Tag Archives: family

Using AWS Step Functions State Machines to Handle Workflow-Driven AWS CodePipeline Actions

Post Syndicated from Marcilio Mendonca original https://aws.amazon.com/blogs/devops/using-aws-step-functions-state-machines-to-handle-workflow-driven-aws-codepipeline-actions/

AWS CodePipeline is a continuous integration and continuous delivery service for fast and reliable application and infrastructure updates. It offers powerful integration with other AWS services, such as AWS CodeBuildAWS CodeDeployAWS CodeCommit, AWS CloudFormation and with third-party tools such as Jenkins and GitHub. These services make it possible for AWS customers to successfully automate various tasks, including infrastructure provisioning, blue/green deployments, serverless deployments, AMI baking, database provisioning, and release management.

Developers have been able to use CodePipeline to build sophisticated automation pipelines that often require a single CodePipeline action to perform multiple tasks, fork into different execution paths, and deal with asynchronous behavior. For example, to deploy a Lambda function, a CodePipeline action might first inspect the changes pushed to the code repository. If only the Lambda code has changed, the action can simply update the Lambda code package, create a new version, and point the Lambda alias to the new version. If the changes also affect infrastructure resources managed by AWS CloudFormation, the pipeline action might have to create a stack or update an existing one through the use of a change set. In addition, if an update is required, the pipeline action might enforce a safety policy to infrastructure resources that prevents the deletion and replacement of resources. You can do this by creating a change set and having the pipeline action inspect its changes before updating the stack. Change sets that do not conform to the policy are deleted.

This use case is a good illustration of workflow-driven pipeline actions. These are actions that run multiple tasks, deal with async behavior and loops, need to maintain and propagate state, and fork into different execution paths. Implementing workflow-driven actions directly in CodePipeline can lead to complex pipelines that are hard for developers to understand and maintain. Ideally, a pipeline action should perform a single task and delegate the complexity of dealing with workflow-driven behavior associated with that task to a state machine engine. This would make it possible for developers to build simpler, more intuitive pipelines and allow them to use state machine execution logs to visualize and troubleshoot their pipeline actions.

In this blog post, we discuss how AWS Step Functions state machines can be used to handle workflow-driven actions. We show how a CodePipeline action can trigger a Step Functions state machine and how the pipeline and the state machine are kept decoupled through a Lambda function. The advantages of using state machines include:

  • Simplified logic (complex tasks are broken into multiple smaller tasks).
  • Ease of handling asynchronous behavior (through state machine wait states).
  • Built-in support for choices and processing different execution paths (through state machine choices).
  • Built-in visualization and logging of the state machine execution.

The source code for the sample pipeline, pipeline actions, and state machine used in this post is available at https://github.com/awslabs/aws-codepipeline-stepfunctions.


This figure shows the components in the CodePipeline-Step Functions integration that will be described in this post. The pipeline contains two stages: a Source stage represented by a CodeCommit Git repository and a Prod stage with a single Deploy action that represents the workflow-driven action.

This action invokes a Lambda function (1) called the State Machine Trigger Lambda, which, in turn, triggers a Step Function state machine to process the request (2). The Lambda function sends a continuation token back to the pipeline (3) to continue its execution later and terminates. Seconds later, the pipeline invokes the Lambda function again (4), passing the continuation token received. The Lambda function checks the execution state of the state machine (5,6) and communicates the status to the pipeline. The process is repeated until the state machine execution is complete. Then the Lambda function notifies the pipeline that the corresponding pipeline action is complete (7). If the state machine has failed, the Lambda function will then fail the pipeline action and stop its execution (7). While running, the state machine triggers various Lambda functions to perform different tasks. The state machine and the pipeline are fully decoupled. Their interaction is handled by the Lambda function.

The Deploy State Machine

The sample state machine used in this post is a simplified version of the use case, with emphasis on infrastructure deployment. The state machine will follow distinct execution paths and thus have different outcomes, depending on:

  • The current state of the AWS CloudFormation stack.
  • The nature of the code changes made to the AWS CloudFormation template and pushed into the pipeline.

If the stack does not exist, it will be created. If the stack exists, a change set will be created and its resources inspected by the state machine. The inspection consists of parsing the change set results and detecting whether any resources will be deleted or replaced. If no resources are being deleted or replaced, the change set is allowed to be executed and the state machine completes successfully. Otherwise, the change set is deleted and the state machine completes execution with a failure as the terminal state.

Let’s dive into each of these execution paths.

Path 1: Create a Stack and Succeed Deployment

The Deploy state machine is shown here. It is triggered by the Lambda function using the following input parameters stored in an S3 bucket.

Create New Stack Execution Path

    "environmentName": "prod",
    "stackName": "sample-lambda-app",
    "templatePath": "infra/Lambda-template.yaml",
    "revisionS3Bucket": "codepipeline-us-east-1-418586629775",
    "revisionS3Key": "StepFunctionsDrivenD/CodeCommit/sjcmExZ"

Note that some values used here are for the use case example only. Account-specific parameters like revisionS3Bucket and revisionS3Key will be different when you deploy this use case in your account.

These input parameters are used by various states in the state machine and passed to the corresponding Lambda functions to perform different tasks. For example, stackName is used to create a stack, check the status of stack creation, and create a change set. The environmentName represents the environment (for example, dev, test, prod) to which the code is being deployed. It is used to prefix the name of stacks and change sets.

With the exception of built-in states such as wait and choice, each state in the state machine invokes a specific Lambda function.  The results received from the Lambda invocations are appended to the state machine’s original input. When the state machine finishes its execution, several parameters will have been added to its original input.

The first stage in the state machine is “Check Stack Existence”. It checks whether a stack with the input name specified in the stackName input parameter already exists. The output of the state adds a Boolean value called doesStackExist to the original state machine input as follows:

  "doesStackExist": true,
  "environmentName": "prod",
  "stackName": "sample-lambda-app",
  "templatePath": "infra/lambda-template.yaml",
  "revisionS3Bucket": "codepipeline-us-east-1-418586629775",
  "revisionS3Key": "StepFunctionsDrivenD/CodeCommit/sjcmExZ",

The following stage, “Does Stack Exist?”, is represented by Step Functions built-in choice state. It checks the value of doesStackExist to determine whether a new stack needs to be created (doesStackExist=true) or a change set needs to be created and inspected (doesStackExist=false).

If the stack does not exist, the states illustrated in green in the preceding figure are executed. This execution path creates the stack, waits until the stack is created, checks the status of the stack’s creation, and marks the deployment successful after the stack has been created. Except for “Stack Created?” and “Wait Stack Creation,” each of these stages invokes a Lambda function. “Stack Created?” and “Wait Stack Creation” are implemented by using the built-in choice state (to decide which path to follow) and the wait state (to wait a few seconds before proceeding), respectively. Each stage adds the results of their Lambda function executions to the initial input of the state machine, allowing future stages to process them.

Path 2: Safely Update a Stack and Mark Deployment as Successful

Safely Update a Stack and Mark Deployment as Successful Execution Path

If the stack indicated by the stackName parameter already exists, a different path is executed. (See the green states in the figure.) This path will create a change set and use wait and choice states to wait until the change set is created. Afterwards, a stage in the execution path will inspect  the resources affected before the change set is executed.

The inspection procedure represented by the “Inspect Change Set Changes” stage consists of parsing the resources affected by the change set and checking whether any of the existing resources are being deleted or replaced. The following is an excerpt of the algorithm, where changeSetChanges.Changes is the object representing the change set changes:

for (var i = 0; i < changeSetChanges.Changes.length; i++) {
    var change = changeSetChanges.Changes[i];
    if (change.Type == "Resource") {
        if (change.ResourceChange.Action == "Delete") {
        if (change.ResourceChange.Action == "Modify") {
            if (change.ResourceChange.Replacement == "True") {

The algorithm returns different values to indicate whether the change set can be safely executed (CAN_SAFELY_UPDATE_EXISTING_STACK or RESOURCES_BEING_DELETED_OR_REPLACED). This value is used later by the state machine to decide whether to execute the change set and update the stack or interrupt the deployment.

The output of the “Inspect Change Set” stage is shown here.

  "environmentName": "prod",
  "stackName": "sample-lambda-app",
  "templatePath": "infra/lambda-template.yaml",
  "revisionS3Bucket": "codepipeline-us-east-1-418586629775",
  "revisionS3Key": "StepFunctionsDrivenD/CodeCommit/sjcmExZ",
  "doesStackExist": true,
  "changeSetName": "prod-sample-lambda-app-change-set-545",
  "changeSetCreationStatus": "complete",

At this point, these parameters have been added to the state machine’s original input:

  • changeSetName, which is added by the “Create Change Set” state.
  • changeSetCreationStatus, which is added by the “Get Change Set Creation Status” state.
  • changeSetAction, which is added by the “Inspect Change Set Changes” state.

The “Safe to Update Infra?” step is a choice state (its JSON spec follows) that simply checks the value of the changeSetAction parameter. If the value is equal to “CAN-SAFELY-UPDATE-EXISTING-STACK“, meaning that no resources will be deleted or replaced, the step will execute the change set by proceeding to the “Execute Change Set” state. The deployment is successful (the state machine completes its execution successfully).

"Safe to Update Infra?": {
      "Type": "Choice",
      "Choices": [
          "Variable": "$.taskParams.changeSetAction",
          "StringEquals": "CAN-SAFELY-UPDATE-EXISTING-STACK",
          "Next": "Execute Change Set"
      "Default": "Deployment Failed"

Path 3: Reject Stack Update and Fail Deployment

Reject Stack Update and Fail Deployment Execution Path

If the changeSetAction parameter is different from “CAN-SAFELY-UPDATE-EXISTING-STACK“, the state machine will interrupt the deployment by deleting the change set and proceeding to the “Deployment Fail” step, which is a built-in Fail state. (Its JSON spec follows.) This state causes the state machine to stop in a failed state and serves to indicate to the Lambda function that the pipeline deployment should be interrupted in a fail state as well.

 "Deployment Failed": {
      "Type": "Fail",
      "Cause": "Deployment Failed",
      "Error": "Deployment Failed"

In all three scenarios, there’s a state machine’s visual representation available in the AWS Step Functions console that makes it very easy for developers to identify what tasks have been executed or why a deployment has failed. Developers can also inspect the inputs and outputs of each state and look at the state machine Lambda function’s logs for details. Meanwhile, the corresponding CodePipeline action remains very simple and intuitive for developers who only need to know whether the deployment was successful or failed.

The State Machine Trigger Lambda Function

The Trigger Lambda function is invoked directly by the Deploy action in CodePipeline. The CodePipeline action must pass a JSON structure to the trigger function through the UserParameters attribute, as follows:

  "s3Bucket": "codepipeline-StepFunctions-sample",
  "stateMachineFile": "state_machine_input.json"

The s3Bucket parameter specifies the S3 bucket location for the state machine input parameters file. The stateMachineFile parameter specifies the file holding the input parameters. By being able to specify different input parameters to the state machine, we make the Trigger Lambda function and the state machine reusable across environments. For example, the same state machine could be called from a test and prod pipeline action by specifying a different S3 bucket or state machine input file for each environment.

The Trigger Lambda function performs two main tasks: triggering the state machine and checking the execution state of the state machine. Its core logic is shown here:

exports.index = function (event, context, callback) {
    try {
        console.log("Event: " + JSON.stringify(event));
        console.log("Context: " + JSON.stringify(context));
        console.log("Environment Variables: " + JSON.stringify(process.env));
        if (Util.isContinuingPipelineTask(event)) {
            monitorStateMachineExecution(event, context, callback);
        else {
            triggerStateMachine(event, context, callback);
    catch (err) {
        failure(Util.jobId(event), callback, context.invokeid, err.message);

Util.isContinuingPipelineTask(event) is a utility function that checks if the Trigger Lambda function is being called for the first time (that is, no continuation token is passed by CodePipeline) or as a continuation of a previous call. In its first execution, the Lambda function will trigger the state machine and send a continuation token to CodePipeline that contains the state machine execution ARN. The state machine ARN is exposed to the Lambda function through a Lambda environment variable called stateMachineArn. Here is the code that triggers the state machine:

function triggerStateMachine(event, context, callback) {
    var stateMachineArn = process.env.stateMachineArn;
    var s3Bucket = Util.actionUserParameter(event, "s3Bucket");
    var stateMachineFile = Util.actionUserParameter(event, "stateMachineFile");
    getStateMachineInputData(s3Bucket, stateMachineFile)
        .then(function (data) {
            var initialParameters = data.Body.toString();
            var stateMachineInputJSON = createStateMachineInitialInput(initialParameters, event);
            console.log("State machine input JSON: " + JSON.stringify(stateMachineInputJSON));
            return stateMachineInputJSON;
        .then(function (stateMachineInputJSON) {
            return triggerStateMachineExecution(stateMachineArn, stateMachineInputJSON);
        .then(function (triggerStateMachineOutput) {
            var continuationToken = { "stateMachineExecutionArn": triggerStateMachineOutput.executionArn };
            var message = "State machine has been triggered: " + JSON.stringify(triggerStateMachineOutput) + ", continuationToken: " + JSON.stringify(continuationToken);
            return continueExecution(Util.jobId(event), continuationToken, callback, message);
        .catch(function (err) {
            console.log("Error triggering state machine: " + stateMachineArn + ", Error: " + err.message);
            failure(Util.jobId(event), callback, context.invokeid, err.message);

The Trigger Lambda function fetches the state machine input parameters from an S3 file, triggers the execution of the state machine using the input parameters and the stateMachineArn environment variable, and signals to CodePipeline that the execution should continue later by passing a continuation token that contains the state machine execution ARN. In case any of these operations fail and an exception is thrown, the Trigger Lambda function will fail the pipeline immediately by signaling a pipeline failure through the putJobFailureResult CodePipeline API.

If the Lambda function is continuing a previous execution, it will extract the state machine execution ARN from the continuation token and check the status of the state machine, as shown here.

function monitorStateMachineExecution(event, context, callback) {
    var stateMachineArn = process.env.stateMachineArn;
    var continuationToken = JSON.parse(Util.continuationToken(event));
    var stateMachineExecutionArn = continuationToken.stateMachineExecutionArn;
        .then(function (response) {
            if (response.status === "RUNNING") {
                var message = "Execution: " + stateMachineExecutionArn + " of state machine: " + stateMachineArn + " is still " + response.status;
                return continueExecution(Util.jobId(event), continuationToken, callback, message);
            if (response.status === "SUCCEEDED") {
                var message = "Execution: " + stateMachineExecutionArn + " of state machine: " + stateMachineArn + " has: " + response.status;
                return success(Util.jobId(event), callback, message);
            var message = "Execution: " + stateMachineExecutionArn + " of state machine: " + stateMachineArn + " has: " + response.status;
            return failure(Util.jobId(event), callback, context.invokeid, message);
        .catch(function (err) {
            var message = "Error monitoring execution: " + stateMachineExecutionArn + " of state machine: " + stateMachineArn + ", Error: " + err.message;
            failure(Util.jobId(event), callback, context.invokeid, message);

If the state machine is in the RUNNING state, the Lambda function will send the continuation token back to the CodePipeline action. This will cause CodePipeline to call the Lambda function again a few seconds later. If the state machine has SUCCEEDED, then the Lambda function will notify the CodePipeline action that the action has succeeded. In any other case (FAILURE, TIMED-OUT, or ABORT), the Lambda function will fail the pipeline action.

This behavior is especially useful for developers who are building and debugging a new state machine because a bug in the state machine can potentially leave the pipeline action hanging for long periods of time until it times out. The Trigger Lambda function prevents this.

Also, by having the Trigger Lambda function as a means to decouple the pipeline and state machine, we make the state machine more reusable. It can be triggered from anywhere, not just from a CodePipeline action.

The Pipeline in CodePipeline

Our sample pipeline contains two simple stages: the Source stage represented by a CodeCommit Git repository and the Prod stage, which contains the Deploy action that invokes the Trigger Lambda function. When the state machine decides that the change set created must be rejected (because it replaces or deletes some the existing production resources), it fails the pipeline without performing any updates to the existing infrastructure. (See the failed Deploy action in red.) Otherwise, the pipeline action succeeds, indicating that the existing provisioned infrastructure was either created (first run) or updated without impacting any resources. (See the green Deploy stage in the pipeline on the left.)

The Pipeline in CodePipeline

The JSON spec for the pipeline’s Prod stage is shown here. We use the UserParameters attribute to pass the S3 bucket and state machine input file to the Lambda function. These parameters are action-specific, which means that we can reuse the state machine in another pipeline action.

  "name": "Prod",
  "actions": [
          "inputArtifacts": [
                  "name": "CodeCommitOutput"
          "name": "Deploy",
          "actionTypeId": {
              "category": "Invoke",
              "owner": "AWS",
              "version": "1",
              "provider": "Lambda"
          "outputArtifacts": [],
          "configuration": {
              "FunctionName": "StateMachineTriggerLambda",
              "UserParameters": "{\"s3Bucket\": \"codepipeline-StepFunctions-sample\", \"stateMachineFile\": \"state_machine_input.json\"}"
          "runOrder": 1


In this blog post, we discussed how state machines in AWS Step Functions can be used to handle workflow-driven actions. We showed how a Lambda function can be used to fully decouple the pipeline and the state machine and manage their interaction. The use of a state machine greatly simplified the associated CodePipeline action, allowing us to build a much simpler and cleaner pipeline while drilling down into the state machine’s execution for troubleshooting or debugging.

Here are two exercises you can complete by using the source code.

Exercise #1: Do not fail the state machine and pipeline action after inspecting a change set that deletes or replaces resources. Instead, create a stack with a different name (think of blue/green deployments). You can do this by creating a state machine transition between the “Safe to Update Infra?” and “Create Stack” stages and passing a new stack name as input to the “Create Stack” stage.

Exercise #2: Add wait logic to the state machine to wait until the change set completes its execution before allowing the state machine to proceed to the “Deployment Succeeded” stage. Use the stack creation case as an example. You’ll have to create a Lambda function (similar to the Lambda function that checks the creation status of a stack) to get the creation status of the change set.

Have fun and share your thoughts!

About the Author

Marcilio Mendonca is a Sr. Consultant in the Canadian Professional Services Team at Amazon Web Services. He has helped AWS customers design, build, and deploy best-in-class, cloud-native AWS applications using VMs, containers, and serverless architectures. Before he joined AWS, Marcilio was a Software Development Engineer at Amazon. Marcilio also holds a Ph.D. in Computer Science. In his spare time, he enjoys playing drums, riding his motorcycle in the Toronto GTA area, and spending quality time with his family.

Amazon Redshift Dense Compute (DC2) Nodes Deliver Twice the Performance as DC1 at the Same Price

Post Syndicated from Quaseer Mujawar original https://aws.amazon.com/blogs/big-data/amazon-redshift-dense-compute-dc2-nodes-deliver-twice-the-performance-as-dc1-at-the-same-price/

Amazon Redshift makes analyzing exabyte-scale data fast, simple, and cost-effective. It delivers advanced data warehousing capabilities, including parallel execution, compressed columnar storage, and end-to-end encryption as a fully managed service, for less than $1,000/TB/year. With Amazon Redshift Spectrum, you can run SQL queries directly against exabytes of unstructured data in Amazon S3 for $5/TB scanned.

Today, we are making our Dense Compute (DC) family faster and more cost-effective with new second-generation Dense Compute (DC2) nodes at the same price as our previous generation DC1. DC2 is designed for demanding data warehousing workloads that require low latency and high throughput. DC2 features powerful Intel E5-2686 v4 (Broadwell) CPUs, fast DDR4 memory, and NVMe-based solid state disks.

We’ve tuned Amazon Redshift to take advantage of the better CPU, network, and disk on DC2 nodes, providing up to twice the performance of DC1 at the same price. Our DC2.8xlarge instances now provide twice the memory per slice of data and an optimized storage layout with 30 percent better storage utilization.

Customer successes

Several flagship customers, ranging from fast growing startups to large Fortune 100 companies, previewed the new DC2 node type. In their tests, DC2 provided up to twice the performance as DC1. Our preview customers saw faster ETL (extract, transform, and load) jobs, higher query throughput, better concurrency, faster reports, and shorter data-to-insights—all at the same cost as DC1. DC2.8xlarge customers also noted that their databases used up to 30 percent less disk space due to our optimized storage format, reducing their costs.

4Cite Marketing, one of America’s fastest growing private companies, uses Amazon Redshift to analyze customer data and determine personalized product recommendations for retailers. “Amazon Redshift’s new DC2 node is giving us a 100 percent performance increase, allowing us to provide faster insights for our retailers, more cost-effectively, to drive incremental revenue,” said Jim Finnerty, 4Cite’s senior vice president of product.

BrandVerity, a Seattle-based brand protection and compliance‎ company, provides solutions to monitor, detect, and mitigate online brand, trademark, and compliance abuse. “We saw a 70 percent performance boost with the DC2 nodes for running Redshift Spectrum queries. As a result, we can analyze far more data for our customers and deliver results much faster,” said Hyung-Joon Kim, principal software engineer at BrandVerity.

“Amazon Redshift is at the core of our operations and our marketing automation tools,” said Jarno Kartela, head of analytics and chief data scientist at DNA Plc, one of the leading Finnish telecommunications groups and Finland’s largest cable operator and pay TV provider. “We saw a 52 percent performance gain in moving to Amazon Redshift’s DC2 nodes. We can now run queries in half the time, allowing us to provide more analytics power and reduce time-to-insight for our analytics and marketing automation users.”

You can read about their experiences on our Customer Success page.

Get started

You can try the new node type using our getting started guide. Just choose dc2.large or dc2.8xlarge in the Amazon Redshift console:

If you have a DC1.large Amazon Redshift cluster, you can restore to a new DC2.large cluster using an existing snapshot. To migrate from DS2.xlarge, DS2.8xlarge, or DC1.8xlarge Amazon Redshift clusters, you can use the resize operation to move data to your new DC2 cluster. For more information, see Clusters and Nodes in Amazon Redshift.

To get the latest Amazon Redshift feature announcements, check out our What’s New page, and subscribe to the RSS feed.

PureVPN Explains How it Helped the FBI Catch a Cyberstalker

Post Syndicated from Andy original https://torrentfreak.com/purevpn-explains-how-it-helped-the-fbi-catch-a-cyberstalker-171016/

Early October, Ryan S. Lin, 24, of Newton, Massachusetts, was arrested on suspicion of conducting “an extensive cyberstalking campaign” against a 24-year-old Massachusetts woman, as well as her family members and friends.

The Department of Justice described Lin’s offenses as a “multi-faceted” computer hacking and cyberstalking campaign. Launched in April 2016 when he began hacking into the victim’s online accounts, Lin allegedly obtained personal photographs and sensitive information about her medical and sexual histories and distributed that information to hundreds of other people.

Details of what information the FBI compiled on Lin can be found in our earlier report but aside from his alleged crimes (which are both significant and repugnant), it was PureVPN’s involvement in the case that caused the most controversy.

In a report compiled by an FBI special agent, it was revealed that the Hong Kong-based company’s logs helped the authorities net the alleged criminal.

“Significantly, PureVPN was able to determine that their service was accessed by the same customer from two originating IP addresses: the RCN IP address from the home Lin was living in at the time, and the software company where Lin was employed at the time,” the agent’s affidavit reads.

Among many in the privacy community, this revelation was met with disappointment. On the PureVPN website the company claims to carry no logs and on a general basis, it’s expected that so-called “no-logging” VPN providers should provide people with some anonymity, at least as far as their service goes. Now, several days after the furor, the company has responded to its critics.

In a fairly lengthy statement, the company begins by confirming that it definitely doesn’t log what websites a user views or what content he or she downloads.

“PureVPN did not breach its Privacy Policy and certainly did not breach your trust. NO browsing logs, browsing habits or anything else was, or ever will be shared,” the company writes.

However, that’s only half the problem. While it doesn’t log user activity (what sites people visit or content they download), it does log the IP addresses that customers use to access the PureVPN service. These, given the right circumstances, can be matched to external activities thanks to logs carried by other web companies.

PureVPN talks about logs held by Google’s Gmail service to illustrate its point.

“A network log is automatically generated every time a user visits a website. For the sake of this example, let’s say a user logged into their Gmail account. Every time they accessed Gmail, the email provider created a network log,” the company explains.

“If you are using a VPN, Gmail’s network log would contain the IP provided by PureVPN. This is one half of the picture. Now, if someone asks Google who accessed the user’s account, Google would state that whoever was using this IP, accessed the account.

“If the user was connected to PureVPN, it would be a PureVPN IP. The inquirer [in the Lin case, the FBI] would then share timestamps and network logs acquired from Google and ask them to be compared with the network logs maintained by the VPN provider.”

Now, if PureVPN carried no logs – literally no logs – it would not be able to help with this kind of inquiry. That was the case last year when the FBI approached Private Internet Access for information and the company was unable to assist.

However, as is made pretty clear by PureVPN’s explanation, the company does log user IP addresses and timestamps which reveal when a user was logged on to the service. It doesn’t matter that PureVPN doesn’t log what the user allegedly did online, since the third-party service already knows that information to the precise second.

Following the example, GMail knows that a user sent an email at 10:22am on Monday October 16 from a PureVPN IP address. So, if PureVPN is approached by the FBI, the company can confirm that User X was using the same IP address at exactly the same time, and his home IP address was XXX.XX.XXX.XX. Effectively, the combined logs link one IP address to the other and the user is revealed. It’s that simple.

It is for this reason that in TorrentFreak’s annual summary of no-logging VPN providers, the very first question we ask every single company reads as follows:

Do you keep ANY logs which would allow you to match an IP-address and a time stamp to a user/users of your service? If so, what information do you hold and for how long?

Clearly, if a company says “yes we log incoming IP addresses and associated timestamps”, any claim to total user anonymity is ended right there and then.

While not completely useless (a logging service will still stop the prying eyes of ISPs and similar surveillance, while also defeating throttling and site-blocking), if you’re a whistle-blower with a job or even your life to protect, this level of protection is entirely inadequate.

The take-home points from this controversy are numerous, but perhaps the most important is for people to read and understand VPN provider logging policies.

Secondly, and just as importantly, VPN providers need to be extremely clear about the information they log. Not tracking browsing or downloading activities is all well and good, but if home IP addresses and timestamps are stored, this needs to be made clear to the customer.

Finally, VPN users should not be evil. There are plenty of good reasons to stay anonymous online but cyberstalking, death threats and ruining people’s lives are not included. Fortunately, the FBI have offline methods for catching this type of offender, and long may that continue.

PureVPN’s blog post is available here.

Source: TF, for the latest info on copyright, file-sharing, torrent sites and ANONYMOUS VPN services.

Predict Billboard Top 10 Hits Using RStudio, H2O and Amazon Athena

Post Syndicated from Gopal Wunnava original https://aws.amazon.com/blogs/big-data/predict-billboard-top-10-hits-using-rstudio-h2o-and-amazon-athena/

Success in the popular music industry is typically measured in terms of the number of Top 10 hits artists have to their credit. The music industry is a highly competitive multi-billion dollar business, and record labels incur various costs in exchange for a percentage of the profits from sales and concert tickets.

Predicting the success of an artist’s release in the popular music industry can be difficult. One release may be extremely popular, resulting in widespread play on TV, radio and social media, while another single may turn out quite unpopular, and therefore unprofitable. Record labels need to be selective in their decision making, and predictive analytics can help them with decision making around the type of songs and artists they need to promote.

In this walkthrough, you leverage H2O.ai, Amazon Athena, and RStudio to make predictions on whether a song might make it to the Top 10 Billboard charts. You explore the GLM, GBM, and deep learning modeling techniques using H2O’s rapid, distributed and easy-to-use open source parallel processing engine. RStudio is a popular IDE, licensed either commercially or under AGPLv3, for working with R. This is ideal if you don’t want to connect to a server via SSH and use code editors such as vi to do analytics. RStudio is available in a desktop version, or a server version that allows you to access R via a web browser. RStudio’s Notebooks feature is used to demonstrate the execution of code and output. In addition, this post showcases how you can leverage Athena for query and interactive analysis during the modeling phase. A working knowledge of statistics and machine learning would be helpful to interpret the analysis being performed in this post.


Your goal is to predict whether a song will make it to the Top 10 Billboard charts. For this purpose, you will be using multiple modeling techniques―namely GLM, GBM and deep learning―and choose the model that is the best fit.

This solution involves the following steps:

  • Install and configure RStudio with Athena
  • Log in to RStudio
  • Install R packages
  • Connect to Athena
  • Create a dataset
  • Create models

Install and configure RStudio with Athena

Use the following AWS CloudFormation stack to install, configure, and connect RStudio on an Amazon EC2 instance with Athena.

Launching this stack creates all required resources and prerequisites:

  • Amazon EC2 instance with Amazon Linux (minimum size of t2.large is recommended)
  • Provisioning of the EC2 instance in an existing VPC and public subnet
  • Installation of Java 8
  • Assignment of an IAM role to the EC2 instance with the required permissions for accessing Athena and Amazon S3
  • Security group allowing access to the RStudio and SSH ports from the internet (I recommend restricting access to these ports)
  • S3 staging bucket required for Athena (referenced within RStudio as ATHENABUCKET)
  • RStudio username and password
  • Setup logs in Amazon CloudWatch Logs (if needed for additional troubleshooting)
  • Amazon EC2 Systems Manager agent, which makes it easy to manage and patch

All AWS resources are created in the US-East-1 Region. To avoid cross-region data transfer fees, launch the CloudFormation stack in the same region. To check the availability of Athena in other regions, see Region Table.

Log in to RStudio

The instance security group has been automatically configured to allow incoming connections on the RStudio port 8787 from any source internet address. You can edit the security group to restrict source IP access. If you have trouble connecting, ensure that port 8787 isn’t blocked by subnet network ACLS or by your outgoing proxy/firewall.

  1. In the CloudFormation stack, choose Outputs, Value, and then open the RStudio URL. You might need to wait for a few minutes until the instance has been launched.
  2. Log in to RStudio with the and password you provided during setup.

Install R packages

Next, install the required R packages from the RStudio console. You can download the R notebook file containing just the code.

#install pacman – a handy package manager for managing installs
if("pacman" %in% rownames(installed.packages()) == FALSE)
h2o.init(nthreads = -1)
##  Connection successful!
## R is connected to the H2O cluster: 
##     H2O cluster uptime:         2 hours 42 minutes 
##     H2O cluster version: 
##     H2O cluster version age:    4 months and 4 days !!! 
##     H2O cluster name:           H2O_started_from_R_rstudio_hjx881 
##     H2O cluster total nodes:    1 
##     H2O cluster total memory:   3.30 GB 
##     H2O cluster total cores:    4 
##     H2O cluster allowed cores:  4 
##     H2O cluster healthy:        TRUE 
##     H2O Connection ip:          localhost 
##     H2O Connection port:        54321 
##     H2O Connection proxy:       NA 
##     H2O Internal Security:      FALSE 
##     R Version:                  R version 3.3.3 (2017-03-06)
## Warning in h2o.clusterInfo(): 
## Your H2O cluster version is too old (4 months and 4 days)!
## Please download and install the latest version from http://h2o.ai/download/
#install aws sdk if not present (pre-requisite for using Athena with an IAM role)
if (!aws_sdk_present()) {


Connect to Athena

Next, establish a connection to Athena from RStudio, using an IAM role associated with your EC2 instance. Use ATHENABUCKET to specify the S3 staging directory.

URL <- 'https://s3.amazonaws.com/athena-downloads/drivers/AthenaJDBC41-1.0.1.jar'
fil <- basename(URL)
#download the file into current working directory
if (!file.exists(fil)) download.file(URL, fil)
#verify that the file has been downloaded successfully
## [1] "AthenaJDBC41-1.0.1.jar"
drv <- JDBC(driverClass="com.amazonaws.athena.jdbc.AthenaDriver", fil, identifier.quote="'")

con <- jdbcConnection <- dbConnect(drv, 'jdbc:awsathena://athena.us-east-1.amazonaws.com:443/',

Verify the connection. The results returned depend on your specific Athena setup.

## <JDBCConnection>
##  [1] "gdelt"               "wikistats"           "elb_logs_raw_native"
##  [4] "twitter"             "twitter2"            "usermovieratings"   
##  [7] "eventcodes"          "events"              "billboard"          
## [10] "billboardtop10"      "elb_logs"            "gdelthist"          
## [13] "gdeltmaster"         "twitter"             "twitter3"

Create a dataset

For this analysis, you use a sample dataset combining information from Billboard and Wikipedia with Echo Nest data in the Million Songs Dataset. Upload this dataset into your own S3 bucket. The table below provides a description of the fields used in this dataset.

Field Description
year Year that song was released
songtitle Title of the song
artistname Name of the song artist
songid Unique identifier for the song
artistid Unique identifier for the song artist
timesignature Variable estimating the time signature of the song
timesignature_confidence Confidence in the estimate for the timesignature
loudness Continuous variable indicating the average amplitude of the audio in decibels
tempo Variable indicating the estimated beats per minute of the song
tempo_confidence Confidence in the estimate for tempo
key Variable with twelve levels indicating the estimated key of the song (C, C#, B)
key_confidence Confidence in the estimate for key
energy Variable that represents the overall acoustic energy of the song, using a mix of features such as loudness
pitch Continuous variable that indicates the pitch of the song
timbre_0_min thru timbre_11_min Variables that indicate the minimum values over all segments for each of the twelve values in the timbre vector
timbre_0_max thru timbre_11_max Variables that indicate the maximum values over all segments for each of the twelve values in the timbre vector
top10 Indicator for whether or not the song made it to the Top 10 of the Billboard charts (1 if it was in the top 10, and 0 if not)

Create an Athena table based on the dataset

In the Athena console, select the default database, sampled, or create a new database.

Run the following create table statement.

create external table if not exists billboard
year int,
songtitle string,
artistname string,
songID string,
artistID string,
timesignature int,
timesignature_confidence double,
loudness double,
tempo double,
tempo_confidence double,
key int,
key_confidence double,
energy double,
pitch double,
timbre_0_min double,
timbre_0_max double,
timbre_1_min double,
timbre_1_max double,
timbre_2_min double,
timbre_2_max double,
timbre_3_min double,
timbre_3_max double,
timbre_4_min double,
timbre_4_max double,
timbre_5_min double,
timbre_5_max double,
timbre_6_min double,
timbre_6_max double,
timbre_7_min double,
timbre_7_max double,
timbre_8_min double,
timbre_8_max double,
timbre_9_min double,
timbre_9_max double,
timbre_10_min double,
timbre_10_max double,
timbre_11_min double,
timbre_11_max double,
Top10 int
LOCATION 's3://aws-bigdata-blog/artifacts/predict-billboard/data'

Inspect the table definition for the ‘billboard’ table that you have created. If you chose a database other than sampledb, replace that value with your choice.

dbGetQuery(con, "show create table sampledb.billboard")
##                                      createtab_stmt
## 1       CREATE EXTERNAL TABLE `sampledb.billboard`(
## 2                                       `year` int,
## 3                               `songtitle` string,
## 4                              `artistname` string,
## 5                                  `songid` string,
## 6                                `artistid` string,
## 7                              `timesignature` int,
## 8                `timesignature_confidence` double,
## 9                                `loudness` double,
## 10                                  `tempo` double,
## 11                       `tempo_confidence` double,
## 12                                       `key` int,
## 13                         `key_confidence` double,
## 14                                 `energy` double,
## 15                                  `pitch` double,
## 16                           `timbre_0_min` double,
## 17                           `timbre_0_max` double,
## 18                           `timbre_1_min` double,
## 19                           `timbre_1_max` double,
## 20                           `timbre_2_min` double,
## 21                           `timbre_2_max` double,
## 22                           `timbre_3_min` double,
## 23                           `timbre_3_max` double,
## 24                           `timbre_4_min` double,
## 25                           `timbre_4_max` double,
## 26                           `timbre_5_min` double,
## 27                           `timbre_5_max` double,
## 28                           `timbre_6_min` double,
## 29                           `timbre_6_max` double,
## 30                           `timbre_7_min` double,
## 31                           `timbre_7_max` double,
## 32                           `timbre_8_min` double,
## 33                           `timbre_8_max` double,
## 34                           `timbre_9_min` double,
## 35                           `timbre_9_max` double,
## 36                          `timbre_10_min` double,
## 37                          `timbre_10_max` double,
## 38                          `timbre_11_min` double,
## 39                          `timbre_11_max` double,
## 40                                     `top10` int)
## 41                             ROW FORMAT DELIMITED 
## 42                         FIELDS TERMINATED BY ',' 
## 43                            STORED AS INPUTFORMAT 
## 44       'org.apache.hadoop.mapred.TextInputFormat' 
## 45                                     OUTPUTFORMAT 
## 46  'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'
## 47                                        LOCATION
## 48    's3://aws-bigdata-blog/artifacts/predict-billboard/data'
## 49                                  TBLPROPERTIES (
## 50            'transient_lastDdlTime'='1505484133')

Run a sample query

Next, run a sample query to obtain a list of all songs from Janet Jackson that made it to the Billboard Top 10 charts.

dbGetQuery(con, " SELECT songtitle,artistname,top10   FROM sampledb.billboard WHERE lower(artistname) =     'janet jackson' AND top10 = 1")
##                       songtitle    artistname top10
## 1                       Runaway Janet Jackson     1
## 2               Because Of Love Janet Jackson     1
## 3                         Again Janet Jackson     1
## 4                            If Janet Jackson     1
## 5  Love Will Never Do (Without You) Janet Jackson 1
## 6                     Black Cat Janet Jackson     1
## 7               Come Back To Me Janet Jackson     1
## 8                       Alright Janet Jackson     1
## 9                      Escapade Janet Jackson     1
## 10                Rhythm Nation Janet Jackson     1

Determine how many songs in this dataset are specifically from the year 2010.

dbGetQuery(con, " SELECT count(*)   FROM sampledb.billboard WHERE year = 2010")
##   _col0
## 1   373

The sample dataset provides certain song properties of interest that can be analyzed to gauge the impact to the song’s overall popularity. Look at one such property, timesignature, and determine the value that is the most frequent among songs in the database. Timesignature is a measure of the number of beats and the type of note involved.

Running the query directly may result in an error, as shown in the commented lines below. This error is a result of trying to retrieve a large result set over a JDBC connection, which can cause out-of-memory issues at the client level. To address this, reduce the fetch size and run again.

#t<-dbGetQuery(con, " SELECT timesignature FROM sampledb.billboard")
#Note:  Running the preceding query results in the following error: 
#Error in .jcall(rp, "I", "fetch", stride, block): java.sql.SQLException: The requested #fetchSize is more than the allowed value in Athena. Please reduce the fetchSize and try #again. Refer to the Athena documentation for valid fetchSize values.
# Use the dbSendQuery function, reduce the fetch size, and run again
r <- dbSendQuery(con, " SELECT timesignature     FROM sampledb.billboard")
dftimesignature<- fetch(r, n=-1, block=100)
## [1] TRUE
## dftimesignature
##    0    1    3    4    5    7 
##   10  143  503 6787  112   19
## [1] 7574

From the results, observe that 6787 songs have a timesignature of 4.

Next, determine the song with the highest tempo.

dbGetQuery(con, " SELECT songtitle,artistname,tempo   FROM sampledb.billboard WHERE tempo = (SELECT max(tempo) FROM sampledb.billboard) ")
##                   songtitle      artistname   tempo
## 1 Wanna Be Startin' Somethin' Michael Jackson 244.307

Create the training dataset

Your model needs to be trained such that it can learn and make accurate predictions. Split the data into training and test datasets, and create the training dataset first.  This dataset contains all observations from the year 2009 and earlier. You may face the same JDBC connection issue pointed out earlier, so this query uses a fetch size.

#BillboardTrain <- dbGetQuery(con, "SELECT * FROM sampledb.billboard WHERE year <= 2009")
#Running the preceding query results in the following error:-
#Error in .verify.JDBC.result(r, "Unable to retrieve JDBC result set for ", : Unable to retrieve #JDBC result set for SELECT * FROM sampledb.billboard WHERE year <= 2009 (Internal error)
#Follow the same approach as before to address this issue.

r <- dbSendQuery(con, "SELECT * FROM sampledb.billboard WHERE year <= 2009")
BillboardTrain <- fetch(r, n=-1, block=100)
## [1] TRUE
##   year           songtitle artistname timesignature
## 1 2009 The Awkward Goodbye    Athlete             3
## 2 2009        Rubik's Cube    Athlete             3
##   timesignature_confidence loudness   tempo tempo_confidence
## 1                    0.732   -6.320  89.614   0.652
## 2                    0.906   -9.541 117.742   0.542
## [1] 7201

Create the test dataset

BillboardTest <- dbGetQuery(con, "SELECT * FROM sampledb.billboard where year = 2010")
##   year              songtitle        artistname key
## 1 2010 This Is the House That Doubt Built A Day to Remember  11
## 2 2010        Sticks & Bricks A Day to Remember  10
##   key_confidence    energy pitch timbre_0_min
## 1          0.453 0.9666556 0.024        0.002
## 2          0.469 0.9847095 0.025        0.000
## [1] 373

Convert the training and test datasets into H2O dataframes

train.h2o <- as.h2o(BillboardTrain)
  |                                                                 |   0%
  |=================================================================| 100%
test.h2o <- as.h2o(BillboardTest)
  |                                                                 |   0%
  |=================================================================| 100%

Inspect the column names in your H2O dataframes.

##  [1] "year"                     "songtitle"               
##  [3] "artistname"               "songid"                  
##  [5] "artistid"                 "timesignature"           
##  [7] "timesignature_confidence" "loudness"                
##  [9] "tempo"                    "tempo_confidence"        
## [11] "key"                      "key_confidence"          
## [13] "energy"                   "pitch"                   
## [15] "timbre_0_min"             "timbre_0_max"            
## [17] "timbre_1_min"             "timbre_1_max"            
## [19] "timbre_2_min"             "timbre_2_max"            
## [21] "timbre_3_min"             "timbre_3_max"            
## [23] "timbre_4_min"             "timbre_4_max"            
## [25] "timbre_5_min"             "timbre_5_max"            
## [27] "timbre_6_min"             "timbre_6_max"            
## [29] "timbre_7_min"             "timbre_7_max"            
## [31] "timbre_8_min"             "timbre_8_max"            
## [33] "timbre_9_min"             "timbre_9_max"            
## [35] "timbre_10_min"            "timbre_10_max"           
## [37] "timbre_11_min"            "timbre_11_max"           
## [39] "top10"

Create models

You need to designate the independent and dependent variables prior to applying your modeling algorithms. Because you’re trying to predict the ‘top10’ field, this would be your dependent variable and everything else would be independent.

Create your first model using GLM. Because GLM works best with numeric data, you create your model by dropping non-numeric variables. You only use the variables in the dataset that describe the numerical attributes of the song in the logistic regression model. You won’t use these variables:  “year”, “songtitle”, “artistname”, “songid”, or “artistid”.

y.dep <- 39
x.indep <- c(6:38)
##  [1]  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
## [24] 29 30 31 32 33 34 35 36 37 38

Create Model 1: All numeric variables

Create Model 1 with the training dataset, using GLM as the modeling algorithm and H2O’s built-in h2o.glm function.

modelh1 <- h2o.glm( y = y.dep, x = x.indep, training_frame = train.h2o, family = "binomial")
  |                                                                 |   0%
  |=====                                                            |   8%
  |=================================================================| 100%

Measure the performance of Model 1, using H2O’s built-in performance function.

## H2OBinomialMetrics: glm
## MSE:  0.09924684
## RMSE:  0.3150347
## LogLoss:  0.3220267
## Mean Per-Class Error:  0.2380168
## AUC:  0.8431394
## Gini:  0.6862787
## R^2:  0.254663
## Null Deviance:  326.0801
## Residual Deviance:  240.2319
## AIC:  308.2319
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0   1    Error     Rate
## 0      255  59 0.187898  =59/314
## 1       17  42 0.288136   =17/59
## Totals 272 101 0.203753  =76/373
## Maximum Metrics: Maximum metrics at their respective thresholds
##                         metric threshold    value idx
## 1                       max f1  0.192772 0.525000 100
## 2                       max f2  0.124912 0.650510 155
## 3                 max f0point5  0.416258 0.612903  23
## 4                 max accuracy  0.416258 0.879357  23
## 5                max precision  0.813396 1.000000   0
## 6                   max recall  0.037579 1.000000 282
## 7              max specificity  0.813396 1.000000   0
## 8             max absolute_mcc  0.416258 0.455251  23
## 9   max min_per_class_accuracy  0.161402 0.738854 125
## 10 max mean_per_class_accuracy  0.124912 0.765006 155
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or ` 
## [1] 0.8431394

The AUC metric provides insight into how well the classifier is able to separate the two classes. In this case, the value of 0.8431394 indicates that the classification is good. (A value of 0.5 indicates a worthless test, while a value of 1.0 indicates a perfect test.)

Next, inspect the coefficients of the variables in the dataset.

dfmodelh1 <- as.data.frame(h2o.varimp(modelh1))
##                       names coefficients sign
## 1              timbre_0_max  1.290938663  NEG
## 2                  loudness  1.262941934  POS
## 3                     pitch  0.616995941  NEG
## 4              timbre_1_min  0.422323735  POS
## 5              timbre_6_min  0.349016024  NEG
## 6                    energy  0.348092062  NEG
## 7             timbre_11_min  0.307331997  NEG
## 8              timbre_3_max  0.302225619  NEG
## 9             timbre_11_max  0.243632060  POS
## 10             timbre_4_min  0.224233951  POS
## 11             timbre_4_max  0.204134342  POS
## 12             timbre_5_min  0.199149324  NEG
## 13             timbre_0_min  0.195147119  POS
## 14 timesignature_confidence  0.179973904  POS
## 15         tempo_confidence  0.144242598  POS
## 16            timbre_10_max  0.137644568  POS
## 17             timbre_7_min  0.126995955  NEG
## 18            timbre_10_min  0.123851179  POS
## 19             timbre_7_max  0.100031481  NEG
## 20             timbre_2_min  0.096127636  NEG
## 21           key_confidence  0.083115820  POS
## 22             timbre_6_max  0.073712419  POS
## 23            timesignature  0.067241917  POS
## 24             timbre_8_min  0.061301881  POS
## 25             timbre_8_max  0.060041698  POS
## 26                      key  0.056158445  POS
## 27             timbre_3_min  0.050825116  POS
## 28             timbre_9_max  0.033733561  POS
## 29             timbre_2_max  0.030939072  POS
## 30             timbre_9_min  0.020708113  POS
## 31             timbre_1_max  0.014228818  NEG
## 32                    tempo  0.008199861  POS
## 33             timbre_5_max  0.004837870  POS
## 34                                    NA <NA>

Typically, songs with heavier instrumentation tend to be louder (have higher values in the variable “loudness”) and more energetic (have higher values in the variable “energy”). This knowledge is helpful for interpreting the modeling results.

You can make the following observations from the results:

  • The coefficient estimates for the confidence values associated with the time signature, key, and tempo variables are positive. This suggests that higher confidence leads to a higher predicted probability of a Top 10 hit.
  • The coefficient estimate for loudness is positive, meaning that mainstream listeners prefer louder songs with heavier instrumentation.
  • The coefficient estimate for energy is negative, meaning that mainstream listeners prefer songs that are less energetic, which are those songs with light instrumentation.

These coefficients lead to contradictory conclusions for Model 1. This could be due to multicollinearity issues. Inspect the correlation between the variables “loudness” and “energy” in the training set.

## [1] 0.7399067

This number indicates that these two variables are highly correlated, and Model 1 does indeed suffer from multicollinearity. Typically, you associate a value of -1.0 to -0.5 or 1.0 to 0.5 to indicate strong correlation, and a value of 0.1 to 0.1 to indicate weak correlation. To avoid this correlation issue, omit one of these two variables and re-create the models.

You build two variations of the original model:

  • Model 2, in which you keep “energy” and omit “loudness”
  • Model 3, in which you keep “loudness” and omit “energy”

You compare these two models and choose the model with a better fit for this use case.

Create Model 2: Keep energy and omit loudness

##  [1] "year"                     "songtitle"               
##  [3] "artistname"               "songid"                  
##  [5] "artistid"                 "timesignature"           
##  [7] "timesignature_confidence" "loudness"                
##  [9] "tempo"                    "tempo_confidence"        
## [11] "key"                      "key_confidence"          
## [13] "energy"                   "pitch"                   
## [15] "timbre_0_min"             "timbre_0_max"            
## [17] "timbre_1_min"             "timbre_1_max"            
## [19] "timbre_2_min"             "timbre_2_max"            
## [21] "timbre_3_min"             "timbre_3_max"            
## [23] "timbre_4_min"             "timbre_4_max"            
## [25] "timbre_5_min"             "timbre_5_max"            
## [27] "timbre_6_min"             "timbre_6_max"            
## [29] "timbre_7_min"             "timbre_7_max"            
## [31] "timbre_8_min"             "timbre_8_max"            
## [33] "timbre_9_min"             "timbre_9_max"            
## [35] "timbre_10_min"            "timbre_10_max"           
## [37] "timbre_11_min"            "timbre_11_max"           
## [39] "top10"
y.dep <- 39
x.indep <- c(6:7,9:38)
##  [1]  6  7  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
## [24] 30 31 32 33 34 35 36 37 38
modelh2 <- h2o.glm( y = y.dep, x = x.indep, training_frame = train.h2o, family = "binomial")
  |                                                                 |   0%
  |=======                                                          |  10%
  |=================================================================| 100%

Measure the performance of Model 2.

## H2OBinomialMetrics: glm
## MSE:  0.09922606
## RMSE:  0.3150017
## LogLoss:  0.3228213
## Mean Per-Class Error:  0.2490554
## AUC:  0.8431933
## Gini:  0.6863867
## R^2:  0.2548191
## Null Deviance:  326.0801
## Residual Deviance:  240.8247
## AIC:  306.8247
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0  1    Error     Rate
## 0      280 34 0.108280  =34/314
## 1       23 36 0.389831   =23/59
## Totals 303 70 0.152815  =57/373
## Maximum Metrics: Maximum metrics at their respective thresholds
##                         metric threshold    value idx
## 1                       max f1  0.254391 0.558140  69
## 2                       max f2  0.113031 0.647208 157
## 3                 max f0point5  0.413999 0.596026  22
## 4                 max accuracy  0.446250 0.876676  18
## 5                max precision  0.811739 1.000000   0
## 6                   max recall  0.037682 1.000000 283
## 7              max specificity  0.811739 1.000000   0
## 8             max absolute_mcc  0.254391 0.469060  69
## 9   max min_per_class_accuracy  0.141051 0.716561 131
## 10 max mean_per_class_accuracy  0.113031 0.761821 157
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`
dfmodelh2 <- as.data.frame(h2o.varimp(modelh2))
##                       names coefficients sign
## 1                     pitch  0.700331511  NEG
## 2              timbre_1_min  0.510270513  POS
## 3              timbre_0_max  0.402059546  NEG
## 4              timbre_6_min  0.333316236  NEG
## 5             timbre_11_min  0.331647383  NEG
## 6              timbre_3_max  0.252425901  NEG
## 7             timbre_11_max  0.227500308  POS
## 8              timbre_4_max  0.210663865  POS
## 9              timbre_0_min  0.208516163  POS
## 10             timbre_5_min  0.202748055  NEG
## 11             timbre_4_min  0.197246582  POS
## 12            timbre_10_max  0.172729619  POS
## 13         tempo_confidence  0.167523934  POS
## 14 timesignature_confidence  0.167398830  POS
## 15             timbre_7_min  0.142450727  NEG
## 16             timbre_8_max  0.093377516  POS
## 17            timbre_10_min  0.090333426  POS
## 18            timesignature  0.085851625  POS
## 19             timbre_7_max  0.083948442  NEG
## 20           key_confidence  0.079657073  POS
## 21             timbre_6_max  0.076426046  POS
## 22             timbre_2_min  0.071957831  NEG
## 23             timbre_9_max  0.071393189  POS
## 24             timbre_8_min  0.070225578  POS
## 25                      key  0.061394702  POS
## 26             timbre_3_min  0.048384697  POS
## 27             timbre_1_max  0.044721121  NEG
## 28                   energy  0.039698433  POS
## 29             timbre_5_max  0.039469064  POS
## 30             timbre_2_max  0.018461133  POS
## 31                    tempo  0.013279926  POS
## 32             timbre_9_min  0.005282143  NEG
## 33                                    NA <NA>

## [1] 0.8431933

You can make the following observations:

  • The AUC metric is 0.8431933.
  • Inspecting the coefficient of the variable energy, Model 2 suggests that songs with high energy levels tend to be more popular. This is as per expectation.
  • As H2O orders variables by significance, the variable energy is not significant in this model.

You can conclude that Model 2 is not ideal for this use , as energy is not significant.

CreateModel 3: Keep loudness but omit energy

##  [1] "year"                     "songtitle"               
##  [3] "artistname"               "songid"                  
##  [5] "artistid"                 "timesignature"           
##  [7] "timesignature_confidence" "loudness"                
##  [9] "tempo"                    "tempo_confidence"        
## [11] "key"                      "key_confidence"          
## [13] "energy"                   "pitch"                   
## [15] "timbre_0_min"             "timbre_0_max"            
## [17] "timbre_1_min"             "timbre_1_max"            
## [19] "timbre_2_min"             "timbre_2_max"            
## [21] "timbre_3_min"             "timbre_3_max"            
## [23] "timbre_4_min"             "timbre_4_max"            
## [25] "timbre_5_min"             "timbre_5_max"            
## [27] "timbre_6_min"             "timbre_6_max"            
## [29] "timbre_7_min"             "timbre_7_max"            
## [31] "timbre_8_min"             "timbre_8_max"            
## [33] "timbre_9_min"             "timbre_9_max"            
## [35] "timbre_10_min"            "timbre_10_max"           
## [37] "timbre_11_min"            "timbre_11_max"           
## [39] "top10"
y.dep <- 39
x.indep <- c(6:12,14:38)
##  [1]  6  7  8  9 10 11 12 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
## [24] 30 31 32 33 34 35 36 37 38
modelh3 <- h2o.glm( y = y.dep, x = x.indep, training_frame = train.h2o, family = "binomial")
  |                                                                 |   0%
  |========                                                         |  12%
  |=================================================================| 100%
## H2OBinomialMetrics: glm
## MSE:  0.0978859
## RMSE:  0.3128672
## LogLoss:  0.3178367
## Mean Per-Class Error:  0.264925
## AUC:  0.8492389
## Gini:  0.6984778
## R^2:  0.2648836
## Null Deviance:  326.0801
## Residual Deviance:  237.1062
## AIC:  303.1062
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0  1    Error     Rate
## 0      286 28 0.089172  =28/314
## 1       26 33 0.440678   =26/59
## Totals 312 61 0.144772  =54/373
## Maximum Metrics: Maximum metrics at their respective thresholds
##                         metric threshold    value idx
## 1                       max f1  0.273799 0.550000  60
## 2                       max f2  0.125503 0.663265 155
## 3                 max f0point5  0.435479 0.628931  24
## 4                 max accuracy  0.435479 0.882038  24
## 5                max precision  0.821606 1.000000   0
## 6                   max recall  0.038328 1.000000 280
## 7              max specificity  0.821606 1.000000   0
## 8             max absolute_mcc  0.435479 0.471426  24
## 9   max min_per_class_accuracy  0.173693 0.745763 120
## 10 max mean_per_class_accuracy  0.125503 0.775073 155
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`
dfmodelh3 <- as.data.frame(h2o.varimp(modelh3))
##                       names coefficients sign
## 1              timbre_0_max 1.216621e+00  NEG
## 2                  loudness 9.780973e-01  POS
## 3                     pitch 7.249788e-01  NEG
## 4              timbre_1_min 3.891197e-01  POS
## 5              timbre_6_min 3.689193e-01  NEG
## 6             timbre_11_min 3.086673e-01  NEG
## 7              timbre_3_max 3.025593e-01  NEG
## 8             timbre_11_max 2.459081e-01  POS
## 9              timbre_4_min 2.379749e-01  POS
## 10             timbre_4_max 2.157627e-01  POS
## 11             timbre_0_min 1.859531e-01  POS
## 12             timbre_5_min 1.846128e-01  NEG
## 13 timesignature_confidence 1.729658e-01  POS
## 14             timbre_7_min 1.431871e-01  NEG
## 15            timbre_10_max 1.366703e-01  POS
## 16            timbre_10_min 1.215954e-01  POS
## 17         tempo_confidence 1.183698e-01  POS
## 18             timbre_2_min 1.019149e-01  NEG
## 19           key_confidence 9.109701e-02  POS
## 20             timbre_7_max 8.987908e-02  NEG
## 21             timbre_6_max 6.935132e-02  POS
## 22             timbre_8_max 6.878241e-02  POS
## 23            timesignature 6.120105e-02  POS
## 24                      key 5.814805e-02  POS
## 25             timbre_8_min 5.759228e-02  POS
## 26             timbre_1_max 2.930285e-02  NEG
## 27             timbre_9_max 2.843755e-02  POS
## 28             timbre_3_min 2.380245e-02  POS
## 29             timbre_2_max 1.917035e-02  POS
## 30             timbre_5_max 1.715813e-02  POS
## 31                    tempo 1.364418e-02  NEG
## 32             timbre_9_min 8.463143e-05  NEG
## 33                                    NA <NA>
## Warning in h2o.find_row_by_threshold(object, t): Could not find exact
## threshold: 0.5 for this set of metrics; using closest threshold found:
## 0.501855569251422. Run `h2o.predict` and apply your desired threshold on a
## probability column.
## [[1]]
## [1] 0.2033898
## [1] 0.8492389

You can make the following observations:

  • The AUC metric is 0.8492389.
  • From the confusion matrix, the model correctly predicts that 33 songs will be top 10 hits (true positives). However, it has 26 false positives (songs that the model predicted would be Top 10 hits, but ended up not being Top 10 hits).
  • Loudness has a positive coefficient estimate, meaning that this model predicts that songs with heavier instrumentation tend to be more popular. This is the same conclusion from Model 2.
  • Loudness is significant in this model.

Overall, Model 3 predicts a higher number of top 10 hits with an accuracy rate that is acceptable. To choose the best fit for production runs, record labels should consider the following factors:

  • Desired model accuracy at a given threshold
  • Number of correct predictions for top10 hits
  • Tolerable number of false positives or false negatives

Next, make predictions using Model 3 on the test dataset.

predict.regh <- h2o.predict(modelh3, test.h2o)
  |                                                                 |   0%
  |=================================================================| 100%
##   predict        p0          p1
## 1       0 0.9654739 0.034526052
## 2       0 0.9654748 0.034525236
## 3       0 0.9635547 0.036445318
## 4       0 0.9343579 0.065642149
## 5       0 0.9978334 0.002166601
## 6       0 0.9779949 0.022005078
## [373 rows x 3 columns]
##   predict
## 1       0
## 2       0
## 3       0
## 4       0
## 5       0
## 6       0
## [373 rows x 1 column]
#Rename the predicted column 
colnames(dpr)[colnames(dpr) == 'predict'] <- 'predict_top10'
##   0   1 
## 312  61

The first set of output results specifies the probabilities associated with each predicted observation.  For example, observation 1 is 96.54739% likely to not be a Top 10 hit, and 3.4526052% likely to be a Top 10 hit (predict=1 indicates Top 10 hit and predict=0 indicates not a Top 10 hit).  The second set of results list the actual predictions made.  From the third set of results, this model predicts that 61 songs will be top 10 hits.

Compute the baseline accuracy, by assuming that the baseline predicts the most frequent outcome, which is that most songs are not Top 10 hits.

##   0   1 
## 314  59

Now observe that the baseline model would get 314 observations correct, and 59 wrong, for an accuracy of 314/(314+59) = 0.8418231.

It seems that Model 3, with an accuracy of 0.8552, provides you with a small improvement over the baseline model. But is this model useful for record labels?

View the two models from an investment perspective:

  • A production company is interested in investing in songs that are more likely to make it to the Top 10. The company’s objective is to minimize the risk of financial losses attributed to investing in songs that end up unpopular.
  • How many songs does Model 3 correctly predict as a Top 10 hit in 2010? Looking at the confusion matrix, you see that it predicts 33 top 10 hits correctly at an optimal threshold, which is more than half the number
  • It will be more useful to the record label if you can provide the production company with a list of songs that are highly likely to end up in the Top 10.
  • The baseline model is not useful, as it simply does not label any song as a hit.

Considering the three models built so far, you can conclude that Model 3 proves to be the best investment choice for the record label.

GBM model

H2O provides you with the ability to explore other learning models, such as GBM and deep learning. Explore building a model using the GBM technique, using the built-in h2o.gbm function.

Before you do this, you need to convert the target variable to a factor for multinomial classification techniques.

gbm.modelh <- h2o.gbm(y=y.dep, x=x.indep, training_frame = train.h2o, ntrees = 500, max_depth = 4, learn_rate = 0.01, seed = 1122,distribution="multinomial")
  |                                                                 |   0%
  |===                                                              |   5%
  |=====                                                            |   7%
  |======                                                           |   9%
  |=======                                                          |  10%
  |======================                                           |  33%
  |=====================================                            |  56%
  |====================================================             |  79%
  |================================================================ |  98%
  |=================================================================| 100%
## H2OBinomialMetrics: gbm
## MSE:  0.09860778
## RMSE:  0.3140188
## LogLoss:  0.3206876
## Mean Per-Class Error:  0.2120263
## AUC:  0.8630573
## Gini:  0.7261146
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0  1    Error     Rate
## 0      266 48 0.152866  =48/314
## 1       16 43 0.271186   =16/59
## Totals 282 91 0.171582  =64/373
## Maximum Metrics: Maximum metrics at their respective thresholds
##                       metric threshold    value idx
## 1                     max f1  0.189757 0.573333  90
## 2                     max f2  0.130895 0.693717 145
## 3               max f0point5  0.327346 0.598802  26
## 4               max accuracy  0.442757 0.876676  14
## 5              max precision  0.802184 1.000000   0
## 6                 max recall  0.049990 1.000000 284
## 7            max specificity  0.802184 1.000000   0
## 8           max absolute_mcc  0.169135 0.496486 104
## 9 max min_per_class_accuracy  0.169135 0.796610 104
## 10 max mean_per_class_accuracy  0.169135 0.805948 104
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `
## Warning in h2o.find_row_by_threshold(object, t): Could not find exact
## threshold: 0.5 for this set of metrics; using closest threshold found:
## 0.501205344484314. Run `h2o.predict` and apply your desired threshold on a
## probability column.
## [[1]]
## [1] 0.1355932
## [1] 0.8630573

This model correctly predicts 43 top 10 hits, which is 10 more than the number predicted by Model 3. Moreover, the AUC metric is higher than the one obtained from Model 3.

As seen above, H2O’s API provides the ability to obtain key statistical measures required to analyze the models easily, using several built-in functions. The record label can experiment with different parameters to arrive at the model that predicts the maximum number of Top 10 hits at the desired level of accuracy and threshold.

H2O also allows you to experiment with deep learning models. Deep learning models have the ability to learn features implicitly, but can be more expensive computationally.

Now, create a deep learning model with the h2o.deeplearning function, using the same training and test datasets created before. The time taken to run this model depends on the type of EC2 instance chosen for this purpose.  For models that require more computation, consider using accelerated computing instances such as the P2 instance type.

  dlearning.modelh <- h2o.deeplearning(y = y.dep,
                                      x = x.indep,
                                      training_frame = train.h2o,
                                      epoch = 250,
                                      hidden = c(250,250),
                                      activation = "Rectifier",
                                      seed = 1122,
  |                                                                 |   0%
  |===                                                              |   4%
  |=====                                                            |   8%
  |========                                                         |  12%
  |==========                                                       |  16%
  |=============                                                    |  20%
  |================                                                 |  24%
  |==================                                               |  28%
  |=====================                                            |  32%
  |=======================                                          |  36%
  |==========================                                       |  40%
  |=============================                                    |  44%
  |===============================                                  |  48%
  |==================================                               |  52%
  |====================================                             |  56%
  |=======================================                          |  60%
  |==========================================                       |  64%
  |============================================                     |  68%
  |===============================================                  |  72%
  |=================================================                |  76%
  |====================================================             |  80%
  |=======================================================          |  84%
  |=========================================================        |  88%
  |============================================================     |  92%
  |==============================================================   |  96%
  |=================================================================| 100%
##    user  system elapsed 
##   1.216   0.020 166.508
## H2OBinomialMetrics: deeplearning
## MSE:  0.1678359
## RMSE:  0.4096778
## LogLoss:  1.86509
## Mean Per-Class Error:  0.3433013
## AUC:  0.7568822
## Gini:  0.5137644
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0  1    Error     Rate
## 0      290 24 0.076433  =24/314
## 1       36 23 0.610169   =36/59
## Totals 326 47 0.160858  =60/373
## Maximum Metrics: Maximum metrics at their respective thresholds
##                       metric threshold    value idx
## 1                     max f1  0.826267 0.433962  46
## 2                     max f2  0.000000 0.588235 239
## 3               max f0point5  0.999929 0.511811  16
## 4               max accuracy  0.999999 0.865952  10
## 5              max precision  1.000000 1.000000   0
## 6                 max recall  0.000000 1.000000 326
## 7            max specificity  1.000000 1.000000   0
## 8           max absolute_mcc  0.999929 0.363219  16
## 9 max min_per_class_accuracy  0.000004 0.662420 145
## 10 max mean_per_class_accuracy  0.000000 0.685334 224
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`
## Warning in h2o.find_row_by_threshold(object, t): Could not find exact
## threshold: 0.5 for this set of metrics; using closest threshold found:
## 0.496293348880151. Run `h2o.predict` and apply your desired threshold on a
## probability column.
## [[1]]
## [1] 0.3898305
## [1] 0.7568822

The AUC metric for this model is 0.7568822, which is less than what you got from the earlier models. I recommend further experimentation using different hyper parameters, such as the learning rate, epoch or the number of hidden layers.

H2O’s built-in functions provide many key statistical measures that can help measure model performance. Here are some of these key terms.

Metric Description
Sensitivity Measures the proportion of positives that have been correctly identified. It is also called the true positive rate, or recall.
Specificity Measures the proportion of negatives that have been correctly identified. It is also called the true negative rate.
Threshold Cutoff point that maximizes specificity and sensitivity. While the model may not provide the highest prediction at this point, it would not be biased towards positives or negatives.
Precision The fraction of the documents retrieved that are relevant to the information needed, for example, how many of the positively classified are relevant

Provides insight into how well the classifier is able to separate the two classes. The implicit goal is to deal with situations where the sample distribution is highly skewed, with a tendency to overfit to a single class.

0.90 – 1 = excellent (A)

0.8 – 0.9 = good (B)

0.7 – 0.8 = fair (C)

.6 – 0.7 = poor (D)

0.5 – 0.5 = fail (F)

Here’s a summary of the metrics generated from H2O’s built-in functions for the three models that produced useful results.

Metric Model 3 GBM Model Deep Learning Model



















1.0 1.0





1.0 1.0





0.2033898 0.1355932



AUC 0.8492389 0.8630573 0.756882

Note: ‘t’ denotes threshold.

Your options at this point could be narrowed down to Model 3 and the GBM model, based on the AUC and accuracy metrics observed earlier.  If the slightly lower accuracy of the GBM model is deemed acceptable, the record label can choose to go to production with the GBM model, as it can predict a higher number of Top 10 hits.  The AUC metric for the GBM model is also higher than that of Model 3.

Record labels can experiment with different learning techniques and parameters before arriving at a model that proves to be the best fit for their business. Because deep learning models can be computationally expensive, record labels can choose more powerful EC2 instances on AWS to run their experiments faster.


In this post, I showed how the popular music industry can use analytics to predict the type of songs that make the Top 10 Billboard charts. By running H2O’s scalable machine learning platform on AWS, data scientists can easily experiment with multiple modeling techniques and interactively query the data using Amazon Athena, without having to manage the underlying infrastructure. This helps record labels make critical decisions on the type of artists and songs to promote in a timely fashion, thereby increasing sales and revenue.

If you have questions or suggestions, please comment below.

Additional Reading

Learn how to build and explore a simple geospita simple GEOINT application using SparkR.

About the Authors

gopalGopal Wunnava is a Partner Solution Architect with the AWS GSI Team. He works with partners and customers on big data engagements, and is passionate about building analytical solutions that drive business capabilities and decision making. In his spare time, he loves all things sports and movies related and is fond of old classics like Asterix, Obelix comics and Hitchcock movies.



Bob Strahan, a Senior Consultant with AWS Professional Services, contributed to this post.



AWS Developer Tools Expands Integration to Include GitHub

Post Syndicated from Balaji Iyer original https://aws.amazon.com/blogs/devops/aws-developer-tools-expands-integration-to-include-github/

AWS Developer Tools is a set of services that include AWS CodeCommit, AWS CodePipeline, AWS CodeBuild, and AWS CodeDeploy. Together, these services help you securely store and maintain version control of your application’s source code and automatically build, test, and deploy your application to AWS or your on-premises environment. These services are designed to enable developers and IT professionals to rapidly and safely deliver software.

As part of our continued commitment to extend the AWS Developer Tools ecosystem to third-party tools and services, we’re pleased to announce AWS CodeStar and AWS CodeBuild now integrate with GitHub. This will make it easier for GitHub users to set up a continuous integration and continuous delivery toolchain as part of their release process using AWS Developer Tools.

In this post, I will walk through the following:


You’ll need an AWS account, a GitHub account, an Amazon EC2 key pair, and administrator-level permissions for AWS Identity and Access Management (IAM), AWS CodeStar, AWS CodeBuild, AWS CodePipeline, Amazon EC2, Amazon S3.


Integrating GitHub with AWS CodeStar

AWS CodeStar enables you to quickly develop, build, and deploy applications on AWS. Its unified user interface helps you easily manage your software development activities in one place. With AWS CodeStar, you can set up your entire continuous delivery toolchain in minutes, so you can start releasing code faster.

When AWS CodeStar launched in April of this year, it used AWS CodeCommit as the hosted source repository. You can now choose between AWS CodeCommit or GitHub as the source control service for your CodeStar projects. In addition, your CodeStar project dashboard lets you centrally track GitHub activities, including commits, issues, and pull requests. This makes it easy to manage project activity across the components of your CI/CD toolchain. Adding the GitHub dashboard view will simplify development of your AWS applications.

In this section, I will show you how to use GitHub as the source provider for your CodeStar projects. I’ll also show you how to work with recent commits, issues, and pull requests in the CodeStar dashboard.

Sign in to the AWS Management Console and from the Services menu, choose CodeStar. In the CodeStar console, choose Create a new project. You should see the Choose a project template page.

CodeStar Project

Choose an option by programming language, application category, or AWS service. I am going to choose the Ruby on Rails web application that will be running on Amazon EC2.

On the Project details page, you’ll now see the GitHub option. Type a name for your project, and then choose Connect to GitHub.

Project details

You’ll see a message requesting authorization to connect to your GitHub repository. When prompted, choose Authorize, and then type your GitHub account password.


This connects your GitHub identity to AWS CodeStar through OAuth. You can always review your settings by navigating to your GitHub application settings.

Installed GitHub Apps

You’ll see AWS CodeStar is now connected to GitHub:

Create project

You can choose a public or private repository. GitHub offers free accounts for users and organizations working on public and open source projects and paid accounts that offer unlimited private repositories and optional user management and security features.

In this example, I am going to choose the public repository option. Edit the repository description, if you like, and then choose Next.

Review your CodeStar project details, and then choose Create Project. On Choose an Amazon EC2 Key Pair, choose Create Project.

Key Pair

On the Review project details page, you’ll see Edit Amazon EC2 configuration. Choose this link to configure instance type, VPC, and subnet options. AWS CodeStar requires a service role to create and manage AWS resources and IAM permissions. This role will be created for you when you select the AWS CodeStar would like permission to administer AWS resources on your behalf check box.

Choose Create Project. It might take a few minutes to create your project and resources.

Review project details

When you create a CodeStar project, you’re added to the project team as an owner. If this is the first time you’ve used AWS CodeStar, you’ll be asked to provide the following information, which will be shown to others:

  • Your display name.
  • Your email address.

This information is used in your AWS CodeStar user profile. User profiles are not project-specific, but they are limited to a single AWS region. If you are a team member in projects in more than one region, you’ll have to create a user profile in each region.

User settings

User settings

Choose Next. AWS CodeStar will create a GitHub repository with your configuration settings (for example, https://github.com/biyer/ruby-on-rails-service).

When you integrate your integrated development environment (IDE) with AWS CodeStar, you can continue to write and develop code in your preferred environment. The changes you make will be included in the AWS CodeStar project each time you commit and push your code.


After setting up your IDE, choose Next to go to the CodeStar dashboard. Take a few minutes to familiarize yourself with the dashboard. You can easily track progress across your entire software development process, from your backlog of work items to recent code deployments.


After the application deployment is complete, choose the endpoint that will display the application.


This is what you’ll see when you open the application endpoint:

The Commit history section of the dashboard lists the commits made to the Git repository. If you choose the commit ID or the Open in GitHub option, you can use a hotlink to your GitHub repository.

Commit history

Your AWS CodeStar project dashboard is where you and your team view the status of your project resources, including the latest commits to your project, the state of your continuous delivery pipeline, and the performance of your instances. This information is displayed on tiles that are dedicated to a particular resource. To see more information about any of these resources, choose the details link on the tile. The console for that AWS service will open on the details page for that resource.


You can also filter issues based on their status and the assigned user.


AWS CodeBuild Now Supports Building GitHub Pull Requests

CodeBuild is a fully managed build service that compiles source code, runs tests, and produces software packages that are ready to deploy. With CodeBuild, you don’t need to provision, manage, and scale your own build servers. CodeBuild scales continuously and processes multiple builds concurrently, so your builds are not left waiting in a queue. You can use prepackaged build environments to get started quickly or you can create custom build environments that use your own build tools.

We recently announced support for GitHub pull requests in AWS CodeBuild. This functionality makes it easier to collaborate across your team while editing and building your application code with CodeBuild. You can use the AWS CodeBuild or AWS CodePipeline consoles to run AWS CodeBuild. You can also automate the running of AWS CodeBuild by using the AWS Command Line Interface (AWS CLI), the AWS SDKs, or the AWS CodeBuild Plugin for Jenkins.

AWS CodeBuild

In this section, I will show you how to trigger a build in AWS CodeBuild with a pull request from GitHub through webhooks.

Open the AWS CodeBuild console at https://console.aws.amazon.com/codebuild/. Choose Create project. If you already have a CodeBuild project, you can choose Edit project, and then follow along. CodeBuild can connect to AWS CodeCommit, S3, BitBucket, and GitHub to pull source code for builds. For Source provider, choose GitHub, and then choose Connect to GitHub.


After you’ve successfully linked GitHub and your CodeBuild project, you can choose a repository in your GitHub account. CodeBuild also supports connections to any public repository. You can review your settings by navigating to your GitHub application settings.

GitHub Apps

On Source: What to Build, for Webhook, select the Rebuild every time a code change is pushed to this repository check box.

Note: You can select this option only if, under Repository, you chose Use a repository in my account.


In Environment: How to build, for Environment image, select Use an image managed by AWS CodeBuild. For Operating system, choose Ubuntu. For Runtime, choose Base. For Version, choose the latest available version. For Build specification, you can provide a collection of build commands and related settings, in YAML format (buildspec.yml) or you can override the build spec by inserting build commands directly in the console. AWS CodeBuild uses these commands to run a build. In this example, the output is the string “hello.”


On Artifacts: Where to put the artifacts from this build project, for Type, choose No artifacts. (This is also the type to choose if you are just running tests or pushing a Docker image to Amazon ECR.) You also need an AWS CodeBuild service role so that AWS CodeBuild can interact with dependent AWS services on your behalf. Unless you already have a role, choose Create a role, and for Role name, type a name for your role.


In this example, leave the advanced settings at their defaults.

If you expand Show advanced settings, you’ll see options for customizing your build, including:

  • A build timeout.
  • A KMS key to encrypt all the artifacts that the builds for this project will use.
  • Options for building a Docker image.
  • Elevated permissions during your build action (for example, accessing Docker inside your build container to build a Dockerfile).
  • Resource options for the build compute type.
  • Environment variables (built-in or custom). For more information, see Create a Build Project in the AWS CodeBuild User Guide.

Advanced settings

You can use the AWS CodeBuild console to create a parameter in Amazon EC2 Systems Manager. Choose Create a parameter, and then follow the instructions in the dialog box. (In that dialog box, for KMS key, you can optionally specify the ARN of an AWS KMS key in your account. Amazon EC2 Systems Manager uses this key to encrypt the parameter’s value during storage and decrypt during retrieval.)

Create parameter

Choose Continue. On the Review page, either choose Save and build or choose Save to run the build later.

Choose Start build. When the build is complete, the Build logs section should display detailed information about the build.


To demonstrate a pull request, I will fork the repository as a different GitHub user, make commits to the forked repo, check in the changes to a newly created branch, and then open a pull request.

Pull request

As soon as the pull request is submitted, you’ll see CodeBuild start executing the build.


GitHub sends an HTTP POST payload to the webhook’s configured URL (highlighted here), which CodeBuild uses to download the latest source code and execute the build phases.

Build project

If you expand the Show all checks option for the GitHub pull request, you’ll see that CodeBuild has completed the build, all checks have passed, and a deep link is provided in Details, which opens the build history in the CodeBuild console.

Pull request


In this post, I showed you how to use GitHub as the source provider for your CodeStar projects and how to work with recent commits, issues, and pull requests in the CodeStar dashboard. I also showed you how you can use GitHub pull requests to automatically trigger a build in AWS CodeBuild — specifically, how this functionality makes it easier to collaborate across your team while editing and building your application code with CodeBuild.

About the author:

Balaji Iyer is an Enterprise Consultant for the Professional Services Team at Amazon Web Services. In this role, he has helped several customers successfully navigate their journey to AWS. His specialties include architecting and implementing highly scalable distributed systems, serverless architectures, large scale migrations, operational security, and leading strategic AWS initiatives. Before he joined Amazon, Balaji spent more than a decade building operating systems, big data analytics solutions, mobile services, and web applications. In his spare time, he enjoys experiencing the great outdoors and spending time with his family.


Kim Dotcom Plots Hollywood Execs’ Downfall in Wake of Weinstein Scandal

Post Syndicated from Andy original https://torrentfreak.com/kim-dotcom-plots-hollywood-execs-downfall-in-wake-of-weinstein-scandal-171011/

It has been nothing short of a disastrous week for movie mogul Harvey Weinstein.

Accused of sexual abuse and harassment by a string of actresses, the latest including Angelina Jolie and Gwyneth Paltrow, the 65-year-old is having his life taken apart.

This week, the influential producer was fired by his own The Weinstein Company, which is now seeking to change its name. And yesterday, following allegations of rape made in The New Yorker magazine, his wife, designer Georgina Chapman, announced she was leaving the Miramax co-founder.

“My heart breaks for all the women who have suffered tremendous pain because of these unforgivable actions,” the 41-year-old told People magazine.

As the scandal continues and more victims come forward, there are signs of a general emboldening of women in Hollywood, some of whom are publicly speaking out about their own experiences. If that continues to gain momentum – and the opportunity is certainly there – one man with his own experiences of Hollywood’s wrath wants to play a prominent role.

“Just the beginning. Sexual abuse and slavery by the Hollywood elites is as common as dirt. Tsunami,” Kim Dotcom wrote on Twitter.

Dotcom initially suggested that via a website, victims of Hollywood abuse could share their stories anonymously, shining light on a topic that is often shrouded in fear and secrecy. But soon the idea was growing legs.

“Looking for a Los Angeles law firm willing to represent hundreds of sexual abuse victims of Hollywood elites, pro-bono. I’ll find funding,” he said.

Within hours, Dotcom announced that he’d found lawyers in the US who are willing to help victims, for free.

“I had talks with Hollywood lawyers. Found a big law firm willing to represent sexual abuse victims, for free. Next, the website,” he teased.

It’s not hard to see why Dotcom is making this battle his own. Aside from any empathy he feels towards victims on a personal level, he sees his family as kindred spirits, people who have also felt the wrath of Hollywood executives.

That being said, the Megaupload founder is extremely clear that framing this as revenge or a personal vendetta would be not only wrong, but also disrespectful to the victims of abuse.

“I want to help victims because I’m a victim,” he told TorrentFreak.

“I’m an abuse victim of Hollywood, not sexual abuse, but certainly abuse of power. It’s time to shine some light on those Hollywood elites who think they are above the law and untouchable.”

Dotcom told NZ Herald that people like Harvey Weinstein rub shoulders with the great and the good, hoping to influence decision-makers for their own personal gain. It’s something Dotcom, his family, and his colleagues have felt the effects of.

“They dine with presidents, donate millions to powerful politicians and buy favors like tax breaks and new copyright legislation, even the Megaupload raid. They think they can destroy lives and businesses with impunity. They think they can get away with anything. But they can’t. We’ll teach them,” he warned.

The Megaupload founder says he has both “the motive and the resources” to help victims and he’s promising to do that with proven skills. Ironically, many of these have been honed as a direct result of Hollywood’s attack on Megaupload and Dotcom’s relentless drive to bounce back with new sites like Mega and his latest K.im / Bitcache project.

“I’m an experienced fundraiser. A high traffic crowdfunding campaign for this cause can raise millions. The costs won’t be an issue,” Dotcom informs TF. “There seems to be an appetite for these cases because defendants usually settle quickly. I have calls with LA firms today and tomorrow.

“Just the beginning. Watch me,” he concludes.

Source: TF, for the latest info on copyright, file-sharing, torrent sites and ANONYMOUS VPN services.

PureVPN Logs Helped FBI Net Alleged Cyberstalker

Post Syndicated from Andy original https://torrentfreak.com/purevpn-logs-helped-fbi-net-alleged-cyberstalker-171009/

Last Thursday, Ryan S. Lin, 24, of Newton, Massachusetts, was arrested on suspicion of conducting “an extensive cyberstalking campaign” against his former roommate, a 24-year-old Massachusetts woman, as well as her family members and friends.

According to the Department of Justice, Lin’s “multi-faceted campaign of computer hacking and cyberstalking” began in April 2016 when he began hacking into the victim’s online accounts, obtaining personal photographs, sensitive information about her medical and sexual histories, and other private details.

It’s alleged that after obtaining the above material, Lin distributed it to hundreds of others. It’s claimed he created fake online profiles showing the victim’s home address while soliciting sexual activity. This caused men to show up at her home.

“Mr. Lin allegedly carried out a relentless cyber stalking campaign against a young woman in a chilling effort to violate her privacy and threaten those around her,” said Acting United States Attorney William D. Weinreb.

“While using anonymizing services and other online tools to avoid attribution, Mr. Lin harassed the victim, her family, friends, co-workers and roommates, and then targeted local schools and institutions in her community. Mr. Lin will now face the consequences of his crimes.”

While Lin awaits his ultimate fate (he appeared in U.S. District Court in Boston Friday), the allegation he used anonymization tools to hide himself online but still managed to get caught raises a number of questions. An affidavit submitted by Special Agent Jeffrey Williams in support of the criminal complaint against Lin provides most of the answers.

Describing Lin’s actions against the victim as “doxing”, Williams begins by noting that while Lin was the initial aggressor, the fact he made the information so widely available raises the possibility that other people got involved with malicious acts later on. Nevertheless, Lin remains the investigation’s prime suspect.

According to the affidavit, Lin is computer savvy having majored in computer science. He allegedly utilized a number of methods to hide his identity and IP address, including TOR, Virtual Private Network (VPN) services and email providers that “do not maintain logs or other records.”

But if that genuinely is the case, how was Lin caught?

First up, it’s worth noting that plenty of Lin’s aggressive and stalking behaviors towards the victim were demonstrated in a physical sense, offline. In that respect, it appears the authorities already had him as the prime suspect and worked back from there.

In one instance, the FBI examined a computer that had been used by Lin at a former workplace. Although Windows had been reinstalled, the FBI managed to find Google Chrome data which indicated Lin had viewed articles about bomb threats he allegedly made. They were also able to determine he’d accessed the victim’s Gmail account and additional data suggested that he’d used a VPN service.

“Artifacts indicated that PureVPN, a VPN service that was used repeatedly in the cyberstalking scheme, was installed on the computer,” the affidavit reads.

From here the Special Agent’s report reveals that the FBI received cooperation from Hong Kong-based PureVPN.

“Significantly, PureVPN was able to determine that their service was accessed by the same customer from two originating IP addresses: the RCN IP address from the home Lin was living in at the time, and the software company where Lin was employed at the time,” the agent’s affidavit reads.

Needless to say, while this information will prove useful to the FBI’s prosecution of Lin, it’s also likely to turn into a huge headache for the VPN provider. The company claims zero-logging, which clearly isn’t the case.

“PureVPN operates a self-managed VPN network that currently stands at 750+ Servers in 141 Countries. But is this enough to ensure complete security?” the company’s marketing statement reads.

“That’s why PureVPN has launched advanced features to add proactive, preventive and complete security. There are no third-parties involved and NO logs of your activities.”

PureVPN privacy graphic

However, if one drills down into the PureVPN privacy policy proper, one sees the following:

Our servers automatically record the time at which you connect to any of our servers. From here on forward, we do not keep any records of anything that could associate any specific activity to a specific user. The time when a successful connection is made with our servers is counted as a ‘connection’ and the total bandwidth used during this connection is called ‘bandwidth’. Connection and bandwidth are kept in record to maintain the quality of our service. This helps us understand the flow of traffic to specific servers so we could optimize them better.

This seems to match what the FBI says – almost. While it says it doesn’t log, PureVPN admits to keeping records of when a user connects to the service and for how long. The FBI clearly states that the service also captures the user’s IP address too. In fact, it appears that PureVPN also logged the IP address belonging to another VPN service (WANSecurity) that was allegedly used by Lin to connect to PureVPN.

That record also helped to complete another circle of evidence. IP addresses used by
Kansas-based WANSecurity and Secure Internet LLC (servers operated by PureVPN) were allegedly used to access Gmail accounts known to be under Lin’s control.

Somewhat ironically, this summer Lin took to Twitter to criticize VPN provider IPVanish (which is not involved in the case) over its no-logging claims.

“There is no such thing as a VPN that doesn’t keep logs,” Lin said. “If they can limit your connections or track bandwidth usage, they keep logs.”

Or, in the case of PureVPN, if they log a connection time and a source IP address, that could be enough to raise the suspicions of the FBI and boost what already appears to be a pretty strong case.

If convicted, Lin faces up to five years in prison and three years of supervised release.

Source: TF, for the latest info on copyright, file-sharing, torrent sites and ANONYMOUS VPN services.

JavaScript got better while I wasn’t looking

Post Syndicated from Eevee original https://eev.ee/blog/2017/10/07/javascript-got-better-while-i-wasnt-looking/

IndustrialRobot has generously donated in order to inquire:

In the last few years there seems to have been a lot of activity with adding emojis to Unicode. Has there been an equal effort to add ‘real’ languages/glyph systems/etc?

And as always, if you don’t have anything to say on that topic, feel free to choose your own. :p


I mean, each release of Unicode lists major new additions right at the top — Unicode 10, Unicode 9, Unicode 8, etc. They also keep fastidious notes, so you can also dig into how and why these new scripts came from, by reading e.g. the proposal for the addition of Zanabazar Square. I don’t think I have much to add here; I’m not a real linguist, I only play one on TV.

So with that out of the way, here’s something completely different!

A brief history of JavaScript

JavaScript was created in seven days, about eight thousand years ago. It was pretty rough, and it stayed rough for most of its life. But that was fine, because no one used it for anything besides having a trail of sparkles follow your mouse on their Xanga profile.

Then people discovered you could actually do a handful of useful things with JavaScript, and it saw a sharp uptick in usage. Alas, it stayed pretty rough. So we came up with polyfills and jQuerys and all kinds of miscellaneous things that tried to smooth over the rough parts, to varying degrees of success.

And… that’s it. That’s pretty much how things stayed for a while.

I have complicated feelings about JavaScript. I don’t hate it… but I certainly don’t enjoy it, either. It has some pretty neat ideas, like prototypical inheritance and “everything is a value”, but it buries them under a pile of annoying quirks and a woefully inadequate standard library. The DOM APIs don’t make things much better — they seem to be designed as though the target language were Java, rarely taking advantage of any interesting JavaScript features. And the places where the APIs overlap with the language are a hilarious mess: I have to check documentation every single time I use any API that returns a set of things, because there are at least three totally different conventions for handling that and I can’t keep them straight.

The funny thing is that I’ve been fairly happy to work with Lua, even though it shares most of the same obvious quirks as JavaScript. Both languages are weakly typed; both treat nonexistent variables and keys as simply false values, rather than errors; both have a single data structure that doubles as both a list and a map; both use 64-bit floating-point as their only numeric type (though Lua added integers very recently); both lack a standard object model; both have very tiny standard libraries. Hell, Lua doesn’t even have exceptions, not really — you have to fake them in much the same style as Perl.

And yet none of this bothers me nearly as much in Lua. The differences between the languages are very subtle, but combined they make a huge impact.

  • Lua has separate operators for addition and concatenation, so + is never ambiguous. It also has printf-style string formatting in the standard library.

  • Lua’s method calls are syntactic sugar: foo:bar() just means foo.bar(foo). Lua doesn’t even have a special this or self value; the invocant just becomes the first argument. In contrast, JavaScript invokes some hand-waved magic to set its contextual this variable, which has led to no end of confusion.

  • Lua has an iteration protocol, as well as built-in iterators for dealing with list-style or map-style data. JavaScript has a special dedicated Array type and clumsy built-in iteration syntax.

  • Lua has operator overloading and (surprisingly flexible) module importing.

  • Lua allows the keys of a map to be any value (though non-scalars are always compared by identity). JavaScript implicitly converts keys to strings — and since there’s no operator overloading, there’s no way to natively fix this.

These are fairly minor differences, in the grand scheme of language design. And almost every feature in Lua is implemented in a ridiculously simple way; in fact the entire language is described in complete detail in a single web page. So writing JavaScript is always frustrating for me: the language is so close to being much more ergonomic, and yet, it isn’t.

Or, so I thought. As it turns out, while I’ve been off doing other stuff for a few years, browser vendors have been implementing all this pie-in-the-sky stuff from “ES5” and “ES6”, whatever those are. People even upgrade their browsers now. Lo and behold, the last time I went to write JavaScript, I found out that a number of papercuts had actually been solved, and the solutions were sufficiently widely available that I could actually use them in web code.

The weird thing is that I do hear a lot about JavaScript, but the feature I’ve seen raved the most about by far is probably… built-in types for working with arrays of bytes? That’s cool and all, but not exactly the most pressing concern for me.

Anyway, if you also haven’t been keeping tabs on the world of JavaScript, here are some things we missed.


MDN docs — supported in Firefox 44, Chrome 41, IE 11, Safari 10

I’m pretty sure I first saw let over a decade ago. Firefox has supported it for ages, but you actually had to opt in by specifying JavaScript version 1.7. Remember JavaScript versions? You know, from back in the days when people actually suggested you write stuff like this:

<SCRIPT LANGUAGE="JavaScript1.2" TYPE="text/javascript">


Anyway, so, let declares a variable — but scoped to the immediately containing block, unlike var, which scopes to the innermost function. The trouble with var was that it was very easy to make misleading:

// foo exists here
while (true) {
    var foo = ...;
// foo exists here too

If you reused the same temporary variable name in a different block, or if you expected to be shadowing an outer foo, or if you were trying to do something with creating closures in a loop, this would cause you some trouble.

But no more, because let actually scopes the way it looks like it should, the way variable declarations do in C and friends. As an added bonus, if you refer to a variable declared with let outside of where it’s valid, you’ll get a ReferenceError instead of a silent undefined value. Hooray!

There’s one other interesting quirk to let that I can’t find explicitly documented. Consider:

let closures = [];
for (let i = 0; i < 4; i++) {
    closures.push(function() { console.log(i); });
for (let j = 0; j < closures.length; j++) {

If this code had used var i, then it would print 4 four times, because the function-scoped var i means each closure is sharing the same i, whose final value is 4. With let, the output is 0 1 2 3, as you might expect, because each run through the loop gets its own i.

But wait, hang on.

The semantics of a C-style for are that the first expression is only evaluated once, at the very beginning. So there’s only one let i. In fact, it makes no sense for each run through the loop to have a distinct i, because the whole idea of the loop is to modify i each time with i++.

I assume this is simply a special case, since it’s what everyone expects. We expect it so much that I can’t find anyone pointing out that the usual explanation for why it works makes no sense. It has the interesting side effect that for no longer de-sugars perfectly to a while, since this will print all 4s:

closures = [];
let i = 0;
while (i < 4) {
    closures.push(function() { console.log(i); });
for (let j = 0; j < closures.length; j++) {

This isn’t a problem — I’m glad let works this way! — it just stands out to me as interesting. Lua doesn’t need a special case here, since it uses an iterator protocol that produces values rather than mutating a visible state variable, so there’s no problem with having the loop variable be truly distinct on each run through the loop.


MDN docs — supported in Firefox 45, Chrome 42, Safari 9, Edge 13

Prototypical inheritance is pretty cool. The way JavaScript presents it is a little bit opaque, unfortunately, which seems to confuse a lot of people. JavaScript gives you enough functionality to make it work, and even makes it sound like a first-class feature with a property outright called prototype… but to actually use it, you have to do a bunch of weird stuff that doesn’t much look like constructing an object or type.

The funny thing is, people with almost any background get along with Python just fine, and Python uses prototypical inheritance! Nobody ever seems to notice this, because Python tucks it neatly behind a class block that works enough like a Java-style class. (Python also handles inheritance without using the prototype, so it’s a little different… but I digress. Maybe in another post.)

The point is, there’s nothing fundamentally wrong with how JavaScript handles objects; the ergonomics are just terrible.

Lo! They finally added a class keyword. Or, rather, they finally made the class keyword do something; it’s been reserved this entire time.

class Vector {
    constructor(x, y) {
        this.x = x;
        this.y = y;

    get magnitude() {
        return Math.sqrt(this.x * this.x + this.y * this.y);

    dot(other) {
        return this.x * other.x + this.y * other.y;

This is all just sugar for existing features: creating a Vector function to act as the constructor, assigning a function to Vector.prototype.dot, and whatever it is you do to make a property. (Oh, there are properties. I’ll get to that in a bit.)

The class block can be used as an expression, with or without a name. It also supports prototypical inheritance with an extends clause and has a super pseudo-value for superclass calls.

It’s a little weird that the inside of the class block has its own special syntax, with function omitted and whatnot, but honestly you’d have a hard time making a class block without special syntax.

One severe omission here is that you can’t declare values inside the block, i.e. you can’t just drop a bar = 3; in there if you want all your objects to share a default attribute. The workaround is to just do this.bar = 3; inside the constructor, but I find that unsatisfying, since it defeats half the point of using prototypes.


MDN docs — supported in Firefox 4, Chrome 5, IE 9, Safari 5.1

JavaScript historically didn’t have a way to intercept attribute access, which is a travesty. And by “intercept attribute access”, I mean that you couldn’t design a value foo such that evaluating foo.bar runs some code you wrote.

Exciting news: now it does. Or, rather, you can intercept specific attributes, like in the class example above. The above magnitude definition is equivalent to:

Object.defineProperty(Vector.prototype, 'magnitude', {
    configurable: true,
    enumerable: true,
    get: function() {
        return Math.sqrt(this.x * this.x + this.y * this.y);


And what even are these configurable and enumerable things? It seems that every single key on every single object now has its own set of three Boolean twiddles:

  • configurable means the property itself can be reconfigured with another call to Object.defineProperty.
  • enumerable means the property appears in for..in or Object.keys().
  • writable means the property value can be changed, which only applies to properties with real values rather than accessor functions.

The incredibly wild thing is that for properties defined by Object.defineProperty, configurable and enumerable default to false, meaning that by default accessor properties are immutable and invisible. Super weird.

Nice to have, though. And luckily, it turns out the same syntax as in class also works in object literals.

Vector.prototype = {
    get magnitude() {
        return Math.sqrt(this.x * this.x + this.y * this.y);

Alas, I’m not aware of a way to intercept arbitrary attribute access.

Another feature along the same lines is Object.seal(), which marks all of an object’s properties as non-configurable and prevents any new properties from being added to the object. The object is still mutable, but its “shape” can’t be changed. And of course you can just make the object completely immutable if you want, via setting all its properties non-writable, or just using Object.freeze().

I have mixed feelings about the ability to irrevocably change something about a dynamic runtime. It would certainly solve some gripes of former Haskell-minded colleagues, and I don’t have any compelling argument against it, but it feels like it violates some unwritten contract about dynamic languages — surely any structural change made by user code should also be able to be undone by user code?

Slurpy arguments

MDN docs — supported in Firefox 15, Chrome 47, Edge 12, Safari 10

Officially this feature is called “rest parameters”, but that’s a terrible name, no one cares about “arguments” vs “parameters”, and “slurpy” is a good word. Bless you, Perl.

function foo(a, b, ...args) {
    // ...

Now you can call foo with as many arguments as you want, and every argument after the second will be collected in args as a regular array.

You can also do the reverse with the spread operator:

let args = [];

It even works in array literals, even multiple times:

let args2 = [...args, ...args];
console.log(args2);  // [1, 2, 3, 1, 2, 3]

Apparently there’s also a proposal for allowing the same thing with objects inside object literals.

Default arguments

MDN docs — supported in Firefox 15, Chrome 49, Edge 14, Safari 10

Yes, arguments can have defaults now. It’s more like Sass than Python — default expressions are evaluated once per call, and later default expressions can refer to earlier arguments. I don’t know how I feel about that but whatever.

function foo(n = 1, m = n + 1, list = []) {

Also, unlike Python, you can have an argument with a default and follow it with an argument without a default, since the default default (!) is and always has been defined as undefined. Er, let me just write it out.

function bar(a = 5, b) {

Arrow functions

MDN docs — supported in Firefox 22, Chrome 45, Edge 12, Safari 10

Perhaps the most humble improvement is the arrow function. It’s a slightly shorter way to write an anonymous function.

(a, b, c) => { ... }
a => { ... }
() => { ... }

An arrow function does not set this or some other magical values, so you can safely use an arrow function as a quick closure inside a method without having to rebind this. Hooray!

Otherwise, arrow functions act pretty much like regular functions; you can even use all the features of regular function signatures.

Arrow functions are particularly nice in combination with all the combinator-style array functions that were added a while ago, like Array.forEach.

[7, 8, 9].forEach(value => {


MDN docs — supported in Firefox 36, Chrome 38, Edge 12, Safari 9

This isn’t quite what I’d call an exciting feature, but it’s necessary for explaining the next one. It’s actually… extremely weird.

symbol is a new kind of primitive (like number and string), not an object (like, er, Number and String). A symbol is created with Symbol('foo'). No, not new Symbol('foo'); that throws a TypeError, for, uh, some reason.

The only point of a symbol is as a unique key. You see, symbols have one very special property: they can be used as object keys, and will not be stringified. Remember, only strings can be keys in JavaScript — even the indices of an array are, semantically speaking, still strings. Symbols are a new exception to this rule.

Also, like other objects, two symbols don’t compare equal to each other: Symbol('foo') != Symbol('foo').

The result is that symbols solve one of the problems that plauges most object systems, something I’ve talked about before: interfaces. Since an interface might be implemented by any arbitrary type, and any arbitrary type might want to implement any number of arbitrary interfaces, all the method names on an interface are effectively part of a single global namespace.

I think I need to take a moment to justify that. If you have IFoo and IBar, both with a method called method, and you want to implement both on the same type… you have a problem. Because most object systems consider “interface” to mean “I have a method called method, with no way to say which interface’s method you mean. This is a hard problem to avoid, because IFoo and IBar might not even come from the same library. Occasionally languages offer a clumsy way to “rename” one method or the other, but the most common approach seems to be for interface designers to avoid names that sound “too common”. You end up with redundant mouthfuls like IFoo.foo_method.

This incredibly sucks, and the only languages I’m aware of that avoid the problem are the ML family and Rust. In Rust, you define all the methods for a particular trait (interface) in a separate block, away from the type’s “own” methods. It’s pretty slick. You can still do obj.method(), and as long as there’s only one method among all the available traits, you’ll get that one. If not, there’s syntax for explicitly saying which trait you mean, which I can’t remember because I’ve never had to use it.

Symbols are JavaScript’s answer to this problem. If you want to define some interface, you can name its methods with symbols, which are guaranteed to be unique. You just have to make sure you keep the symbol around somewhere accessible so other people can actually use it. (Or… not?)

The interesting thing is that JavaScript now has several of its own symbols built in, allowing user objects to implement features that were previously reserved for built-in types. For example, you can use the Symbol.hasInstance symbol — which is simply where the language is storing an existing symbol and is not the same as Symbol('hasInstance')! — to override instanceof:

// oh my god don't do this though
class EvenNumber {
    static [Symbol.hasInstance](obj) {
        return obj % 2 == 0;
console.log(2 instanceof EvenNumber);  // true
console.log(3 instanceof EvenNumber);  // false

Oh, and those brackets around Symbol.hasInstance are a sort of reverse-quoting — they indicate an expression to use where the language would normally expect a literal identifier. I think they work as object keys, too, and maybe some other places.

The equivalent in Python is to implement a method called __instancecheck__, a name which is not special in any way except that Python has reserved all method names of the form __foo__. That’s great for Python, but doesn’t really help user code. JavaScript has actually outclassed (ho ho) Python here.

Of course, obj[BobNamespace.some_method]() is not the prettiest way to call an interface method, so it’s not perfect. I imagine this would be best implemented in user code by exposing a polymorphic function, similar to how Python’s len(obj) pretty much just calls obj.__len__().

I only bring this up because it’s the plumbing behind one of the most incredible things in JavaScript that I didn’t even know about until I started writing this post. I’m so excited oh my gosh. Are you ready? It’s:

Iteration protocol

MDN docs — supported in Firefox 27, Chrome 39, Safari 10; still experimental in Edge

Yes! Amazing! JavaScript has first-class support for iteration! I can’t even believe this.

It works pretty much how you’d expect, or at least, how I’d expect. You give your object a method called Symbol.iterator, and that returns an iterator.

What’s an iterator? It’s an object with a next() method that returns the next value and whether the iterator is exhausted.

Wait, wait, wait a second. Hang on. The method is called next? Really? You didn’t go for Symbol.next? Python 2 did exactly the same thing, then realized its mistake and changed it to __next__ in Python 3. Why did you do this?

Well, anyway. My go-to test of an iterator protocol is how hard it is to write an equivalent to Python’s enumerate(), which takes a list and iterates over its values and their indices. In Python it looks like this:

for i, value in enumerate(['one', 'two', 'three']):
    print(i, value)
# 0 one
# 1 two
# 2 three

It’s super nice to have, and I’m always amazed when languages with “strong” “support” for iteration don’t have it. Like, C# doesn’t. So if you want to iterate over a list but also need indices, you need to fall back to a C-style for loop. And if you want to iterate over a lazy or arbitrary iterable but also need indices, you need to track it yourself with a counter. Ridiculous.

Here’s my attempt at building it in JavaScript.

function enumerate(iterable) {
    // Return a new iter*able* object with a Symbol.iterator method that
    // returns an iterator.
    return {
        [Symbol.iterator]: function() {
            let iterator = iterable[Symbol.iterator]();
            let i = 0;

            return {
                next: function() {
                    let nextval = iterator.next();
                    if (! nextval.done) {
                        nextval.value = [i, nextval.value];
                    return nextval;
for (let [i, value] of enumerate(['one', 'two', 'three'])) {
    console.log(i, value);
// 0 one
// 1 two
// 2 three

Incidentally, for..of (which iterates over a sequence, unlike for..in which iterates over keys — obviously) is finally supported in Edge 12. Hallelujah.

Oh, and let [i, value] is destructuring assignment, which is also a thing now and works with objects as well. You can even use the splat operator with it! Like Python! (And you can use it in function signatures! Like Python! Wait, no, Python decided that was terrible and removed it in 3…)

let [x, y, ...others] = ['apple', 'orange', 'cherry', 'banana'];

It’s a Halloween miracle. 🎃


MDN docs — supported in Firefox 26, Chrome 39, Edge 13, Safari 10

That’s right, JavaScript has goddamn generators now. It’s basically just copying Python and adding a lot of superfluous punctuation everywhere. Not that I’m complaining.

Also, generators are themselves iterable, so I’m going to cut to the chase and rewrite my enumerate() with a generator.

function enumerate(iterable) {
    return {
        [Symbol.iterator]: function*() {
            let i = 0;
            for (let value of iterable) {
                yield [i, value];
for (let [i, value] of enumerate(['one', 'two', 'three'])) {
    console.log(i, value);
// 0 one
// 1 two
// 2 three

Amazing. function* is a pretty strange choice of syntax, but whatever? I guess it also lets them make yield only act as a keyword inside a generator, for ultimate backwards compatibility.

JavaScript generators support everything Python generators do: yield* yields every item from a subsequence, like Python’s yield from; generators can return final values; you can pass values back into the generator if you iterate it by hand. No, really, I wasn’t kidding, it’s basically just copying Python. It’s great. You could now built asyncio in JavaScript!

In fact, they did that! JavaScript now has async and await. An async function returns a Promise, which is also a built-in type now. Amazing.

Sets and maps

MDN docs for MapMDN docs for Set — supported in Firefox 13, Chrome 38, IE 11, Safari 7.1

I did not save the best for last. This is much less exciting than generators. But still exciting.

The only data structure in JavaScript is the object, a map where the strings are keys. (Or now, also symbols, I guess.) That means you can’t readily use custom values as keys, nor simulate a set of arbitrary objects. And you have to worry about people mucking with Object.prototype, yikes.

But now, there’s Map and Set! Wow.

Unfortunately, because JavaScript, Map couldn’t use the indexing operators without losing the ability to have methods, so you have to use a boring old method-based API. But Map has convenient methods that plain objects don’t, like entries() to iterate over pairs of keys and values. In fact, you can use a map with for..of to get key/value pairs. So that’s nice.

Perhaps more interesting, there’s also now a WeakMap and WeakSet, where the keys are weak references. I don’t think JavaScript had any way to do weak references before this, so that’s pretty slick. There’s no obvious way to hold a weak value, but I guess you could substitute a WeakSet with only one item.

Template literals

MDN docs — supported in Firefox 34, Chrome 41, Edge 12, Safari 9

Template literals are JavaScript’s answer to string interpolation, which has historically been a huge pain in the ass because it doesn’t even have string formatting in the standard library.

They’re just strings delimited by backticks instead of quotes. They can span multiple lines and contain expressions.

console.log(`one plus
two is ${1 + 2}`);

Someone decided it would be a good idea to allow nesting more sets of backticks inside a ${} expression, so, good luck to syntax highlighters.

However, someone also had the most incredible idea ever, which was to add syntax allowing user code to do the interpolation — so you can do custom escaping, when absolutely necessary, which is virtually never, because “escaping” means you’re building a structured format by slopping strings together willy-nilly instead of using some API that works with the structure.

function html(literals, ...values) {
    let ret = [];
    literals.forEach((literal, i) => {
        if (i > 0) {
            // Is there seriously still not a built-in function for doing this?
            // Well, probably because you SHOULDN'T BE DOING IT
            ret.push(values[i - 1]
                .replace(/&/g, '&amp;')
                .replace(/</g, '&lt;')
                .replace(/>/g, '&gt;')
                .replace(/"/g, '&quot;')
                .replace(/'/g, '&apos;'));
    return ret.join('');
let username = 'Bob<script>';
let result = html`<b>Hello, ${username}!</b>`;
// <b>Hello, Bob&lt;script&gt;!</b>

It’s a shame this feature is in JavaScript, the language where you are least likely to need it.

Trailing commas

Remember how you couldn’t do this for ages, because ass-old IE considered it a syntax error and would reject the entire script?

    a: 'one',
    b: 'two',
    c: 'three',  // <- THIS GUY RIGHT HERE

Well now it’s part of the goddamn spec and if there’s anything in this post you can rely on, it’s this. In fact you can use AS MANY GODDAMN TRAILING COMMAS AS YOU WANT. But only in arrays.

[1, 2, 3,,,,,,,,,,,,,,,,,,,,,,,,,]

Apparently that has the bizarre side effect of reserving extra space at the end of the array, without putting values there.

And more, probably

Like strict mode, which makes a few silent “errors” be actual errors, forces you to declare variables (no implicit globals!), and forbids the completely bozotic with block.

Or String.trim(), which trims whitespace off of strings.

Or… Math.sign()? That’s new? Seriously? Well, okay.

Or the Proxy type, which lets you customize indexing and assignment and calling. Oh. I guess that is possible, though this is a pretty weird way to do it; why not just use symbol-named methods?

You can write Unicode escapes for astral plane characters in strings (or identifiers!), as \u{XXXXXXXX}.

There’s a const now? I extremely don’t care, just name it in all caps and don’t reassign it, come on.

There’s also a mountain of other minor things, which you can peruse at your leisure via MDN or the ECMAScript compatibility tables (note the links at the top, too).

That’s all I’ve got. I still wouldn’t say I’m a big fan of JavaScript, but it’s definitely making an effort to clean up some goofy inconsistencies and solve common problems. I think I could even write some without yelling on Twitter about it now.

On the other hand, if you’re still stuck supporting IE 10 for some reason… well, er, my condolences.

Raspberry Coulis’s night vision camera

Post Syndicated from Rachel Churcher original https://www.raspberrypi.org/blog/night-vision-camera/

We’ve all been there: zombies at the door, Daleks on the driveway, creatures from the Upside Down in the walls. You want to be able to monitor their movements, but how do you do that without attracting their attention? Wesley Archer (AKA Raspberry Coulis) has the answer: a Pi-powered Night Vision Camera, perfect for catching unearthly creatures on the prowl — and for wildlife spotting, birdwatching, and home security too, I guess…

Wesley's Pi-powered Night Vision Camera

Wesley’s Pi-powered Night Vision Camera

Black box

To build his Night Vision Camera, Wesley ordered an infrared Lisiparoi LED Light Ring, a Cyntech Raspberry Pi case, a Pi NoIR Camera Module, and USB WiFi adapter. He based this project around a Raspberry Pi Model B that was in need of a good home, but there’s no reason why you couldn’t use a more up-to-date model with built-in wireless networking.

Wesley set about adapting the case to hold the camera and the infrared light, cutting a hole for the camera lens and attaching the LED Light Ring to the front of the box. The light ring acts as an infrared floodlight, invisible to the zombies, but when combined with the NoIR Camera Module, capable of capturing useful images in the dark. His blog includes helpful step-by-step instructions for this process — unlike Wesley, we recommend using a metal file from your toolbox, and not a nail file belonging to another member of your family. If you’re surrounded by the undead hordes, domestic harmony is essential.

Cyntech Raspberry Pi case with a hole for the Camera Module - Night Vision Camera

Cutting holes in the case. Please don’t use your spouse’s nail file for this job …


When your hardware is complete, it’s time to install the software. Wesley chose MotionEyeOS to run his camera, and his blog explains the process of downloading and installing the software on your Pi. When everything is set up, and the Pi is connected to your WiFi network, all you need is the Pi’s IP address to view the feed from the camera. Type the IP address into a browser on the same WiFi network, log in, and you’ll soon be spotting intruders (supernatural or otherwise), or possibly watching the fluffy residents of your bird box. Whatever makes you happy.


While a camera with night vision is obviously useful, both the Lisiparoi Light Ring and the Camera Module are available for use with the visible spectrum. You can order the Light Ring with infrared or standard white LEDs, and the standard Camera Module works with visible light. If you don’t mind attracting attention, both options could be used to monitor your perimeter for threats.

Saving the world

We think this project would be an amazing inspiration for the current Pioneers challenge, Only you can save us! Set up a camera to control entry to your secret bunker, even when the lights fail. Fend off attacks from zombies, Daleks, or giant spiders, and help save humanity from catastrophe!

Cute knitted zombies dancing - Night Vision Camera

Save yourselves!

Have you built a security system to keep your property safe from marauding zombies? Or even from regular burglars? Has your Pioneers team used infrared monitoring in your build? Tell us about it in the comments!

The post Raspberry Coulis’s night vision camera appeared first on Raspberry Pi.

Microcell through a mobile hotspot

Post Syndicated from Robert Graham original http://blog.erratasec.com/2017/10/microcell-through-mobile-hotspot.html

I accidentally acquired a tree farm 20 minutes outside of town. For utilities, it gets electricity and basic phone. It doesn’t get water, sewer, cable, or DSL (i.e. no Internet). Also, it doesn’t really get cell phone service. While you can get SMS messages up there, you usually can’t get a call connected, or hold a conversation if it does.

We have found a solution — an evil solution. We connect an AT&T “Microcell“, which provides home cell phone service through your Internet connection, to an AT&T Mobile Hotspot, which provides an Internet connection through your cell phone service.

Now, you may be laughing at this, because it’s a circular connection. It’s like trying to make a sailboat go by blowing on the sails, or lifting up a barrel to lighten the load in the boat.

But it actually works.

Since we get some, but not enough, cellular signal, we setup a mast 20 feet high with a directional antenna pointed to the cell tower 7.5 miles to the southwest, connected to a signal amplifier. It’s still an imperfect solution, as we are still getting terrain distortions in the signal, but it provides a good enough signal-to-noise ratio to get a solid connection.

We then connect that directional antenna directly to a high-end Mobile Hotspot. This gives us a solid 2mbps connection with a latency under 30milliseconds. This is far lower than the 50mbps you can get right next to a 4G/LTE tower, but it’s still pretty good for our purposes.

We then connect the AT&T Microcell to the Mobile Hotspot, via WiFi.

To avoid the circular connection, we lock the frequencies for the Mobile Hotspot to 4G/LTE, and to 3G for the Microcell. This prevents the Mobile Hotspot locking onto the strong 3G signal from the Microcell. It also prevents the two from causing noise to the other.

This works really great. We now get a strong cell signal on our phones even 400 feet from the house through some trees. We can be all over the property, out in the lake, down by the garden, and so on, and have our phones work as normal. It’s only AT&T, but that’s what the whole family uses.

You might be asking why we didn’t just use a normal signal amplifier, like they use on corporate campus. It boosts all the analog frequencies, making any cell phone service works.

We’ve tried this, and it works a bit, allowing cell phones to work inside the house pretty well. But they don’t work outside the house, which is where we spend a lot of time. In addition, while our newer phones work, my sister’s iPhone 5 doesn’t. We have no idea what’s going on. Presumably, we could hire professional installers and stuff to get everything working, but nobody would quote us a price lower than $25,000 to even come look at the property.

Another possible solution is satellite Internet. There are two satellites in orbit that cover the United States with small “spot beams” delivering high-speed service (25mbps downloads). However, the latency is 500milliseconds, which makes it impractical for low-latency applications like phone calls.

While I know a lot about the technology in theory, I find myself hopelessly clueless in practice. I’ve been playing with SDR (“software defined radio”) to try to figure out exactly where to locate and point the directional antenna, but I’m not sure I’ve come up with anything useful. In casual tests, it seems rotating the antenna from vertical to horizontal increases the signal-to-noise ratio a bit, which seems counter intuitive, and should not happen. So I’m completely lost.

Anyway, I thought I’d write this up as a blogpost, in case anybody has better suggestion. Or, instead of signals, suggestions to get wired connectivity. Properties a half mile away get DSL, I wish I knew who to talk to at the local phone company to pay them money to extend Internet to our property.

Phone works in all this area now

Announcing the 2017-18 European Astro Pi challenge!

Post Syndicated from David Honess original https://www.raspberrypi.org/blog/announcing-2017-18-astro-pi/

Astro Pi is back! Today we’re excited to announce the 2017-18 European Astro Pi challenge in partnership with the European Space Agency (ESA). We are searching for the next generation of space scientists.


Enjoy the videos and music you love, upload original content, and share it all with friends, family, and the world on YouTube.

Astro Pi is an annual science and coding competition where student-written code is run on the International Space Station under the oversight of an ESA astronaut. The challenge is open to students from all 22 ESA member countries, including — for the first time — associate members Canada and Slovenia.

The format of the competition is changing slightly this year, and we also have a brand-new non-competitive mission in which participants are guaranteed to have their code run on the ISS for 30 seconds!

Mission Zero

Until now, students have worked on Astro Pi projects in an extra-curricular context and over multiple sessions. For teachers and students who don’t have much spare capacity, we wanted to provide an accessible activity that teams can complete in just one session.

So we came up with Mission Zero for young people no older than 14. To complete it, form a team of two to four people and use our step-by-step guide to help you write a simple Python program that shows your personal message and the ambient temperature on the Astro Pi. If you adhere to a few rules, your code is guaranteed to run in space for 30 seconds, and you’ll receive a certificate showing the exact time period during which your code has run in space. No special hardware is needed for this mission, since everything is done in a web browser.

Mission Zero is open until 26 November 2017! Find out more.

Mission Space Lab

Students aged up to 19 can take part in Mission Space Lab. Form a team of two to six people, and work like real space scientists to design your own experiment. Receive free kit to work with, and write the Python code to carry out your experiment.

There are two themes for Mission Space Lab teams to choose from for their projects:

  • Life in space
    You will make use of Astro Pi Vis (“Ed”) in the European Columbus module. You can use all of its sensors, but you cannot record images or videos.
  • Life on Earth
    You will make use of Astro Pi IR (“Izzy”), which will be aimed towards the Earth through a window. You can use all of its sensors and its camera.

The Astro Pi kit, delivered to Space Lab teams by ESA

If you achieve flight status, your code will be uploaded to the ISS and run for three hours (two orbits). All the data that your code records in space will be downloaded and returned to you for analysis. Then submit a short report on your findings to be in with a chance to win exclusive, money-can’t-buy prizes! You can also submit your project for a Bronze CREST Award.

Mission Space Lab registration is open until 29 October 2017, and accepted teams will continue to spring 2018. Find out more.

How do I get started?

There are loads of materials available that will help you begin your Astro Pi journey — check out the Getting started with the Sense HAT resource and this video explaining how to build the flight case.


If you have any questions, please post them in the comments below. We’re standing by to answer them!

The post Announcing the 2017-18 European Astro Pi challenge! appeared first on Raspberry Pi.

Friday Squid Blogging: Another Giant Squid Caught off the Coast of Kerry

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2017/09/friday_squid_bl_592.html

The Flannery family have caught four giant squid, two this year.

As usual, you can also use this squid post to talk about the security stories in the news that I haven’t covered.

Read my blog posting guidelines here.

Surviving Your First Year

Post Syndicated from Gleb Budman original https://www.backblaze.com/blog/startup-stages-surviving-your-first-year/

Surviving Your First Year

This post by Backblaze’s CEO and co-founder Gleb Budman is the fifth in a series about entrepreneurship. You can choose posts in the series from the list below:

  1. How Backblaze got Started: The Problem, The Solution, and the Stuff In-Between
  2. Building a Competitive Moat: Turning Challenges Into Advantages
  3. From Idea to Launch: Getting Your First Customers
  4. How to Get Your First 1,000 Customers
  5. Surviving Your First Year

Use the Join button above to receive notification of new posts in this series.

In my previous posts, I talked about coming up with an idea, determining the solution, and getting your first customers. But you’re building a company, not a product. Let’s talk about what the first year should look like.

The primary goals for that first year are to: 1) set up the company; 2) build, launch, and learn; and 3) survive.

Setting Up the Company

The company you’re building is more than the product itself, and you’re not going to do it alone. You don’t want to spend too much time on this since getting customers is key, but if you don’t set up the basics, there are all sorts of issues down the line.

startup idea board

Find Your Co-Founders & Determine Roles

You may already have the idea, but who do you need to execute it? At Backblaze, we needed people to build the web experience, the client backup application, and the server/storage side. We also needed someone to handle the business/marketing aspects, and we felt that the design and user experience were critical. As a result, we started with five co-founders: three engineers, a designer, and me for the business and marketing.

Of course not every role needs to be filled by a co-founder. You can hire employees for positions as well. But think through the strategic skills you’ll need to launch and consider co-founders with those skill sets.

Too many people think they can just “work together” on everything. Don’t. Determine roles as quickly as possible so that it’s clear who is responsible for what work and which decisions. We were lucky in that we had worked together and thus knew what each person would do, but even so we assigned titles early on to clarify roles.

Takeaway:   Fill critical roles and explicitly split roles and responsibilities.

Get Your Legal Basics In Place

When we’re excited about building a product, legal basics are often the last thing we want to deal with. You don’t need to go overboard, but it’s critical to get certain things done.

  1. Determine ownership split. What is the percentage breakdown of the company that each of the founders will own? It can be a tough discussion, but it only becomes more difficult later when there is more value and people have put more time into it. At Backblaze we split the equity equally five ways. This is uncommon. The benefit of this is that all the founders feel valued and “in it together.” The benefit of the more common split where someone has a dominant share is that person is typically empowered to be the ultimate decision-maker. Slicing Pie provides some guidance on how to think about splitting equity. Regardless of which way you want you go, don’t put it off.
  2. Incorporate. Hard to be a company if you’re not. There are various formats, but if you plan to raise angel/venture funding, a Delaware-based C-corp is standard.
  3. Deal With Stock. At a minimum, issue stock to the founders, have each one buy their shares, and file an 83(b). Buying your shares at this stage might be $100. Filing the 83(b) election marks the date at which you purchased your shares, and shows that you bought them for what they were worth. This one piece of paper paper can make the difference between paying long-term capital gains rates (~20%) or income tax rates (~40%).
  4. Assign Intellectual Property. Ask everyone to sign a Proprietary Information and Inventions Assignment (“PIIA”). This document says that what they do at the company is owned by the company. Early on we had a friend who came by and brainstormed ideas. We thought of it as interesting banter. He later said he owned part of our storage design. While we worked it out together, a PIIA makes ownership clear.

The ownership split can be worked out by the founders directly. For the other items, I would involve lawyers. Some law firms will set up the basics and defer payment until you raise money or the business can pay for services out of operations. Gunderson Dettmer did that for us (ask for Bennett Yee). Cooley will do this on a casey-by-case basis as well.

Takeaway:  Don’t let the excitement of building a company distract you from filing the basic legal documents required to protect and grow your company.

Get Health Insurance

This item may seem out of place, but not having health insurance can easily bankrupt you personally, and that certainly won’t bode well for your company. While you can buy individual health insurance, it will often be less expensive to buy it as a company. Also, it will make recruiting employees more difficult if you do not offer healthcare. When we contacted brokers they asked us to send the W-2 of each employee that wanted coverage, but the founders weren’t taking a salary at first. To work around this, make the founders ‘officers’ of the company, and the healthcare brokers can then insure them. (Of course, you need to be ok with your co-founders being officers, but hopefully, that is logical anyway.)

Takeaway:  Don’t take your co-founders’ physical and financial health for granted. Health insurance can serve as both individual protection and a recruiting tool for future employees.

Building, Launching & Learning

Getting the company set up gives you the foundation, but ultimately a company with no product and no customers isn’t very interesting.


Ideally, you have one person on the team focusing on all of the items above and everyone else can be heads-down building product. There is a lot to say about building product, but for this post, I’ll just say that your goal is to get something out the door that is good enough to start collecting feedback. It doesn’t have to have every feature you dream of and doesn’t have to support 1 billion users on day one.


If you’re building a car or rocket, that may take some time. But with the availability of open-source software and cloud services, most startups should launch inside of a year.

Launching forces a scoping of the feature set to what’s critical, rallies the company around a goal, starts building awareness of your company and solution, and pushes forward the learning process. Backblaze launched in public beta on June 2, 2008, eight months after the founders all started working on it full-time.

Takeaway:  Focus on the most important features and launch.

Learn & Iterate

As much as we think we know about the customers and their needs, the launch process and beyond opens up all sorts of insights. This early period is critical to collect feedback and iterate, especially while both the product and company are still quite malleable. We initially planned on building peer-to-peer and local backup immediately on the heels of our online offering, but after launching found minimal demand for those features. On the other hand, there was tremendous demand from companies and resellers.

Takeaway:  Use the critical post-launch period to collect feedback and iterate.


“Live to fight another day.” If the company doesn’t survive, it’s hard to change the world. Let’s talk about some of the survival components.

Consider What You As A Founding Team Want & How You Work

Are you doing this because you hope to get rich? See yourself on the cover of Fortune? Make your own decisions? Work from home all the time? Founder fighting is the number one reason companies fail; the founders need to be on the same page as much as possible.

At Backblaze we agreed very early on that we wanted three things:

  1. Build products we were proud of
  2. Have fun
  3. Make money

This has driven various decisions over the years and has evolved into being part of the culture. For example, while Backblaze is absolutely a company with a profit motive, we do not compromise the product to make more money. Other directions are not bad; they’re just different.

Do you want a lifestyle business? Or want to build a billion dollar business? Want to run it forever or build it for a couple years and do something else?

Pretend you’re getting married to each other. Do some introspection and talk about your vision of the future a lot. Do you expect everyone to work 20 or 100 hours every week? In the office or remote? How do you like to work? What pet peeves do you have?

When getting married each person brings the “life they’ve known,” often influenced by the life their parents lived. Together they need to decide which aspects of their previous lives they want to keep, toss, or change. As founders coming together, you have the same opportunity for your new company.

Takeaway:  In order for a company to survive, the founders must agree on what they want the company to be. Have the discussions early.

Determine How You Will Fund Your Business

Raising venture capital is often seen as the only path, and considered the most important thing to start doing on day one. However, there are a variety of options for funding your business, including using money from savings, part-time work, friends & family money, loans, angels, and customers. Consider the right option for you, your founding team, and your business.

Conserve Cash

Whichever option you choose for funding your business, chances are high that you will not be flush with cash on day one. In certain situations, you actually don’t want to conserve cash because you’ve raised $100m and now you want to run as fast as you can to capture a market — cash is plentiful and time is not. However, with the exception of founder struggles, running out of cash is the most common way companies go under. There are many ways to conserve cash — limit hiring of employees and consultants, use lawyers and accountants sparingly, don’t spend on advertising, work from a home office, etc. The most important way is to simply ensure that you and your team are cash conscious, challenging decisions that commit you to spending cash.

Backblaze spent a total of $94,122 to get to public beta launch. That included building the backup application, our own server infrastructure, the website with account/billing/restore functionality, the marketing involved in getting to launch, and all the steps above in setting up the company, paying for healthcare, etc. The five founders took no salary during this time (which, of course, would have cost dramatically more), so most of this money went to computers, servers, hard drives, and other infrastructure.

Takeaway:  Minimize cash burn — it extends your runway and gives you options.

Slowly Flesh Out Your Team

We started with five co-founders, and thus a fairly fleshed-out team. A year in, we only added one person, a Mac architect. Three months later we shipped a beta of our Mac version, which has resulted in more than 50% of our revenue.

Minimizing hiring is key to cash conservation, and hiring ahead of getting market feedback is risky since you may realize that the talent you need will change. However, once you start getting feedback, think about the key people that you need to move your company forward. But be rigorous in determining whether they’re critical. We didn’t hire our first customer support person until all five founders were spending 20% of their time on it.

Takeaway:  Don’t hire in anticipation of market growth; hire to fuel the growth.

Keep Your Spirits Up

Startups are roller coasters of emotion. There have been some serious articles about founders suffering from depression and worse. The idea phase is exhilarating, then there is the slog of building. The launch is a blast, but the week after there are crickets.

On June 2, 2008, we launched in public beta with great press and hordes of customers. But a few months later we were signing up only about 10 new customers per month. That’s $50 new monthly recurring revenue (MRR) after a year of work and no salary.

On August 25, 2008, we brought on our Mac architect. Two months later, on October 26, 2008, Apple launched Time Machine — completely free and built-in backup for all Macs.

There were plenty of times when our prospects looked bleak. In the rearview mirror it’s easy to say, “well sure, but now you have lots of customers,” or “yes, but Time Machine doesn’t do cloud backup.” But at the time neither of these were a given.

Takeaway:  Getting up each day and believing that as a team you’ll figure it out will let you get to the point where you can look in the rearview mirror and say, “It looked bleak back then.”

Succeeding in Your First Year

I titled the post “Surviving Your First Year,” but if you manage to, 1) set up the company; 2) build, launch, and learn; and 3) survive, you will have done more than survive: you’ll have truly succeeded in your first year.

The post Surviving Your First Year appeared first on Backblaze Blog | Cloud Storage & Cloud Backup.

Have Friends Who Don’t Back Up? Share This Post!

Post Syndicated from Roderick Bauer original https://www.backblaze.com/blog/beginner-guide-to-computer-backup/

pointing out how to backup a computer

We’ve all been there.

A friend or family member comes to you knowing you’re a knowledgeable computer user and tells you that he has lost all the data on his computer.

You say, “Sure, I’ll help you get your computer working again. We’ll just restore your backup to a new drive or a new computer.”

Your friend looks at his feet and says, “I didn’t have a backup.”

You have to tell your friend that it’s very possible that without a backup that data is lost forever. It’s too late for a lecture about how he should have made regular backups of his computer. Your friend just wants his data back and he’s looking to you to help him.

You wish you could help. You realize that the time you could have helped was before the loss happened; when you could have helped your friend start making regular backups.

Yes, we’ve all been there. In fact, it’s how Backblaze got started.

You Can Be a Hero to a Friend by Sharing This Post

If you share this post with a friend or family member, you could avoid the situation where your friend loses his data and you wish you could help but can’t.

The following information will help your friend get started backing up in the easiest way possible — no fuss, no decisions, and no buying storage drives or plugging in cables.

The guide begins here:

Getting Started Backing Up

Your friend or family member has shared this guide with you because he or she believes you might benefit from backing up your computer. Don’t consider this an intervention, just a friendly tip that will save you lots of headaches, sorrow, and maybe money. With the right backup solution, it’s easy to protect your data against accidental deletion, theft, natural disaster, or malware, including ransomware.

Your friend was smart to send this to you, which probably means that you’re a smart person as well, so we’ll get right to the point. You likely know you should be backing up, but like all of us, don’t always get around to everything we should be doing.

You need a backup solution that is:

  1. Affordable
  2. Easy
  3. Never runs out of storage space
  4. Backs up everything automatically
  5. Restores files easily

Why Cloud Backup is the Best Solution For You

Backblaze Personal Backup was created for everyone who knows they should back up, but doesn’t. It backs up to the cloud, meaning that your data is protected in our secure data centers. A simple installation gets you started immediately, with no decisions about what or where to back up. It just works. And it’s just $5 a month to back up everything. Other services might limit the amount of data, the types of files, or both. With Backblaze, there’s no limit on the amount of data you can back up from your computer.

You can get started immediately with a free 15 day trial of Backblaze Unlimited Backup. In fewer than 5 minutes you’ll be all set.

Congratulations, You’re Done!

You can now celebrate. Your data is backed up and secure.

That’s it, and all you really need to get started backing up. We’ve included more details below, but frankly, the above is all you need to be safely and securely backed up.

You can tell the person who sent this to you that you’re now safely backed up and have moved on to other things, like what advice you can give them to help improve their life. Seriously, you might want to buy the person who sent this to you a coffee or another treat. They deserve it.

Here’s more information if you’d like to learn more about backing up.

Share or Email This Post to a Friend

Do your friend and yourself a favor and share this post. On the left side of the page (or at the bottom of the post) are buttons you can use to share this post on Twitter, Facebook, LinkedIn, and Google+, or to email it directly to your friend. It will take just a few seconds and could save your friend’s data.

It could also save you from having to give someone the bad news that her finances, photos, manuscript, or other work are gone forever. That would be nice.

But your real reward will be in knowing you did the right thing.

Tell us in the comments how it went. We’d like to hear.

The post Have Friends Who Don’t Back Up? Share This Post! appeared first on Backblaze Blog | Cloud Storage & Cloud Backup.

A Million ‘Pirate’ Boxes Sold in the UK During The Last Two Years

Post Syndicated from Andy original https://torrentfreak.com/a-million-pirate-boxes-sold-in-the-uk-during-the-last-two-years-170919/

With the devices hitting the headlines on an almost weekly basis, it probably comes as no surprise that ‘pirate’ set-top boxes are quickly becoming public enemy number one with video rightsholders.

Typically loaded with the legal Kodi software but augmented with third-party addons, these often Android-based pieces of hardware drag piracy out of the realm of the computer savvy and into the living rooms of millions.

One of the countries reportedly most affected by this boom is the UK. The consumption of these devices among the general public is said to have reached epidemic proportions, and anecdotal evidence suggests that terms like Kodi and Showbox are now household terms.

Today we have another report to digest, this time from the Federation Against Copyright Theft, or FACT as they’re often known. Titled ‘Cracking Down on Digital Piracy,’ the report provides a general overview of the piracy scene, tackling well-worn topics such as how release groups and site operators work, among others.

The report is produced by FACT after consultation with the Police Intellectual Property Crime Unit, Intellectual Property Office, Police Scotland, and anti-piracy outfit Entura International. It begins by noting that the vast majority of the British public aren’t involved in the consumption of infringing content.

“The most recent stats show that 75% of Brits who look at content online abide by the law and don’t download or stream it illegally – up from 70% in 2013. However, that still leaves 25% who do access material illegally,” the report reads.

The report quickly heads to the topic of ‘pirate’ set-top boxes which is unsurprising, not least due to FACT’s current focus as a business entity.

While it often positions itself alongside government bodies (which no doubt boosts its status with the general public), FACT is a private limited company serving The Premier League, another company desperate to stamp out the use of infringing devices.

Nevertheless, it’s difficult to argue with some of the figures cited in the report.

“At a conservative estimate, we believe a million set-top boxes with software added
to them to facilitate illegal downloads have been sold in the UK in the last couple
of years,” the Intellectual Property Office reveals.

Interestingly, given a growing tech-savvy public, FACT’s report notes that ready-configured boxes are increasingly coming into the country.

“Historically, individuals and organized gangs have added illegal apps and add-ons onto the boxes once they have been imported, to allow illegal access to premium channels. However more recently, more boxes are coming into the UK complete with illegal access to copyrighted content via apps and add-ons already installed,” FACT notes.

“Boxes are often stored in ‘fulfillment houses’ along with other illegal electrical items and sold on social media. The boxes are either sold as one-off purchases, or with a monthly subscription to access paid-for channels.”

While FACT press releases regularly blur the lines when people are prosecuted for supplying set-top boxes in general, it’s important to note that there are essentially two kinds of products on offer to the public.

The first relies on Kodi-type devices which provide on-going free access to infringing content. The second involves premium IPTV subscriptions which are a whole different level of criminality. Separating the two when reading news reports can be extremely difficult, but it’s a hugely important to recognize the difference when assessing the kinds of sentences set-top box suppliers are receiving in the UK.

Nevertheless, FACT correctly highlights that the supply of both kinds of product are on the increase, with various parties recognizing the commercial opportunities.

“A significant number of home-grown British criminals are now involved in this type of crime. Some of them import the boxes wholesale through entirely legal channels, and modify them with illegal software at home. Others work with sophisticated criminal networks across Europe to bring the boxes into the UK.

“They then sell these boxes online, for example through eBay or Facebook, sometimes managing to sell hundreds or thousands of boxes before being caught,” the company adds.

The report notes that in some cases the sale of infringing set-top boxes occurs through cottage industry, with suppliers often working on their own or with small groups of friends and family. Invetiably, perhaps, larger scale operations are reported to be part of networks with connections to other kinds of crime, such as dealing in drugs.

“In contrast to drugs, streaming devices provide a relatively steady and predictable revenue stream for these criminals – while still being lucrative, often generating hundreds of thousands of pounds a year, they are seen as a lower risk activity with less likelihood of leading to arrest or imprisonment,” FACT reports.

While there’s certainly the potential to earn large sums from ‘pirate’ boxes and premium IPTV services, operating on the “hundreds of thousands of pounds a year” scale in the UK would attract a lot of unwanted attention. That’s not saying that it isn’t already, however.

Noting that digital piracy has evolved hugely over the past three or four years, the report says that the cases investigated so far are just the “tip of the iceberg” and that many other cases are in the early stages and will only become known to the public in the months and years ahead.

Indeed, the Intellectual Property Office hints that some kind of large-scale enforcement action may be on the horizon.

“We have identified a significant criminal business model which we have discussed and shared with key law enforcement partners. I can’t go into detail on this, but as investigations take their course, you will see the scale,” an IPO spokesperson reveals.

While details are necessarily scarce, a source familiar with this area told TF that he would be very surprised if the targets aren’t the growing handful of commercial UK-based IPTV re-sellers who offer full subscription TV services for a few pounds per month.

“They’re brazen. Watch this space,” he said.

FACT’s full report, Cracking Down on Digital Piracy, can be downloaded here (pdf)

Source: TF, for the latest info on copyright, file-sharing, torrent sites and ANONYMOUS VPN services.