removing complexities around infrastructure management.

Using Amazon SQS as a cloud-native messaging service for queues, you don’t employ any of the above mentioned components. As described in a prior post, an SQS queue can act as a load balancer in itself. The consumers, or target services, don’t need an HTTP server, but simply ask a queue for available messages. If you use AWS Lambda for your consumers, this process is even simpler, as the Lambda functions are automatically invoked when messages appear in an SQS queue. See Using AWS Lambda with Amazon SQS to learn more.

The same applies to Serverless architectures implementing a publish/subscribe pattern. Lambda function executions can be directly triggered by SNS messages. Without AWS Lambda, you need load balancers and web servers in your backend implementations to receive SNS notifications, as those are injected via web hooks into your services. SNS also provides the fan-out functionality that you would otherwise have to build using an intermediary component to implement a recipient list of subscribers.

Reliability, resilience

For synchronous systems, if a service crashes while it processes the payload of an API request, the information is lost. A good way to prevent this on a microservice is to explicitly persist an incoming request immediately after receiving it. Then process and reprocess, until the request is finally marked as resolved.

This approach requires additional work, and it requires the microservice to not crash while persisting an incoming API request. The microservice sending a request must also resend if the target service doesn’t acknowledge receipt. For example, it doesn’t respond with a successful HTTP status code, or the connection drops.

When sending messages to a queue, this additional work is addressed by the messaging infrastructure. A message will remain in a queue unless a consumer explicitly states that processing is finished by acknowledging the message reception. As long as message reception is not acknowledged by a consumer, it will stay in the queue. Messages can be retained in an SQS queue for a maximum of 14 days.

Scale out latency

Under increased load, your services must scale out to process the requests. You must then consider scale-out latency, which may be managed for you with serverless implementations. It takes a few moments from when an Auto Scaling group triggers the launch of additional instances until these are ready to operate. Also launching new container tasks takes time. When your scaling threshold is not optimal and the scaling event occurs late, your available resources may be unable to serve all incoming requests. These requests may be lost or answered with HTTP status code 5xx.

Using message queues that buffer messages during a scaling event help prevent this. Even in use cases where the EUC is waiting for an immediate response, this is the more reliable architecture. If your infrastructure needs time to scale out and you are not able to process all requests in time, the requests are persisted.

When messaging is your only choice

What happens when your services must respond to peak loads at scale?

For many applications, the scale-out latency, including load balancer pre-warming, will eventually become too large to handle steeply ascending loads fast enough. With a serverless architecture, exposing your Lambda functions with API Gateway can handle steeply ascending loads. But you must still consider downstream systems, which may be easily overwhelmed.

In these scenarios, where rapid scaling without overwhelming downstream systems is important, messaging may be your best choice. Message queues help protect your downstream services by buffering incoming payloads for consumption at the pace of the consuming service. This helps not only for the communications between microservices, but also when peak loads flood your client-facing API. Often, the most important goal is to accept an incoming request, while the actual processing of that request can happen later. You decouple these steps from each other by using queues.

Serverless messaging systems like Amazon SQS and Amazon SNS can respond quickly to support high scale. These are often the best solution when scale is unpredictable.  While the instance-based messaging system, Amazon MQ, provides compatibility with open standards, it requires manual scaling for large workloads, unlike serverless messaging services.

Conclusion

We hope you got some inspiration to also employ asynchronous messaging for your microservices communications architecture. In blog XYZ we provide concrete examples of these patterns. For more information, feel free to consume the following resources:

  1. AWS whitepaper: Implementing Microservices on AWS
  2. AWS blog: Implementing enterprise integration patterns with AWS messaging services: point-to-point channels
  3. AWS blog: Implementing enterprise integration patterns with AWS messaging services: publish-subscribe channels
  4. AWS blog: Building Scalable Applications and Microservices: Adding Messaging to Your Toolbox

Read the next blog in the series,  Application Integration Patterns for Microservices: Fan-out Strategies.

New for AWS Lambda – SQS FIFO as an event source

Post Syndicated from James Beswick original https://aws.amazon.com/blogs/compute/new-for-aws-lambda-sqs-fifo-as-an-event-source/

AWS Lambda first announced support for Amazon SQS standard queues as an event source in April 2018. This allows builders to develop serverless applications using queues to directly invoke Lambda functions. Today, we have expanded this feature to include SQS FIFO queues. This makes it easier to create serverless applications using queues where the order of messages is important.

Ordered events and operations are critical for some types of event-driven application. For example, in stock market trading applications, the order of events for a trade is essential to determine the time, price, and parties in a transaction. Similarly, in applications managing pricing or inventory, the order of events is key for maintaining the data integrity of the system.

Using SQS FIFO with Lambda is straight-forward, with only minor differences in configuration compared with standard SQS queues. This blog post explains how serverless developers can get started using SQS FIFO queues in their Lambda functions.

How it works

SQS FIFO queues are similar to standard queues but use two additional attributes in the SendMessage API. These attributes are also returned in the event payload to the consuming Lambda function:

  • MessageGroupID: this required field enables multiple message groups within a single queue. If you do not need this functionality, provide the same MessageGroupId value for all messages. Messages within a single group are processed in a FIFO fashion.
  • MessageDeduplicationID: this token is used to deduplicate messages within a 5-minute interval. If two messages with the same MessageDuplicationID are sent, both messages are accepted successfully but only one is delivered. Learn more about the best practices for using this attribute.

You can submit a message into an SQS FIFO queue using the AWS Management Console, AWS CLI, AWS SAM, or AWS SDK. This SDK example uses Node.js to submit a single message:

const AWS = require('aws-sdk')
AWS.config.update({region: 'us-west-2'})
const sqs = new AWS.SQS({apiVersion: '2012-11-05'})

const send = async (groupId, messageId) => {
  return await sqs.sendMessage({
    MessageGroupId: `group-${groupId}`,
    MessageDeduplicationId: `m-${groupId}-${messageId}`,
    MessageBody: `${messageId}`,
    QueueUrl: "https://sqs.us-west-2.amazonaws.com/123456789012/myTestQueue.fifo"
  }).promise()
}

const main = async () => {
  await send ('A', '1')
}
main()

When the Lambda function is triggered by the SQS FIFO queue, these two attributes are delivered as part of the event payload:

SQS FIFO payload

When configured this way, the SQS FIFO queue sends as many messages as possible to the consuming Lambda function, subject to the Batch size configured in the trigger. The Batch size determines the number of items to read from the queue, up to a maximum of 10 items, though a single batch is smaller if the queue has fewer items.

For example, if you submit six messages to a queue with a Batch size of 10 using two separate MessageGroupId values, the consuming Lambda function would receive all six items:

 

SQS FIFO example #1

As all the messages are processed in a single batch, only one Lambda function is invoked. A more complex example shows how the Lambda service uses concurrency to process a busy SQS FIFO queue more efficiently.

SQS FIFO example #2

In this example, there are three message groups, and a Lambda function with an SQS FIFO trigger. The Batch size for the Lambda trigger is 5, and there are 53 messages in total:

  • Message Group A consists of 18 messages, A1-A18
  • Message Group B consists of 12 messages, B1-B12
  • Message Group C consists of 23 messages, C1-C23

There are three rules governing the order of messages leaving a FIFO queue, to help understand the processing behavior:

  1. Return the oldest message where no other message with the same MessageGroupId is in flight.
  2. Return as many messages with the same MessageGroupId as possible.
  3. If a message batch is still not full, go back to the first rule. As a result, it’s possible for a single batch to contain messages from multiple MessageGroupIds.

In this case, the SQS FIFO queue receives these messages, and orders the items within each message group. In processing, the following events occur:

  1. Consumer #1 receives 5 messages in batch #1 (C1-C5).
  2. Consumer #1 receives 5 messages in batch #2 (B1-B5).
  3. As there are messages in flight for group B, and there are items in the queue from other groups ready for processing, Lambda scales up. Consumer #2 now receives 5 messages in batch #2 (C6-C10).
  4. Consumer #3 receives batch #3 (A1-A5).
  5. Each Lambda instance will continue receiving batches of 5 messages until the queue is empty.

Lambda automatically scales out horizontally to consume the messages in the queue. It will try to consume the queue as quickly and efficiently as possible by maximizing concurrency within the bounds of each service. As queue traffic fluctuates, the Lambda service scales the polling operations based on the number of inflight messages.

In SQS FIFO queues, using more than one MessageGroupId enables Lambda to scale up and process more items in the queue using a greater concurrency limit. Total concurrency is equal to or less than the number of unique MessageGroupIds in the SQS FIFO queue. Learn more about AWS Lambda Function Scaling and how it applies to your event source.

Amazon SQS FIFO queues ensure that the order of processing follows the message order within a message group. However, it does not guarantee only once delivery when used as a Lambda trigger. If only once delivery is important in your serverless application, it’s recommended to make your function idempotent. You could achieve this by tracking a unique attribute of the message using a scalable, low-latency control database like Amazon DynamoDB.

Triggering Lambda from SQS

1. Navigate to the SQS console and choose Create New Queue.

2. Names for FIFO queues must end in .fifo. For Queue Name, enter myTestQueue.fifo and select FIFO Queue. Choose Quick-Create Queue.

Create New Queue

3. From the Lambda console, choose Create function.

4. Enter mySQStest for Function name, and leave the Node.js 12.x runtime selected. Open the Choose or create an execution role panel and select Create a new role from AWS policy templates.

5. Enter mySQStest for Role name and select Amazon SQS poller permissions from the Policy templates drop-down. Choose Create function.

Create Lambda function

6. Once the function is created, you can configure the trigger. Choose Add trigger.

Add trigger

7. In the Select a trigger drop-down, choose SQS. In the SQS queue drop-down, select the FIFO queue created earlier. Leave Batch size as the default value, then choose Add.

Add trigger dialog

8. After a few seconds, the SQS trigger is ready.

SQS FIFO trigger ready

The Lambda function is now invoked when new messages are available in the SQS FIFO queue.

Conclusion

Now, you can process messages from Amazon SQS FIFO queues with Lambda, helping bring scalable serverless processing to applications needing first-in, first-out message processing.

You can get started with SQS FIFO queue as a Lambda event source via AWS Management Console, AWS CLI, AWS SAM, AWS CloudFormation, or AWS SDK for Lambda. It is available in all AWS Regions where AWS Lambda is available. You pay only for the SQS API operations performed by the Lambda service on your behalf, and the Lambda requests and duration used to process your messages.

For more information on where AWS Lambda is available, see the AWS Region Table. To learn more, see Using AWS Lambda with Amazon SQS FIFO in the AWS Lambda Developer Guide.

 

 

Designing durable serverless apps with DLQs for Amazon SNS, Amazon SQS, AWS Lambda

Post Syndicated from Rachel Richardson original https://aws.amazon.com/blogs/compute/designing-durable-serverless-apps-with-dlqs-for-amazon-sns-amazon-sqs-aws-lambda/

This post is courtesy of Otavio Ferreira, Sr Manager, SNS.

In a postal system, a dead-letter office is a facility for processing undeliverable mail. In pub/sub messaging, a dead-letter queue (DLQ) is a queue to which messages published to a topic can be sent, in case those messages cannot be delivered to a subscribed endpoint.

Amazon SNS supports DLQs, making your applications more resilient and durable upon delivery failure modes.

Understanding message delivery failures and retries

The delivery of a message fails when it’s not possible for Amazon SNS to access the subscribed endpoint. There are two reasons why this might happen:

  • Client errors, where the client is SNS (the message sender).
  • Server errors, where the server is the system that hosts the subscription endpoint (the message receiver), such as Amazon SQS or AWS Lambda.

Client errors

Client errors happen when SNS has stale subscription metadata. One common cause of client errors is when you (the endpoint owner) delete the endpoint. For example, you might delete the SQS queue that is subscribed to your SNS topic, without also deleting the SNS subscription corresponding to the queue. Another common cause is when you change the resource policy attached to your endpoint in a way that prevents SNS from delivering messages to that endpoint.

These errors are considered client errors because the client has attempted the delivery of a message to a destination that, from the client’s perspective, is no longer accessible. SNS does not retry the delivery of messages that failed as the result of client errors.

Server errors

Server errors happen when the system that powers the subscribed endpoint is unavailable, or when it returns an exception response indicating that it failed to process a valid request from SNS.

When server errors occur, SNS retries the failed deliveries according to a backoff function, which can be either linear or exponential. When a server error occurs for an AWS managed endpoint, backed by either SQS or Lambda, then SNS retries the delivery for up to 100,015 times, over 23 days.

Server errors can also happen with customer managed endpoints, namely HTTP, SMS, email, and mobile push endpoints. SNS also retries the delivery for these types of endpoints. HTTP endpoints support customer-defined retry policies, while SNS sets an internal delivery retry policy for SMS, email, and mobile push endpoints to 50 times, over 6 hours.

Delivery retries

SNS may receive a client error, or continue to receive a server error for a message beyond the number of retries defined by the corresponding retry policy. In that event, SNS discards the message. Setting a DLQ to your SNS subscription enables you to keep this message, regardless of the type of error, either client or server. DLQs give you more control over messages that cannot be delivered.

For more information on the delivery retry policy for each delivery protocol supported by SNS, see Amazon SNS Message Delivery Retry.

Using DLQs for AWS services

SNS, SQS, and Lambda support DLQs, addressing different failure modes. All DLQs are regular queues powered by SQS.

In SNS, DLQs store the messages that failed to be delivered to subscribed endpoints. For more information, see Amazon SNS Dead-Letter Queues.

In SQS, DLQs store the messages that failed to be processed by your consumer application. This failure mode can happen when producers and consumers fail to interpret aspects of the protocol that they use to communicate. In that case, the consumer receives the message from the queue, but fails to process it, as the message doesn’t have the structure or content that the consumer expects. The consumer can’t delete the message from the queue either. After exhausting the receive count in the redrive policy, SQS can sideline the message to the DLQ. For more information, see Amazon SQS Dead-Letter Queues.

In Lambda, DLQs store the messages that resulted in failed asynchronous executions of your Lambda function. An execution can result in an error for several reasons. Your code might raise an exception, time out, or run out of memory. The runtime executing your code might encounter an error and stop. Your function might hit its concurrency limit and be throttled. Regardless of the error type, when the error occurs, your code might have run completely, partially, or not at all. By default, Lambda retries an asynchronous execution twice. After exhausting the retries, Lambda can sideline the message to the DQL. For more information, see AWS Lambda Dead-Letter Queues.

When you have a fan-out architecture, with SQS queues and Lambda functions subscribed to an SNS topic, we recommend that you set DLQs to your SNS subscriptions, and to your destination queues and functions as well. This approach gives your application resilience against message delivery failures, message processing failures, and function execution failures too.

Applying DLQs in a use case

Here’s how everything comes together. The following diagram shows a serverless backend architecture that supports a car rental application. This is a durable serverless architecture based on DLQs for SNS, SQS, and Lambda.

Dead Letter Queue - DLQ SNS use case with architecture diagram

When a customer places an order to rent a car, the application sends that request to an API, which is powered by Amazon API Gateway. The REST API is backed by an SNS topic named Rental-Orders, and deployed onto an Amazon VPC subnet. The topic then fans out that order to the following two subscribed endpoints, for parallel processing:

  • An SQS queue, named Rental-Fulfilment, which feeds the integration with an internal fulfilment system hosted on Amazon EC2.
  • A Lambda function, named Rental-Billing, which processes and loads the customer order into a third-party billing system, also hosted on Amazon EC2.

To increase the durability of this serverless backend API, the following DLQs have been set up:

  • Two SNS DLQs, namely Rental-Fulfilment-Fanout-DLQ and Rental-Billing-Fanout-DLQ, which store the order in case either the subscribed SQS queue or Lambda function ever becomes unreachable.
  • An SQS DLQ, named Rental-Fulfilment-DLQ, which stores the order when the fulfilment system fails to process the order.
  • A Lambda DLQ, named Rental-Billing-DLQ, which stores the order when the function fails to process and load the order into the billing system.

When the DLQ captures the message, you can inspect the message for troubleshooting purposes. After you address the error at hand, you can poll the DLQ to retry the processing of the message.

Setting up DLQs for subscriptions, queues, and functions can be done using the AWS Management Console, SDK, CLI, API, or AWS CloudFormation. You can use the SDK, CLI, and API for polling the DLQs as well.

Configuring DLQs for subscriptions

You can attach a DLQ to an SNS subscription by setting the subscription’s RedrivePolicy parameter. The policy is a JSON object that refers to the DLQ ARN. The ARN must point to an SQS queue in the same AWS account as that of the SNS subscription. Also, both the DLQ and the subscription must be in the same AWS Region.

Here’s how you can configure one of the SNS DLQs applied in the car rental application example, presented earlier.

The following JSON object is a CloudFormation template that subscribes the SQS queue Rental-Fulfilment to the SNS topic Rental-Orders. The template also sets a RedrivePolicy that targets Rental-Fulfilment-Fanout-DLQ as a DLQ.

Lastly, the template sets a FilterPolicy value. It makes SNS deliver a message to the subscribed queue only if the published message carries an attribute named order-status with value set to either confirmed or canceled. As Amazon SNS Message Filtering happens before message delivery, messages that are filtered out aren’t sent to that subscription’s DLQ.

Internally, the CloudFormation template uses the SNS Subscribe API action for deploying the subscription and setting both policies, all part of the same API request.

{  
   "Resources": {
      "mySubscription": {
         "Type" : "AWS::SNS::Subscription",
         "Properties" : {
            "Protocol": "sqs",
            "Endpoint": "arn:aws:sqs:us-east-1:123456789012:Rental-Fulfilment",
            "TopicArn": "arn:aws:sns:us-east-1:123456789012:Rental-Orders",
            "RedrivePolicy": {
               "deadLetterTargetArn": 
                  "arn:aws:sqs:us-east-1:123456789012:Rental-Fulfilment-Fanout-DLQ"
            },
            "FilterPolicy": { 
               "order-status": [ "confirmed", "canceled" ]
            }
         }
      }
   }
}

Maybe the SNS topic and subscription are already deployed. In that case, you can use the SNS SetSubscriptionAttributes API action to set the RedrivePolicy, as shown by the following code examples, based on the AWS CLI and the AWS SDK for Java.

$ aws sns set-subscription-attributes 
   --region us-east-1
   --subscription-arn arn:aws:sns:us-east-1:123456789012:Rental-Orders:44019880-ffa0-4067-9cb4-b974443bcck2
   --attribute-name RedrivePolicy 
   --attribute-value '{"deadLetterTargetArn":"arn:aws:sqs:us-east-1:123456789012:Rental-Fulfilment-Fanout-DLQ"}'
AmazonSNS sns = AmazonSNSClientBuilder.defaultClient();

String subscriptionArn = "arn:aws:sns:us-east-1:123456789012:Rental-Orders:44019880-ffa0-4067-9cb4-b974443bcck2";

String redrivePolicy = "{\"deadLetterTargetArn\":\"arn:aws:sqs:us-east-1:123456789012:Rental-Fulfilment-Fanout-DLQ\"}";

SetSubscriptionAttributesRequest request = new SetSubscriptionAttributesRequest(
  subscriptionArn, 
  "RedrivePolicy", 
  redrivePolicy
);

sns.setSubscriptionAttributes(request);

Monitoring DLQs

You can use Amazon CloudWatch metrics and alarms to monitor the DLQs associated with your SNS subscriptions. In the car rental example, you can monitor the DLQs to be notified when the API failed to distribute any car rental order to the fulfillment or billing systems.

As regular SQS queues, the DLQs in SNS emit a number of metrics to CloudWatch, in 5-minute data points, such as NumberOfMessagesSent, NumberOfMessagesReceived and NumberOfMessagesDeleted. You can use these SQS metrics to be notified upon activity in your DLQs in SNS, so you may trigger a message recovery protocol.

You might have a case where you expect the DLQ to be always empty. In that case, create an CloudWatch alarm on NumberOfMessagesSent, set the alarm threshold to zero, and provide a separate SNS topic to be notified when the alarm goes off. The SNS topic, in its turn, can delivery your alarm notification to any endpoint type that you choose, such as email address, phone number, or mobile pager app.

Additionally, SNS itself provides its own set of metrics that are relevant to DLQs. Specifically, SNS metrics include the following:

  • NumberOfNotificationsRedrivenToDlq – Used when sending the message to the DLQ succeeds.
  • NumberOfNotificationsFailedToRedriveToDlq – Used when sending the message to the DLQ fails. This can happen because the DLQ either doesn’t exist anymore or doesn’t have the required access permissions to allow SNS to send messages to it. For more information about setting up the required access policy, see Giving Permissions for Amazon SNS to Send Messages to Amazon SQS.

Debugging with DLQs

Use CloudWatch Logs to see the exceptions that caused your SNS deliveries to fail and your messages to be sidelined to DLQs. In the car rental example, you can inspect the rental orders in the DLQs, as well as the logs associated with these queues. Then you can understand why those orders failed to be fanned out to the fulfilment or billing systems.

SNS can log both successful and failed deliveries in CloudWatch. You can enable Amazon SNS Delivery Status Logging by setting three SNS topic attributes, which are delivery protocol-specific. As an example, for SNS deliveries to SQS queues, you must set the following topic attributes: SQSSuccessFeedbackRoleArn,  SQSFailureFeedbackRoleArn, and SQSSuccessFeedbackSampleRate.

The following JSON object represents a successful SNS delivery in an CloudWatch Logs entry. The status code logged is 200 (SUCCESS). The attribute RedrivePolicy shows that the SNS subscription in question had its DLQ set.

{
  "notification": {
    "messageMD5Sum": "7bb3327ac55e49485bad42e159ca4d4b",
    "messageId": "e8c2bb09-235c-5f5d-b583-efd8df0f7d74",
    "topicArn": "arn:aws:sns:us-east-1:123456789012:Rental-Orders",
    "timestamp": "2019-10-04 05:13:55.876"
  },
  "delivery": {
    "deliveryId": "6adf232e-fb12-5062-a564-27ff3741051f",
    "redrivePolicy": "{\"deadLetterTargetArn\": \"arn:aws:sqs:us-east-1:123456789012:Rental-Fulfilment-Fanout-DLQ\"}",
    "destination": "arn:aws:sqs:us-east-1:123456789012:Rental-Fulfilment",
    "providerResponse": "{\"sqsRequestId\":\"b2608a46-ccc4-51cc-003d-de972097debc\",\"sqsMessageId\":\"05fecd22-60a1-4d7d-bb79-026d49700b5a\"}",
    "dwellTimeMs": 58,
    "attempts": 1,
    "statusCode": 200
  },
  "status": "SUCCESS"
}

The following JSON object represents a failed SNS delivery in CloudWatch Logs. In the following code example, the subscribed queue doesn’t exist. As a client error, the status code logged is 400 (FAILURE). Again, the RedrivePolicy attribute refers to a DLQ.

{
  "notification": {
    "messageMD5Sum": "81c395cbd350da6bedfe3b24db9517b0",
    "messageId": "9959db9d-25c8-57a6-9439-8e5be8f71a1f",
    "topicArn": "arn:aws:sns:us-east-1:123456789012:Rental-Orders",
    "timestamp": "2019-10-04 05:16:51.116"
  },
  "delivery": {
    "deliveryId": "be743821-4c2c-5acc-a586-6cf0807f6fb1",
    "redrivePolicy": "{\"deadLetterTargetArn\": \"arn:aws:sqs:us-east-1:123456789012:Rental-Fulfilment-Fanout-DLQ\"}",
    "destination": "arn:aws:sqs:us-east-1:123456789012:Rental-Fulfilment",
    "providerResponse": "{\"ErrorCode\":\"AWS.SimpleQueueService.NonExistentQueue\", \"ErrorMessage\":\"The specified queue does not exist or you do not have access to it.\",\"sqsRequestId\":\"Unrecoverable\"}",
    "dwellTimeMs": 53,
    "attempts": 1,
    "statusCode": 400
  },
  "status": "FAILURE"
}

When the message delivery fails and there is a DLQ attached to the subscription, the message is sent to the DLQ and an additional entry is logged in CloudWatch. This new entry is specific to the delivery to the DLQ and refers to the DLQ ARN as the destination, as shown in the following JSON object.

{
  "notification": {
    "messageMD5Sum": "81c395cbd350da6bedfe3b24db9517b0",
    "messageId": "8959db9d-25c8-57a6-9439-8e5be8f71a1f",
    "topicArn": "arn:aws:sns:us-east-1:123456789012:Rental-Orders",
    "timestamp": "2019-10-04 05:16:52.876"
  },
  "delivery": {
    "deliveryId": "a877c79f-a3ee-5105-9bbd-92596eae0232",
    "destination":"arn:aws:sqs:us-east-1:123456789012:Rental-Fulfilment-Fanout-DLQ",
    "providerResponse": "{\"sqsRequestId\":\"8cef1af5-e86a-519e-ad36-4f33252aa5ec\",\"sqsMessageId\":\"2b742c5c-0750-4ec5-a717-b95897adda8e\"}",
    "dwellTimeMs": 51,
    "attempts": 1,
    "statusCode": 200
  },
  "status": "SUCCESS"
}

By analyzing Amazon CloudWatch Logs entries, you can understand why an SNS message was moved to a DLQ, and then take the required set of steps to recover the message. When you enable delivery status logging in SNS, you can configure the sample rate in which deliveries are logged, from 0% to 100%.

Encrypting DLQs

When your SNS subscription targets an SQS encrypted queue, then you probably want your DLQ to be an SQS encrypted queue as well. This configuration provides consistency in the form that your messages are encrypted at rest.

To follow this security recommendation, give the CMK you used to encrypt your DLQ a key policy that grants the SNS service principal access to AWS KMS API actions. For example, see the following sample key policy:

{
    "Sid": "GrantSnsAccessToKms",
    "Effect": "Allow",
    "Principal": { "Service": "sns.amazonaws.com" },
    "Action": [ "kms:Decrypt", "kms:GenerateDataKey*" ],
    "Resource": "*"
}

If you have an SNS encrypted topic, but a subscription in this topic points to a DLQ that isn’t an SQS encrypted queue, then messages sidelined to the DLQ aren’t encrypted at rest.

For more information, see Enabling Server-Side Encryption (SSE) for an Amazon SNS Topic with an Amazon SQS Encrypted Queue Subscribed.

Summary

DLQs for SNS, SQS, and Lambda increase the resiliency and durability of your applications. These DLQs address different failure modes, and can be used together.

  • SNS DLQs store messages that failed to be delivered to subscribed endpoints.
  • SQS DLQs store messages that the consumer system failed to process.
  • Lambda DLQs store the messages that resulted in failed asynchronous executions of your functions.

Setting up DLQs for subscriptions, queues, and functions can be done using the AWS Management Console, SDK, CLI, API, or CloudFormation. DLQs are available in all AWS Regions. Start today by running the tutorials: