Tag Archives: bulk

Amazon Neptune Generally Available

Post Syndicated from Randall Hunt original https://aws.amazon.com/blogs/aws/amazon-neptune-generally-available/

Amazon Neptune is now Generally Available in US East (N. Virginia), US East (Ohio), US West (Oregon), and EU (Ireland). Amazon Neptune is a fast, reliable, fully-managed graph database service that makes it easy to build and run applications that work with highly connected datasets. At the core of Neptune is a purpose-built, high-performance graph database engine optimized for storing billions of relationships and querying the graph with millisecond latencies. Neptune supports two popular graph models, Property Graph and RDF, through Apache TinkerPop Gremlin and SPARQL, allowing you to easily build queries that efficiently navigate highly connected datasets. Neptune can be used to power everything from recommendation engines and knowledge graphs to drug discovery and network security. Neptune is fully-managed with automatic minor version upgrades, backups, encryption, and fail-over. I wrote about Neptune in detail for AWS re:Invent last year and customers have been using the preview and providing great feedback that the team has used to prepare the service for GA.

Now that Amazon Neptune is generally available there are a few changes from the preview:

Launching an Amazon Neptune Cluster

Launching a Neptune cluster is as easy as navigating to the AWS Management Console and clicking create cluster. Of course you can also launch with CloudFormation, the CLI, or the SDKs.

You can monitor your cluster health and the health of individual instances through Amazon CloudWatch and the console.

Additional Resources

We’ve created two repos with some additional tools and examples here. You can expect continuous development on these repos as we add additional tools and examples.

  • Amazon Neptune Tools Repo
    This repo has a useful tool for converting GraphML files into Neptune compatible CSVs for bulk loading from S3.
  • Amazon Neptune Samples Repo
    This repo has a really cool example of building a collaborative filtering recommendation engine for video game preferences.

Purpose Built Databases

There’s an industry trend where we’re moving more and more onto purpose-built databases. Developers and businesses want to access their data in the format that makes the most sense for their applications. As cloud resources make transforming large datasets easier with tools like AWS Glue, we have a lot more options than we used to for accessing our data. With tools like Amazon Redshift, Amazon Athena, Amazon Aurora, Amazon DynamoDB, and more we get to choose the best database for the job or even enable entirely new use-cases. Amazon Neptune is perfect for workloads where the data is highly connected across data rich edges.

I’m really excited about graph databases and I see a huge number of applications. Looking for ideas of cool things to build? I’d love to build a web crawler in AWS Lambda that uses Neptune as the backing store. You could further enrich it by running Amazon Comprehend or Amazon Rekognition on the text and images found and creating a search engine on top of Neptune.

As always, feel free to reach out in the comments or on twitter to provide any feedback!


The Benefits of Side Projects

Post Syndicated from Bozho original https://techblog.bozho.net/the-benefits-of-side-projects/

Side projects are the things you do at home, after work, for your own “entertainment”, or to satisfy your desire to learn new stuff, in case your workplace doesn’t give you that opportunity (or at least not enough of it). Side projects are also a way to build stuff that you think is valuable but not necessarily “commercialisable”. Many side projects are open-sourced sooner or later and some of them contribute to the pool of tools at other people’s disposal.

I’ve outlined one recommendation about side projects before – do them with technologies that are new to you, so that you learn important things that will keep you better positioned in the software world.

But there are more benefits than that – serendipitous benefits, for example. And I’d like to tell some personal stories about that. I’ll focus on a few examples from my list of side projects to show how, through a sort-of butterfly effect, they helped shape my career.

The computoser project, no matter how cool algorithmic music composition, didn’t manage to have much of a long term impact. But it did teach me something apart from niche musical theory – how to read a bulk of scientific papers (mostly computer science) and understand them without being formally trained in the particular field. We’ll see how that was useful later.

Then there was the “State alerts” project – a website that scraped content from public institutions in my country (legislation, legislation proposals, decisions by regulators, new tenders, etc.), made them searchable, and “subscribable” – so that you get notified when a keyword of interest is mentioned in newly proposed legislation, for example. (I obviously subscribed for “information technologies” and “electronic”).

And that project turned out to have a significant impact on the following years. First, I chose a new technology to write it with – Scala. Which turned out to be of great use when I started working at TomTom, and on the 3rd day I was transferred to a Scala project, which was way cooler and much more complex than the original one I was hired for. It was a bit ironic, as my colleagues had just read that “I don’t like Scala” a few weeks earlier, but nevertheless, that was one of the most interesting projects I’ve worked on, and it went on for two years. Had I not known Scala, I’d probably be gone from TomTom much earlier (as the other project was restructured a few times), and I would not have learned many of the scalability, architecture and AWS lessons that I did learn there.

But the very same project had an even more important follow-up. Because if its “civic hacking” flavour, I was invited to join an informal group of developers (later officiated as an NGO) who create tools that are useful for society (something like MySociety.org). That group gathered regularly, discussed both tools and policies, and at some point we put up a list of policy priorities that we wanted to lobby policy makers. One of them was open source for the government, the other one was open data. As a result of our interaction with an interim government, we donated the official open data portal of my country, functioning to this day.

As a result of that, a few months later we got a proposal from the deputy prime minister’s office to “elect” one of the group for an advisor to the cabinet. And we decided that could be me. So I went for it and became advisor to the deputy prime minister. The job has nothing to do with anything one could imagine, and it was challenging and fascinating. We managed to pass legislation, including one that requires open source for custom projects, eID and open data. And all of that would not have been possible without my little side project.

As for my latest side project, LogSentinel – it became my current startup company. And not without help from the previous two mentioned above – the computer science paper reading was of great use when I was navigating the crypto papers landscape, and from the government job I not only gained invaluable legal knowledge, but I also “got” a co-founder.

Some other side projects died without much fanfare, and that’s fine. But the ones above shaped my “story” in a way that would not have been possible otherwise.

And I agree that such serendipitous chain of events could have happened without side projects – I could’ve gotten these opportunities by meeting someone at a bar (unlikely, but who knows). But we, as software engineers, are capable of tilting chance towards us by utilizing our skills. Side projects are our “extracurricular activities”, and they often lead to unpredictable, but rather positive chains of events. They would rarely be the only factor, but they are certainly great at unlocking potential.

The post The Benefits of Side Projects appeared first on Bozho's tech blog.

A Peek Behind the Mail Curtain

Post Syndicated from marcelatoath original https://yahooeng.tumblr.com/post/174023151641


By Libby Lin, Principal Product Manager

Well, we actually won’t show you how we create the magic in our big OATH consumer mail factory. But nevertheless we wanted to share how interested developers could leverage some of our unique features we offer for our Yahoo and AOL Mail customers.

To drive experiences like our travel and shopping smart views or message threading, we tag qualified mails with something we call DECOS and THREADID. While we will not indulge in explaining how exactly we use them internally, we wanted to share how they can be used and accessed through IMAP.

So let’s just look at a sample IMAP command chain. We’ll just assume that you are familiar with the IMAP protocol at this point and you know how to properly talk to an IMAP server.

So here’s how you would retrieve DECO and THREADIDs for specific messages:


   openssl s_client -crlf -connect imap.mail.yahoo.com:993


   a login username password

   a OK LOGIN completed


   a list “” “*”

   * LIST (\Junk \HasNoChildren) “/” “Bulk Mail”

   * LIST (\Archive \HasNoChildren) “/” “Archive”

   * LIST (\Drafts \HasNoChildren) “/” “Draft”

   * LIST (\HasNoChildren) “/” “Inbox”

   * LIST (\HasNoChildren) “/” “Notes”

   * LIST (\Sent \HasNoChildren) “/” “Sent”

   * LIST (\Trash \HasChildren) “/” “Trash”

   * LIST (\HasNoChildren) “/” “Trash/l2”

   * LIST (\HasChildren) “/” “test level 1”

   * LIST (\HasNoChildren) “/” “test level 1/nestedfolder”

   * LIST (\HasNoChildren) “/” “test level 1/test level 2”

   * LIST (\HasNoChildren) “/” “&T2BZfXso-”

   * LIST (\HasNoChildren) “/” “&gQKAqk7WWr12hA-”

   a OK LIST completed


   a select inbox

   * 94 EXISTS

   * 0 RECENT

   * OK [UIDVALIDITY 1453335194] UIDs valid

   * OK [UIDNEXT 40213] Predicted next UID

   * FLAGS (\Answered \Deleted \Draft \Flagged \Seen $Forwarded $Junk $NotJunk)

   * OK [PERMANENTFLAGS (\Answered \Deleted \Draft \Flagged \Seen $Forwarded $Junk $NotJunk)] Permanent flags


   a OK [READ-WRITE] SELECT completed; now in selected state


   a uid search 1:*

   * SEARCH 1 2 3 4 11 12 14 23 24 75 76 77 78 114 120 121 124 128 129 130 132 133 134 135 136 137 138 40139 40140 40141 40142 40143 40144 40145 40146 40147 40148     40149 40150 40151 40152 40153 40154 40155 40156 40157 40158 40159 40160 40161 40162 40163 40164 40165 40166 40167 40168 40172 40173 40174 40175 40176     40177 40178 40179 40182 40183 40184 40185 40186 40187 40188 40190 40191 40192 40193 40194 40195 40196 40197 40198 40199 40200 40201 40202 40203 40204     40205 40206 40207 40208 40209 40211 40212

   a OK UID SEARCH completed


   a uid fetch 40212 (X-MSG-DECOS X-MSG-ID X-MSG-THREADID)

   * 94 FETCH (UID 40212 X-MSG-THREADID “108” X-MSG-ID “ACfIowseFt7xWtj0og0L2G0T1wM” X-MSG-DECOS (“FTI” “F1” “EML”))

   a OK UID FETCH completed

Brutus 2: the gaming PC case of your dreams

Post Syndicated from Janina Ander original https://www.raspberrypi.org/blog/brutus-2-gaming-pc-case/

Attention, case modders: take a look at the Brutus 2, an extremely snazzy computer case with a partly transparent, animated side panel that’s powered by a Pi. Daniel Otto and Carsten Lehman have a current crowdfunder for the case; their video is in German, but the looks of the build speak for themselves. There are some truly gorgeous effects here.

der BRUTUS 2 by 3nb Gaming

Vorbestellungen ab sofort auf https://www.startnext.com/brutus2 Weitere Infos zu uns auf: https://3nb.de https://www.facebook.com/3nb.de https://www.instagram.com/3nb.de Über 3nb: – GbR aus Leipzig, gegründet 2017 – wir kommen aus den Bereichen Elektronik und Informatik – erstes Produkt: der Brutus One ein Gaming PC mit transparentem Display in der Seite Kurzinfo Brutus 2: – Markencomputergehäuse für Gaming- /Casemoddingszene – Besonderheit: animiertes Seitenfenster angesteuert mit einem Raspberry Pi – Vorteile von unserem Case: o Case ist einzeln lieferbar und nicht nur als komplett-PC o kein Leistungsverbrauch der Grafikkarte dank integriertem Raspberry Pi o bessere Darstellung von Texten und Grafiken durch unscharfen Hintergrund

What’s case modding?

Case modding just means modifying your computer or gaming console’s case, and it’s very popular in the gaming community. Some mods are functional, while others improve the way the case looks. Lots of dedicated gamers don’t only want a powerful computer, they also want it to look amazing — at home, or at LAN parties and games tournaments.

The Brutus 2 case

The Brutus 2 case is made by Daniel and Carsten’s startup, 3nb electronics, and it’s a product that is officially Powered by Raspberry Pi. Its standout feature is the semi-transparent TFT screen, which lets you play any video clip you choose while keeping your gaming hardware on display. It looks incredibly cool. All the graphics for the case’s screen are handled by a Raspberry Pi, so it doesn’t use any of your main PC’s GPU power and your gaming won’t suffer.

Brutus 2 PC case powered by Raspberry Pi

The software

To use Brutus 2, you just need to run a small desktop application on your PC to choose what you want to display on the case. A number of neat animations are included, and you can upload your own if you want.

So far, the app only runs on Windows, but 3nb electronics are planning to make the code open-source, so you can modify it for other operating systems, or to display other file types. This is true to the spirit of the case modding and Raspberry Pi communities, who love adapting, retrofitting, and overhauling projects and code to fit their needs.

Brutus 2 PC case powered by Raspberry Pi

Daniel and Carsten say that one of their campaign’s stretch goals is to implement more functionality in the Brutus 2 app. So in the future, the case could also show things like CPU temperature, gaming stats, and in-game messages. Of course, there’s nothing stopping you from integrating features like that yourself.

If you have any questions about the case, you can post them directly to Daniel and Carsten here.

The crowdfunding campaign

The Brutus 2 campaign on Startnext is currently halfway to its first funding goal of €10000, with over three weeks to go until it closes. If you’re quick, you still be may be able to snatch one of the early-bird offers. And if your whole guild NEEDS this, that’s OK — there are discounts for bulk orders.

The post Brutus 2: the gaming PC case of your dreams appeared first on Raspberry Pi.

Under-Fire “Kodi Box” Company “Sold to Chinese Investor” For US$8.82m

Post Syndicated from Andy original https://torrentfreak.com/under-fire-kodi-box-company-sold-to-chinese-investor-for-us8-82m-180426/

Back in 2016, an article appeared in Kiwi media discussing the rise of a new company pledging to beat media giant Sky TV at its own game.

My Box NZ owner Krish Reddy told the publication he was selling Android boxes loaded with Kodi software and augmented with third-party addons.

Without any hint of fear, he stated that these devices enabled customers to access movies, TV shows and live channels for free, after shelling out a substantial US$182 for the box first, that is.

“Why pay $80 minimum per month for Sky when for one payment you can have it free for good?” a claim on the company’s website asked.

Noting that he’d been importing the boxes from China, Reddy suggested that his lawyers hadn’t found any problem with the business plan.

“I don’t see why [Sky] would contact me but if they do contact me and … if there’s something of theirs that they feel I’ve unlawfully taken then yeah … but as it stands I don’t [have any concerns],” he said.

At this point, Reddy said he’d been selling the boxes for just six weeks and had shifted around 80 units. To get coverage from a national newspaper at this stage of the game must’ve been very much appreciated but Reddy didn’t stop there.

In a bulk advertising email sent out to 50,000 people, Reddy described his boxes as “better than Sky”. However, by design or misfortune, the email managed to land in the inboxes of 50 Sky TV staff and directors, something that didn’t go unnoticed by the TV giant.

With Reddy claiming sales of 8,000 units, Sky ran out of patience last April. In a letter from its lawyers, the pay-TV company said Reddy’s devices breached copyright law and the Fair Trading Act. Reddy responded by calling the TV giant “a playground bully”, again denying that he was breaking the law.

“From a legal perspective, what we do is completely within the law. We advertise Sky television channels being available through our website and social media platforms as these are available via streams which you can find through My Box,” he said.

“The content is already available, I’m not going out there and bringing the content so how am I infringing the copyright… the content is already there, if someone uses the box to search for the content, that’s what it is.”

The initial compensation demand from Sky against Reddy’s company My Box ran to NZD$1.4m, around US$1m. It was an amount that had the potential rise by millions if matters got drawn out and/or escalated. But despite picking a terrible opponent in a battle he was unlikely to win, Reddy refused to give up.

“[Sky’s] point of view is they own copyright and I’m destroying the market by giving people content for free. To me it is business; I have got something that is new … that’s competition,” he said.

The Auckland High Court heard the case against My Box last month with Judge Warwick Smith reserving his judgment and Reddy still maintaining that his business is entirely legal. Sales were fantastic, he said, with 20,000 devices sold to customers in 12 countries.

Then something truly amazing happened.

A company up to its eyeballs in litigation, selling a commodity product that an amateur can buy and configure at home for US$40, reportedly got a chance of a lifetime. Reddy revealed to Stuff that a Chinese investor had offered to buy his company for an eye-watering NZ$10 million (US$7.06m).

“We have to thank Sky,” he said. “If they had left us alone we would just have been selling a few boxes, but the controversy made us world famous.”

Reddy noted he’d been given 21 days to respond to the offer, but refused to name the company. Interestingly, he also acknowledged that if My Box lost its case, the company would be liable for damages. However, that wouldn’t bother the potential investor.

“It makes no difference to them whether we win or lose, because their operations won’t be in New Zealand,” Reddy said.

According to the entrepreneur, that’s how things are playing out.

The Chinese firm – which Reddy is still refusing to name – has apparently accepted a counter offer from Reddy of US$8.8m for My Box. As a result, Reddy will wrap up his New Zealand operations within the next 90 days and his six employees will be rendered unemployed.

Given that anyone with the ability to install Kodi and a few addons before putting a box in the mail could replicate Reddy’s business model, the multi-million dollar offer for My Box was never anything less than a bewildering business proposition. That someone carried through with it an even higher price is so fantastic as to be almost unbelievable.

In a sea of unhappy endings for piracy-enabled Kodi box sellers globally, this is the only big win to ever grace the headlines. Assuming this really is the end of the story (and that might not be the case) it will almost certainly be the last.

Source: TF, for the latest info on copyright, file-sharing, torrent sites and more. We also have VPN reviews, discounts, offers and coupons.

AWS AppSync – Production-Ready with Six New Features

Post Syndicated from Jeff Barr original https://aws.amazon.com/blogs/aws/aws-appsync-production-ready-with-six-new-features/

If you build (or want to build) data-driven web and mobile apps and need real-time updates and the ability to work offline, you should take a look at AWS AppSync. Announced in preview form at AWS re:Invent 2017 and described in depth here, AWS AppSync is designed for use in iOS, Android, JavaScript, and React Native apps. AWS AppSync is built around GraphQL, an open, standardized query language that makes it easy for your applications to request the precise data that they need from the cloud.

I’m happy to announce that the preview period is over and that AWS AppSync is now generally available and production-ready, with six new features that will simplify and streamline your application development process:

Console Log Access – You can now see the CloudWatch Logs entries that are created when you test your GraphQL queries, mutations, and subscriptions from within the AWS AppSync Console.

Console Testing with Mock Data – You can now create and use mock context objects in the console for testing purposes.

Subscription Resolvers – You can now create resolvers for AWS AppSync subscription requests, just as you can already do for query and mutate requests.

Batch GraphQL Operations for DynamoDB – You can now make use of DynamoDB’s batch operations (BatchGetItem and BatchWriteItem) across one or more tables. in your resolver functions.

CloudWatch Support – You can now use Amazon CloudWatch Metrics and CloudWatch Logs to monitor calls to the AWS AppSync APIs.

CloudFormation Support – You can now define your schemas, data sources, and resolvers using AWS CloudFormation templates.

A Brief AppSync Review
Before diving in to the new features, let’s review the process of creating an AWS AppSync API, starting from the console. I click Create API to begin:

I enter a name for my API and (for demo purposes) choose to use the Sample schema:

The schema defines a collection of GraphQL object types. Each object type has a set of fields, with optional arguments:

If I was creating an API of my own I would enter my schema at this point. Since I am using the sample, I don’t need to do this. Either way, I click on Create to proceed:

The GraphQL schema type defines the entry points for the operations on the data. All of the data stored on behalf of a particular schema must be accessible using a path that begins at one of these entry points. The console provides me with an endpoint and key for my API:

It also provides me with guidance and a set of fully functional sample apps that I can clone:

When I clicked Create, AWS AppSync created a pair of Amazon DynamoDB tables for me. I can click Data Sources to see them:

I can also see and modify my schema, issue queries, and modify an assortment of settings for my API.

Let’s take a quick look at each new feature…

Console Log Access
The AWS AppSync Console already allows me to issue queries and to see the results, and now provides access to relevant log entries.In order to see the entries, I must enable logs (as detailed below), open up the LOGS, and check the checkbox. Here’s a simple mutation query that adds a new event. I enter the query and click the arrow to test it:

I can click VIEW IN CLOUDWATCH for a more detailed view:

To learn more, read Test and Debug Resolvers.

Console Testing with Mock Data
You can now create a context object in the console where it will be passed to one of your resolvers for testing purposes. I’ll add a testResolver item to my schema:

Then I locate it on the right-hand side of the Schema page and click Attach:

I choose a data source (this is for testing and the actual source will not be accessed), and use the Put item mapping template:

Then I click Select test context, choose Create New Context, assign a name to my test content, and click Save (as you can see, the test context contains the arguments from the query along with values to be returned for each field of the result):

After I save the new Resolver, I click Test to see the request and the response:

Subscription Resolvers
Your AWS AppSync application can monitor changes to any data source using the @aws_subscribe GraphQL schema directive and defining a Subscription type. The AWS AppSync client SDK connects to AWS AppSync using MQTT over Websockets and the application is notified after each mutation. You can now attach resolvers (which convert GraphQL payloads into the protocol needed by the underlying storage system) to your subscription fields and perform authorization checks when clients attempt to connect. This allows you to perform the same fine grained authorization routines across queries, mutations, and subscriptions.

To learn more about this feature, read Real-Time Data.

Batch GraphQL Operations
Your resolvers can now make use of DynamoDB batch operations that span one or more tables in a region. This allows you to use a list of keys in a single query, read records multiple tables, write records in bulk to multiple tables, and conditionally write or delete related records across multiple tables.

In order to use this feature the IAM role that you use to access your tables must grant access to DynamoDB’s BatchGetItem and BatchPutItem functions.

To learn more, read the DynamoDB Batch Resolvers tutorial.

CloudWatch Logs Support
You can now tell AWS AppSync to log API requests to CloudWatch Logs. Click on Settings and Enable logs, then choose the IAM role and the log level:

CloudFormation Support
You can use the following CloudFormation resource types in your templates to define AWS AppSync resources:

AWS::AppSync::GraphQLApi – Defines an AppSync API in terms of a data source (an Amazon Elasticsearch Service domain or a DynamoDB table).

AWS::AppSync::ApiKey – Defines the access key needed to access the data source.

AWS::AppSync::GraphQLSchema – Defines a GraphQL schema.

AWS::AppSync::DataSource – Defines a data source.

AWS::AppSync::Resolver – Defines a resolver by referencing a schema and a data source, and includes a mapping template for requests.

Here’s a simple schema definition in YAML form:

    Type: "AWS::AppSync::GraphQLSchema"
      - AppSyncGraphQLApi
      ApiId: !GetAtt AppSyncGraphQLApi.ApiId
      Definition: |
        schema {
          query: Query
          mutation: Mutation
        type Query {
          singlePost(id: ID!): Post
          allPosts: [Post]
        type Mutation {
          putPost(id: ID!, title: String!): Post
        type Post {
          id: ID!
          title: String!

Available Now
These new features are available now and you can start using them today! Here are a couple of blog posts and other resources that you might find to be of interest:




Tinkernut’s hidden Coke bottle spy cam

Post Syndicated from Alex Bate original https://www.raspberrypi.org/blog/tinkernuts-spy-cam/

Go undercover and keep an eye on your stuff with this brilliant secret Coke bottle spy cam from Tinkernut!

Secret Coke Bottle SPY CAM! – Weekend Hacker #1803

SPECIAL NOTE*** THE FULL TUTORIAL WILL BE AVAILABLE NEXT WEEK April Fools! What a terrible day. So many pranks. You can’t believe anything you read. People invading your space. The mental and physical anguish of enduring the day. It’s time to fight back! Let’s catch the perps in action by making a device that always watches.

Keeping tabs

A Raspberry Pi Zero W, a small camera, and a rechargeable Lithium Polymer (LiPo) battery constitute the bulk of this project’s tech. A pair of 3D-printed parts, and gelatine-solidified Coke Zero make up the fake fizzy body.

Tinkernut Coke bottle Raspberry Pi Spy Cam

“So let’s make this video as short as possible and just buy a cheap pre-made spy cam off of Amazon. Just kidding,” Tinkernut jokes in the tutorial video for the project, before going through the step-by-step process of using the Raspberry Pi to “DIY this the right way”.

After accessing the Zero W from his laptop via SSH, Tinkernut opted for using the rpi_camera_surveillance_system Python script written by GitHub user RuiSantosdotme to control the spy cam. Luckily, this meant no additional library setup, and basically no lag on the video feed.

What we want to do is create a script that activates the camera and serves it to a web page so that we can access it from any web browser. There are plenty of different ways to do this (Motion, Raspivid, etc), but I found a simple Python script that does everything I need it to do and doesn’t require any extra software or libraries to install. The best thing about it is that the lag time is practically unnoticeable.

With the code in place, every boot-up of the Raspberry Pi automatically launches both the script and a web page of the live video, allowing for constant monitoring of potential sneaks and thieves.

Tinkernut Coke bottle Raspberry Pi Spy Cam

The projects is powered by a 1500mAh LiPo battery and the Adafruit LiPo charger. It also includes a simple on/off switch, which Tinkernut wired to the charger and the Pi’s PP1 and PP6 connector pads.

Tinkernut Coke bottle Raspberry Pi Spy Cam

Tinkernut decided to use a Coke Zero bottle for the build, incorporating 3D-printed parts to house the Pi, and a mix of Coke and gelatine to create a realistic-looking filling for the bottle. However, the setup can be transferred to pretty much any hollow item in your home, say, a cookie jar or a cracker box. So get creative and get spying!

A complete spy cam how-to

If you’d like to make your own secret spy cam, you can find a tutorial for Tinkernut’s build at hackster.io, or follow along with his video below. Also make sure to subscribe his YouTube channel to be updated on all his newest builds — they’re rather splendid.

BUILD: Coke Bottle SPY CAM! – Tinkernut Workbench

Learn how to take a regular Coke Zero bottle, cram a Raspberry Pi and webcam inside of it, and have it still look like a regular Coke Zero bottle. Why would you want to do this? To spy on those irritating April Fooligans!!!

And if you’re interested in more spy-themed digital making projects, check out our complete 007 how-to guide for links to tutorials such as our Sense HAT puzzle box, Parent detector, and Laser tripwire.

The post Tinkernut’s hidden Coke bottle spy cam appeared first on Raspberry Pi.

A geometric Rust adventure

Post Syndicated from Eevee original https://eev.ee/blog/2018/03/30/a-geometric-rust-adventure/

Hi. Yes. Sorry. I’ve been trying to write this post for ages, but I’ve also been working on a huge writing project, and apparently I have a very limited amount of writing mana at my disposal. I think this is supposed to be a Patreon reward from January. My bad. I hope it’s super great to make up for the wait!

I recently ported some math code from C++ to Rust in an attempt to do a cool thing with Doom. Here is my story.

The problem

I presented it recently as a conundrum (spoilers: I solved it!), but most of those details are unimportant.

The short version is: I have some shapes. I want to find their intersection.

Really, I want more than that: I want to drop them all on a canvas, intersect everything with everything, and pluck out all the resulting polygons. The input is a set of cookie cutters, and I want to press them all down on the same sheet of dough and figure out what all the resulting contiguous pieces are. And I want to know which cookie cutter(s) each piece came from.

But intersection is a good start.

Example of the goal.  Given two squares that overlap at their corners, I want to find the small overlap piece, plus the two L-shaped pieces left over from each square

I’m carefully referring to the input as shapes rather than polygons, because each one could be a completely arbitrary collection of lines. Obviously there’s not much you can do with shapes that aren’t even closed, but at the very least, I need to handle concavity and multiple disconnected polygons that together are considered a single input.

This is a non-trivial problem with a lot of edge cases, and offhand I don’t know how to solve it robustly. I’m not too eager to go figure it out from scratch, so I went hunting for something I could build from.

(Infuriatingly enough, I can just dump all the shapes out in an SVG file and any SVG viewer can immediately solve the problem, but that doesn’t quite help me. Though I have had a few people suggest I just rasterize the whole damn problem, and after all this, I’m starting to think they may have a point.)

Alas, I couldn’t find a Rust library for doing this. I had a hard time finding any library for doing this that wasn’t a massive fully-featured geometry engine. (I could’ve used that, but I wanted to avoid non-Rust dependencies if possible, since distributing software is already enough of a nightmare.)

A Twitter follower directed me towards a paper that described how to do very nearly what I wanted and nothing else: “A simple algorithm for Boolean operations on polygons” by F. Martínez (2013). Being an academic paper, it’s trapped in paywall hell; sorry about that. (And as I understand it, none of the money you’d pay to get the paper would even go to the authors? Is that right? What a horrible and predatory system for discovering and disseminating knowledge.)

The paper isn’t especially long, but it does describe an awful lot of subtle details and is mostly written in terms of its own reference implementation. Rather than write my own implementation based solely on the paper, I decided to try porting the reference implementation from C++ to Rust.

And so I fell down the rabbit hole.

The basic algorithm

Thankfully, the author has published the sample code on his own website, if you want to follow along. (It’s the bottom link; the same author has, confusingly, published two papers on the same topic with similar titles, four years apart.)

If not, let me describe the algorithm and how the code is generally laid out. The algorithm itself is based on a sweep line, where a vertical line passes across the plane and ✨ does stuff ✨ as it encounters various objects. This implementation has no physical line; instead, it keeps track of which segments from the original polygon would be intersecting the sweep line, which is all we really care about.

A vertical line is passing rightwards over a couple intersecting shapes.  The line current intersects two of the shapes' sides, and these two sides are the "sweep list"

The code is all bundled inside a class with only a single public method, run, because… that’s… more object-oriented, I guess. There are several helper methods, and state is stored in some attributes. A rough outline of run is:

  1. Run through all the line segments in both input polygons. For each one, generate two SweepEvents (one for each endpoint) and add them to a std::deque for storage.

    Add pointers to the two SweepEvents to a std::priority_queue, the event queue. This queue uses a custom comparator to order the events from left to right, so the top element is always the leftmost endpoint.

  2. Loop over the event queue (where an “event” means the sweep line passed over the left or right end of a segment). Encountering a left endpoint means the sweep line is newly touching that segment, so add it to a std::set called the sweep list. An important point is that std::set is ordered, and the sweep list uses a comparator that keeps segments in order vertically.

    Encountering a right endpoint means the sweep line is leaving a segment, so that segment is removed from the sweep list.

  3. When a segment is added to the sweep list, it may have up to two neighbors: the segment above it and the segment below it. Call possibleIntersection to check whether it intersects either of those neighbors. (This is nearly sufficient to find all intersections, which is neat.)

  4. If possibleIntersection detects an intersection, it will split each segment into two pieces then and there. The old segment is shortened in-place to become the left part, and a new segment is created for the right part. The new endpoints at the point of intersection are added to the event queue.

  5. Some bookkeeping is done along the way to track which original polygons each segment is inside, and eventually the segments are reconstructed into new polygons.

Hopefully that’s enough to follow along. It took me an inordinately long time to tease this out. The comments aren’t especially helpful.

    std::deque<SweepEvent> eventHolder;    // It holds the events generated during the computation of the boolean operation

Syntax and basic semantics

The first step was to get something that rustc could at least parse, which meant translating C++ syntax to Rust syntax.

This was surprisingly straightforward! C++ classes become Rust structs. (There was no inheritance here, thankfully.) All the method declarations go away. Method implementations only need to be indented and wrapped in impl.

I did encounter some unnecessarily obtuse uses of the ternary operator:

(prevprev != sl.begin()) ? --prevprev : prevprev = sl.end();

Rust doesn’t have a ternary — you can use a regular if block as an expression — so I expanded these out.

C++ switch blocks become Rust match blocks, but otherwise function basically the same. Rust’s enums are scoped (hallelujah), so I had to explicitly spell out where enum values came from.

The only really annoying part was changing function signatures; C++ types don’t look much at all like Rust types, save for the use of angle brackets. Rust also doesn’t pass by implicit reference, so I needed to sprinkle a few &s around.

I would’ve had a much harder time here if this code had relied on any remotely esoteric C++ functionality, but thankfully it stuck to pretty vanilla features.

Language conventions

This is a geometry problem, so the sample code unsurprisingly has its own home-grown point type. Rather than port that type to Rust, I opted to use the popular euclid crate. Not only is it code I didn’t have to write, but it already does several things that the C++ code was doing by hand inline, like dot products and cross products. And all I had to do was add one line to Cargo.toml to use it! I have no idea how anyone writes C or C++ without a package manager.

The C++ code used getters, i.e. point.x (). I’m not a huge fan of getters, though I do still appreciate the need for them in lowish-level systems languages where you want to future-proof your API and the language wants to keep a clear distinction between attribute access and method calls. But this is a point, which is nothing more than two of the same numeric type glued together; what possible future logic might you add to an accessor? The euclid authors appear to side with me and leave the coordinates as public fields, so I took great joy in removing all the superfluous parentheses.

Polygons are represented with a Polygon class, which has some number of Contours. A contour is a single contiguous loop. Something you’d usually think of as a polygon would only have one, but a shape with a hole would have two: one for the outside, one for the inside. The weird part of this arrangement was that Polygon implemented nearly the entire STL container interface, then waffled between using it and not using it throughout the rest of the code. Rust lets anything in the same module access non-public fields, so I just skipped all that and used polygon.contours directly. Hell, I think I made contours public.

Finally, the SweepEvent type has a pol field that’s declared as an enum PolygonType (either SUBJECT or CLIPPING, to indicate which of the two inputs it is), but then some other code uses the same field as a numeric index into a polygon’s contours. Boy I sure do love static typing where everything’s a goddamn integer. I wanted to extend the algorithm to work on arbitrarily many input polygons anyway, so I scrapped the enum and this became a usize.

Then I got to all the uses of STL. I have only a passing familiarity with the C++ standard library, and this code actually made modest use of it, which caused some fun days-long misunderstandings.

As mentioned, the SweepEvents are stored in a std::deque, which is never read from. It took me a little thinking to realize that the deque was being used as an arena: it’s the canonical home for the structs so pointers to them can be tossed around freely. (It can’t be a std::vector, because that could reallocate and invalidate all the pointers; std::deque is probably a doubly-linked list, and guarantees no reallocation.)

Rust’s standard library does have a doubly-linked list type, but I knew I’d run into ownership hell here later anyway, so I think I replaced it with a Rust Vec to start with. It won’t compile either way, so whatever. We’ll get back to this in a moment.

The list of segments currently intersecting the sweep line is stored in a std::set. That type is explicitly ordered, which I’m very glad I knew already. Rust has two set types, HashSet and BTreeSet; unsurprisingly, the former is unordered and the latter is ordered. Dropping in BTreeSet and fixing some method names got me 90% of the way there.

Which brought me to the other 90%. See, the C++ code also relies on finding nodes adjacent to the node that was just inserted, via STL iterators.

next = prev = se->posSL = it = sl.insert(se).first;
(prev != sl.begin()) ? --prev : prev = sl.end();

I freely admit I’m bad at C++, but this seems like something that could’ve used… I don’t know, 1 comment. Or variable names more than two letters long. What it actually does is:

  1. Add the current sweep event (se) to the sweep list (sl), which returns a pair whose first element is an iterator pointing at the just-inserted event.

  2. Copies that iterator to several other variables, including prev and next.

  3. If the event was inserted at the beginning of the sweep list, set prev to the sweep list’s end iterator, which in C++ is a legal-but-invalid iterator meaning “the space after the end” or something. This is checked for in later code, to see if there is a previous event to look at. Otherwise, decrement prev, so it’s now pointing at the event immediately before the inserted one.

  4. Increment next normally. If the inserted event is last, then this will bump next to the end iterator anyway.

In other words, I need to get the previous and next elements from a BTreeSet. Rust does have bidirectional iterators, which BTreeSet supports… but BTreeSet::insert only returns a bool telling me whether or not anything was inserted, not the position. I came up with this:

let mut maybe_below = active_segments.range(..segment).last().map(|v| *v);
let mut maybe_above = active_segments.range(segment..).next().map(|v| *v);

The range method returns an iterator over a subset of the tree. The .. syntax makes a range (where the right endpoint is exclusive), so ..segment finds the part of the tree before the new segment, and segment.. finds the part of the tree after it. (The latter would start with the segment itself, except I haven’t inserted it yet, so it’s not actually there.)

Then the standard next() and last() methods on bidirectional iterators find me the element I actually want. But the iterator might be empty, so they both return an Option. Also, iterators tend to return references to their contents, but in this case the contents are already references, and I don’t want a double reference, so the map call dereferences one layer — but only if the Option contains a value. Phew!

This is slightly less efficient than the C++ code, since it has to look up where segment goes three times rather than just one. I might be able to get it down to two with some more clever finagling of the iterator, but microsopic performance considerations were a low priority here.

Finally, the event queue uses a std::priority_queue to keep events in a desired order and efficiently pop the next one off the top.

Except priority queues act like heaps, where the greatest (i.e., last) item is made accessible.

Sorting out sorting

C++ comparison functions return true to indicate that the first argument is less than the second argument. Sweep events occur from left to right. You generally implement sorts so that the first thing comes, erm, first.

But sweep events go in a priority queue, and priority queues surface the last item, not the first. This C++ code handled this minor wrinkle by implementing its comparison backwards.

struct SweepEventComp : public std::binary_function<SweepEvent, SweepEvent, bool> { // for sorting sweep events
// Compare two sweep events
// Return true means that e1 is placed at the event queue after e2, i.e,, e1 is processed by the algorithm after e2
bool operator() (const SweepEvent* e1, const SweepEvent* e2)
    if (e1->point.x () > e2->point.x ()) // Different x-coordinate
        return true;
    if (e2->point.x () > e1->point.x ()) // Different x-coordinate
        return false;
    if (e1->point.y () != e2->point.y ()) // Different points, but same x-coordinate. The event with lower y-coordinate is processed first
        return e1->point.y () > e2->point.y ();
    if (e1->left != e2->left) // Same point, but one is a left endpoint and the other a right endpoint. The right endpoint is processed first
        return e1->left;
    // Same point, both events are left endpoints or both are right endpoints.
    if (signedArea (e1->point, e1->otherEvent->point, e2->otherEvent->point) != 0) // not collinear
        return e1->above (e2->otherEvent->point); // the event associate to the bottom segment is processed first
    return e1->pol > e2->pol;

Maybe it’s just me, but I had a hell of a time just figuring out what problem this was even trying to solve. I still have to reread it several times whenever I look at it, to make sure I’m getting the right things backwards.

Making this even more ridiculous is that there’s a second implementation of this same sort, with the same name, in another file — and that one’s implemented forwards. And doesn’t use a tiebreaker. I don’t entirely understand how this even compiles, but it does!

I painstakingly translated this forwards to Rust. Unlike the STL, Rust doesn’t take custom comparators for its containers, so I had to implement ordering on the types themselves (which makes sense, anyway). I wrapped everything in the priority queue in a Reverse, which does what it sounds like.

I’m fairly pleased with Rust’s ordering model. Most of the work is done in Ord, a trait with a cmp() method returning an Ordering (one of Less, Equal, and Greater). No magic numbers, no need to implement all six ordering methods! It’s incredible. Ordering even has some handy methods on it, so the usual case of “order by this, then by this” can be written as:

return self.point().x.cmp(&other.point().x)

Well. Just kidding! It’s not quite that easy. You see, the points here are composed of floats, and floats have the fun property that not all of them are comparable. Specifically, NaN is not less than, greater than, or equal to anything else, including itself. So IEEE 754 float ordering cannot be expressed with Ord. Unless you want to just make up an answer for NaN, but Rust doesn’t tend to do that.

Rust’s float types thus implement the weaker PartialOrd, whose method returns an Option<Ordering> instead. That makes the above example slightly uglier:

return self.point().x.partial_cmp(&other.point().x).unwrap()

Also, since I use unwrap() here, this code will panic and take the whole program down if the points are infinite or NaN. Don’t do that.

This caused some minor inconveniences in other places; for example, the general-purpose cmp::min() doesn’t work on floats, because it requires an Ord-erable type. Thankfully there’s a f64::min(), which handles a NaN by returning the other argument.

(Cool story: for the longest time I had this code using f32s. I’m used to translating int to “32 bits”, and apparently that instinct kicked in for floats as well, even floats spelled double.)

The only other sorting adventure was this:

// Due to overlapping edges the resultEvents array can be not wholly sorted
bool sorted = false;
while (!sorted) {
    sorted = true;
    for (unsigned int i = 0; i < resultEvents.size (); ++i) {
        if (i + 1 < resultEvents.size () && sec (resultEvents[i], resultEvents[i+1])) {
            std::swap (resultEvents[i], resultEvents[i+1]);
            sorted = false;

(I originally misread this comment as saying “the array cannot be wholly sorted” and had no idea why that would be the case, or why the author would then immediately attempt to bubble sort it.)

I’m still not sure why this uses an ad-hoc sort instead of std::sort. But I’m used to taking for granted that general-purpose sorting implementations are tuned to work well for almost-sorted data, like Python’s. Maybe C++ is untrustworthy here, for some reason. I replaced it with a call to .sort() and all seemed fine.

Phew! We’re getting there. Finally, my code appears to type-check.

But now I see storm clouds gathering on the horizon.

Ownership hell

I have a problem. I somehow run into this problem every single time I use Rust. The solutions are never especially satisfying, and all the hacks I might use if forced to write C++ turn out to be unsound, which is even more annoying because rustc is just sitting there with this smug “I told you so expression” and—

The problem is ownership, which Rust is fundamentally built on. Any given value must have exactly one owner, and Rust must be able to statically convince itself that:

  1. No reference to a value outlives that value.
  2. If a mutable reference to a value exists, no other references to that value exist at the same time.

This is the core of Rust. It guarantees at compile time that you cannot lose pointers to allocated memory, you cannot double-free, you cannot have dangling pointers.

It also completely thwarts a lot of approaches you might be inclined to take if you come from managed languages (where who cares, the GC will take care of it) or C++ (where you just throw pointers everywhere and hope for the best apparently).

For example, pointer loops are impossible. Rust’s understanding of ownership and lifetimes is hierarchical, and it simply cannot express loops. (Rust’s own doubly-linked list type uses raw pointers and unsafe code under the hood, where “unsafe” is an escape hatch for the usual ownership rules. Since I only recently realized that pointers to the inside of a mutable Vec are a bad idea, I figure I should probably not be writing unsafe code myself.)

This throws a few wrenches in the works.

Problem the first: pointer loops

I immediately ran into trouble with the SweepEvent struct itself. A SweepEvent pulls double duty: it represents one endpoint of a segment, but each left endpoint also handles bookkeeping for the segment itself — which means that most of the fields on a right endpoint are unused. Also, and more importantly, each SweepEvent has a pointer to the corresponding SweepEvent at the other end of the same segment. So a pair of SweepEvents point to each other.

Rust frowns upon this. In retrospect, I think I could’ve kept it working, but I also think I’m wrong about that.

My first step was to wrench SweepEvent apart. I moved all of the segment-stuff (which is virtually all of it) into a single SweepSegment type, and then populated the event queue with a SweepEndpoint tuple struct, similar to:

enum SegmentEnd {

struct SweepEndpoint<'a>(&'a SweepSegment, SegmentEnd);

This makes SweepEndpoint essentially a tuple with a name. The 'a is a lifetime and says, more or less, that a SweepEndpoint cannot outlive the SweepSegment it references. Makes sense.

Problem solved! I no longer have mutually referential pointers. But I do still have pointers (well, references), and they have to point to something.

Problem the second: where’s all the data

Which brings me to the problem I always run into with Rust. I have a bucket of things, and I need to refer to some of them multiple times.

I tried half a dozen different approaches here and don’t clearly remember all of them, but I think my core problem went as follows. I translated the C++ class to a Rust struct with some methods hanging off of it. A simplified version might look like this.

struct Algorithm {
    arena: LinkedList<SweepSegment>,
    event_queue: BinaryHeap<SweepEndpoint>,

Ah, hang on — SweepEndpoint needs to be annotated with a lifetime, so Rust can enforce that those endpoints don’t live longer than the segments they refer to. No problem?

struct Algorithm<'a> {
    arena: LinkedList<SweepSegment>,
    event_queue: BinaryHeap<SweepEndpoint<'a>>,

Okay! Now for some methods.

fn run(&mut self) {
    self.arena.push_back(SweepSegment{ data: 5 });
    self.event_queue.push(SweepEndpoint(self.arena.back().unwrap(), SegmentEnd::Left));
    self.event_queue.push(SweepEndpoint(self.arena.back().unwrap(), SegmentEnd::Right));
    for event in &self.event_queue {
        println!("{:?}", event)

Aaand… this doesn’t work. Rust “cannot infer an appropriate lifetime for autoref due to conflicting requirements”. The trouble is that self.arena.back() takes a reference to self.arena, and then I put that reference in the event queue. But I promised that everything in the event queue has lifetime 'a, and I don’t actually know how long self lives here; I only know that it can’t outlive 'a, because that would invalidate the references it holds.

A little random guessing let me to change &mut self to &'a mut self — which is fine because the entire impl block this lives in is already parameterized by 'a — and that makes this compile! Hooray! I think that’s because I’m saying self itself has exactly the same lifetime as the references it holds onto, which is true, since it’s referring to itself.

Let’s get a little more ambitious and try having two segments.

fn run(&'a mut self) {
    self.arena.push_back(SweepSegment{ data: 5 });
    self.event_queue.push(SweepEndpoint(self.arena.back().unwrap(), SegmentEnd::Left));
    self.event_queue.push(SweepEndpoint(self.arena.back().unwrap(), SegmentEnd::Right));
    self.arena.push_back(SweepSegment{ data: 17 });
    self.event_queue.push(SweepEndpoint(self.arena.back().unwrap(), SegmentEnd::Left));
    self.event_queue.push(SweepEndpoint(self.arena.back().unwrap(), SegmentEnd::Right));
    for event in &self.event_queue {
        println!("{:?}", event)

Whoops! Rust complains that I’m trying to mutate self.arena while other stuff is referring to it. And, yes, that’s true — I have references to it in the event queue, and Rust is preventing me from potentially deleting everything from the queue when references to it still exist. I’m not actually deleting anything here, of course (though I could be if this were a Vec!), but Rust’s type system can’t encode that (and I dread the thought of a type system that can).

I struggled with this for a while, and rapidly encountered another complete showstopper:

fn run(&'a mut self) {

fn mutate_something(&'a mut self) {}

Rust objects that I’m trying to borrow self mutably, twice — once for the first call, once for the second.

But why? A borrow is supposed to end automatically once it’s no longer used, right? Maybe if I throw some braces around it for scope… nope, that doesn’t help either.

It’s true that borrows usually end automatically, but here I have explicitly told Rust that mutate_something() should borrow with the lifetime 'a, which is the same as the lifetime in run(). So the first call explicitly borrows self for at least the rest of the method. Removing the lifetime from mutate_something() does fix this error, but if that method tries to add new segments, I’m back to the original problem.

Oh no. The mutation in the C++ code is several calls deep. Porting it directly seems nearly impossible.

The typical solution here — at least, the first thing people suggest to me on Twitter — is to wrap basically everything everywhere in Rc<RefCell<T>>, which gives you something that’s reference-counted (avoiding questions of ownership) and defers borrow checks until runtime (avoiding questions of mutable borrows). But that seems pretty heavy-handed here — not only does RefCell add .borrow() noise anywhere you actually want to interact with the underlying value, but do I really need to refcount these tiny structs that only hold a handful of floats each?

I set out to find a middle ground.

Solution, kind of

I really, really didn’t want to perform serious surgery on this code just to get it to build. I still didn’t know if it worked at all, and now I had to rearrange it without being able to check if I was breaking it further. (This isn’t Rust’s fault; it’s a natural problem with porting between fairly different paradigms.)

So I kind of hacked it into working with minimal changes, producing a grotesque abomination which I’m ashamed to link to. Here’s how!

First, I got rid of the class. It turns out this makes lifetime juggling much easier right off the bat. I’m pretty sure Rust considers everything in a struct to be destroyed simultaneously (though in practice it guarantees it’ll destroy fields in order), which doesn’t leave much wiggle room. Locals within a function, on the other hand, can each have their own distinct lifetimes, which solves the problem of expressing that the borrows won’t outlive the arena.

Speaking of the arena, I solved the mutability problem there by switching to… an arena! The typed-arena crate (a port of a type used within Rust itself, I think) is an allocator — you give it a value, and it gives you back a reference, and the reference is guaranteed to be valid for as long as the arena exists. The method that does this is sneaky and takes &self rather than &mut self, so Rust doesn’t know you’re mutating the arena and won’t complain. (One drawback is that the arena will never free anything you give to it, but that’s not a big problem here.)

My next problem was with mutation. The main loop repeatedly calls possibleIntersection with pairs of segments, which can split either or both segment. Rust definitely doesn’t like that — I’d have to pass in two &muts, both of which are mutable references into the same arena, and I’d have a bunch of immutable references into that arena in the sweep list and elsewhere. This isn’t going to fly.

This is kind of a shame, and is one place where Rust seems a little overzealous. Something like this seems like it ought to be perfectly valid:

let mut v = vec![1u32, 2u32];
let a = &mut v[0];
let b = &mut v[1];
// do stuff with a, b

The trouble is, Rust only knows the type signature, which here is something like index_mut(&'a mut self, index: usize) -> &'a T. Nothing about that says that you’re borrowing distinct elements rather than some core part of the type — and, in fact, the above code is only safe because you’re borrowing distinct elements. In the general case, Rust can’t possibly know that. It seems obvious enough from the different indexes, but nothing about the type system even says that different indexes have to return different values. And what if one were borrowed as &mut v[1] and the other were borrowed with v.iter_mut().next().unwrap()?

Anyway, this is exactly where people start to turn to RefCell — if you’re very sure you know better than Rust, then a RefCell will skirt the borrow checker while still enforcing at runtime that you don’t have more than one mutable borrow at a time.

But half the lines in this algorithm examine the endpoints of a segment! I don’t want to wrap the whole thing in a RefCell, or I’ll have to say this everywhere:

if segment1.borrow().point.x < segment2.borrow().point.x { ... }


But wait — this code only mutates the points themselves in one place. When a segment is split, the original segment becomes the left half, and a new segment is created to be the right half. There’s no compelling need for this; it saves an allocation for the left half, but it’s not critical to the algorithm.

Thus, I settled on a compromise. My segment type now looks like this:

struct SegmentPacket {
    // a bunch of flags and whatnot used in the algorithm
struct SweepSegment {
    left_point: MapPoint,
    right_point: MapPoint,
    faces_outwards: bool,
    index: usize,
    order: usize,
    packet: RefCell<SegmentPacket>,

I do still need to call .borrow() or .borrow_mut() to get at the stuff in the “packet”, but that’s far less common, so there’s less noise overall. And I don’t need to wrap it in Rc because it’s part of a type that’s allocated in the arena and passed around only via references.

This still leaves me with the problem of how to actually perform the splits.

I’m not especially happy with what I came up with, I don’t know if I can defend it, and I suspect I could do much better. I changed possibleIntersection so that rather than performing splits, it returns the points at which each segment needs splitting, in the form (usize, Option<MapPoint>, Option<MapPoint>). (The usize is used as a flag for calling code and oughta be an enum, but, isn’t yet.)

Now the top-level function is responsible for all arena management, and all is well.

Except, er. possibleIntersection is called multiple times, and I don’t want to copy-paste a dozen lines of split code after each call. I tried putting just that code in its own function, which had the world’s most godawful signature, and that didn’t work because… uh… hm. I can’t remember why, exactly! Should’ve written that down.

I tried a local closure next, but closures capture their environment by reference, so now I had references to a bunch of locals for as long as the closure existed, which meant I couldn’t mutate those locals. Argh. (This seems a little silly to me, since the closure’s references cannot possibly be used for anything if the closure isn’t being called, but maybe I’m missing something. Or maybe this is just a limitation of lifetimes.)

Increasingly desperate, I tried using a macro. But… macros are hygienic, which means that any new name you use inside a macro is different from any name outside that macro. The macro thus could not see any of my locals. Usually that’s good, but here I explicitly wanted the macro to mess with my locals.

I was just about to give up and go live as a hermit in a cabin in the woods, when I discovered something quite incredible. You can define local macros! If you define a macro inside a function, then it can see any locals defined earlier in that function. Perfect!

macro_rules! _split_segment (
    ($seg:expr, $pt:expr) => (
            let pt = $pt;
            let seg = $seg;
            // ... waaay too much code ...

loop {
    // ...
    // This is possibleIntersection, renamed because Rust rightfully complains about camelCase
    let cross = handle_intersections(Some(segment), maybe_above);
    if let Some(pt) = cross.1 {
        segment = _split_segment!(segment, pt);
    if let Some(pt) = cross.2 {
        maybe_above = Some(_split_segment!(maybe_above.unwrap(), pt));
    // ...

(This doesn’t actually quite match the original algorithm, which has one case where a segment can be split twice. I realized that I could just do the left-most split, and a later iteration would perform the other split. I sure hope that’s right, anyway.)

It’s a bit ugly, and I ran into a whole lot of implicit behavior from the C++ code that I had to fix — for example, the segment is sometimes mutated just before it’s split, purely as a shortcut for mutating the left part of the split. But it finally compiles! And runs! And kinda worked, a bit!


I still had a lot of work to do.

For one, this code was designed for intersecting two shapes, not mass-intersecting a big pile of shapes. The basic algorithm doesn’t care about how many polygons you start with — all it sees is segments — but the code for constructing the return value needed some heavy modification.

The biggest change by far? The original code traced each segment once, expecting the result to be only a single shape. I had to change that to trace each side of each segment once, since the vast bulk of the output consists of shapes which share a side. This violated a few assumptions, which I had to hack around.

I also ran into a couple very bad edge cases, spent ages debugging them, then found out that the original algorithm had a subtle workaround that I’d commented out because it was awkward to port but didn’t seem to do anything. Whoops!

The worst was a precision error, where a vertical line could be split on a point not quite actually on the line, which wreaked all kinds of havoc. I worked around that with some tasteful rounding, which is highly dubious but makes the output more appealing to my squishy human brain. (I might switch to the original workaround, but I really dislike that even simple cases can spit out points at 1500.0000000000003. The whole thing is parameterized over the coordinate type, so maybe I could throw a rational type in there and cross my fingers?)

All that done, I finally, finally, after a couple months of intermittent progress, got what I wanted!

This is Doom 2’s MAP01. The black area to the left of center is where the player starts. Gray areas indicate where the player can walk from there, with lighter shades indicating more distant areas, where “distance” is measured by the minimum number of line crossings. Red areas can’t be reached at all.

(Note: large playable chunks of the map, including the exit room, are red. That’s because those areas are behind doors, and this code doesn’t understand doors yet.)

(Also note: The big crescent in the lower-right is also black because I was lazy and looked for the player’s starting sector by checking the bbox, and that sector’s bbox happens to match.)

The code that generated this had to go out of its way to delete all the unreachable zones around solid walls. I think I could modify the algorithm to do that on the fly pretty easily, which would probably speed it up a bit too. Downside is that the algorithm would then be pretty specifically tied to this problem, and not usable for any other kind of polygon intersection, which I would think could come up elsewhere? The modifications would be pretty minor, though, so maybe I could confine them to a closure or something.

Some final observations

It runs surprisingly slowly. Like, multiple seconds. Unless I add --release, which speeds it up by a factor of… some number with multiple digits. Wahoo. Debug mode has a high price, especially with a lot of calls in play.

The current state of this code is on GitHub. Please don’t look at it. I’m very sorry.

Honestly, most of my anguish came not from Rust, but from the original code relying on lots of fairly subtle behavior without bothering to explain what it was doing or even hint that anything unusual was going on. God, I hate C++.

I don’t know if the Rust community can learn from this. I don’t know if I even learned from this. Let’s all just quietly forget about it.

Now I just need to figure this one out…

Security of Cloud HSMBackups

Post Syndicated from Balaji Iyer original https://aws.amazon.com/blogs/architecture/security-of-cloud-hsmbackups/

Today, our customers use AWS CloudHSM to meet corporate, contractual and regulatory compliance requirements for data security by using dedicated Hardware Security Module (HSM) instances within the AWS cloud. CloudHSM delivers all the benefits of traditional HSMs including secure generation, storage, and management of cryptographic keys used for data encryption that are controlled and accessible only by you.

As a managed service, it automates time-consuming administrative tasks such as hardware provisioning, software patching, high availability, backups and scaling for your sensitive and regulated workloads in a cost-effective manner. Backup and restore functionality is the core building block enabling scalability, reliability and high availability in CloudHSM.

You should consider using AWS CloudHSM if you require:

  • Keys stored in dedicated, third-party validated hardware security modules under your exclusive control
  • FIPS 140-2 compliance
  • Integration with applications using PKCS#11, Java JCE, or Microsoft CNG interfaces
  • High-performance in-VPC cryptographic acceleration (bulk crypto)
  • Financial applications subject to PCI regulations
  • Healthcare applications subject to HIPAA regulations
  • Streaming video solutions subject to contractual DRM requirements

We recently released a whitepaper, “Security of CloudHSM Backups” that provides in-depth information on how backups are protected in all three phases of the CloudHSM backup lifecycle process: Creation, Archive, and Restore.

About the Author

Balaji Iyer is a senior consultant in 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, operational security, large scale migrations, and leading strategic AWS initiatives.

SoFi, the underwater robotic fish

Post Syndicated from Alex Bate original https://www.raspberrypi.org/blog/robotic-fish/

With the Greenland shark finally caught on video for the very first time, scientists and engineers are discussing the limitations of current marine monitoring technology. One significant advance comes from the CSAIL team at Massachusetts Institute of Technology (MIT): SoFi, the robotic fish.

A Robotic Fish Swims in the Ocean

More info: http://bit.ly/SoFiRobot Paper: http://robert.katzschmann.eu/wp-content/uploads/2018/03/katzschmann2018exploration.pdf

The untethered SoFi robot

Last week, the Computer Science and Artificial Intelligence Laboratory (CSAIL) team at MIT unveiled SoFi, “a soft robotic fish that can independently swim alongside real fish in the ocean.”

MIT CSAIL underwater fish SoFi using Raspberry Pi

Directed by a Super Nintendo controller and acoustic signals, SoFi can dive untethered to a maximum of 18 feet for a total of 40 minutes. A Raspberry Pi receives input from the controller and amplifies the ultrasound signals for SoFi via a HiFiBerry. The controller, Raspberry Pi, and HiFiBerry are sealed within a waterproof, cast-moulded silicone membrane filled with non-conductive mineral oil, allowing for underwater equalisation.

MIT CSAIL underwater fish SoFi using Raspberry Pi

The ultrasound signals, received by a modem within SoFi’s head, control everything from direction, tail oscillation, pitch, and depth to the onboard camera.

As explained on MIT’s news blog, “to make the robot swim, the motor pumps water into two balloon-like chambers in the fish’s tail that operate like a set of pistons in an engine. As one chamber expands, it bends and flexes to one side; when the actuators push water to the other channel, that one bends and flexes in the other direction.”

MIT CSAIL underwater fish SoFi using Raspberry Pi

Ocean exploration

While we’ve seen many autonomous underwater vehicles (AUVs) using onboard Raspberry Pis, SoFi’s ability to roam untethered with a wireless waterproof controller is an exciting achievement.

“To our knowledge, this is the first robotic fish that can swim untethered in three dimensions for extended periods of time. We are excited about the possibility of being able to use a system like this to get closer to marine life than humans can get on their own.” – CSAIL PhD candidate Robert Katzschmann

As the MIT news post notes, SoFi’s simple, lightweight setup of a single camera, a motor, and a smartphone lithium polymer battery set it apart it from existing bulky AUVs that require large motors or support from boats.

For more in-depth information on SoFi and the onboard tech that controls it, find the CSAIL team’s paper here.

The post SoFi, the underwater robotic fish appeared first on Raspberry Pi.

Gettys: The Blind Men and the Elephant

Post Syndicated from corbet original https://lwn.net/Articles/747084/rss

Jim Gettys provides
an extensive look at the FQ_CoDel queue-management algorithm
as a big
piece of the solution to bufferbloat problems. “Simple
‘request/response’ or time based protocols are preferentially scheduled
relative to bulk data transport. This means that your VOIP packets, your
TCP handshakes, cryptographic associations, your button press in your game,
your DHCP or other basic network protocols all get preferential service
without the complexity of extensive packet classification, even under very
heavy load of other ongoing flows. Your phone call can work well despite
large downloads or video use.

Give Your WordPress Blog a Voice With Our New Amazon Polly Plugin

Post Syndicated from Jeff Barr original https://aws.amazon.com/blogs/aws/give-your-wordpress-blog-a-voice-with-our-new-amazon-polly-plugin/

I first told you about Polly in late 2016 in my post Amazon Polly – Text to Speech in 47 Voices and 24 Languages. After that AWS re:Invent launch, we added support for Korean, five new voices, and made Polly available in all Regions in the aws partition. We also added whispering, speech marks, a timbre effect, and dynamic range compression.

New WordPress Plugin
Today we are launching a WordPress plugin that uses Polly to create high-quality audio versions of your blog posts. You can access the audio from within the post or in podcast form using a feature that we call Amazon Pollycast! Both options make your content more accessible and can help you to reach a wider audience. This plugin was a joint effort between the AWS team our friends at AWS Advanced Technology Partner WP Engine.

As you will see, the plugin is easy to install and configure. You can use it with installations of WordPress that you run on your own infrastructure or on AWS. Either way, you have access to all of Polly’s voices along with a wide variety of configuration options. The generated audio (an MP3 file for each post) can be stored alongside your WordPress content, or in Amazon Simple Storage Service (S3), with optional support for content distribution via Amazon CloudFront.

Installing the Plugin
I did not have an existing WordPress-powered blog, so I begin by launching a Lightsail instance using the WordPress 4.8.1 blueprint:

Then I follow these directions to access my login credentials:

Credentials in hand, I log in to the WordPress Dashboard:

The plugin makes calls to AWS, and needs to have credentials in order to do so. I hop over to the IAM Console and created a new policy. The policy allows the plugin to access a carefully selected set of S3 and Polly functions (find the full policy in the README):

Then I create an IAM user (wp-polly-user). I enter the name and indicate that it will be used for Programmatic Access:

Then I attach the policy that I just created, and click on Review:

I review my settings (not shown) and then click on Create User. Then I copy the two values (Access Key ID and Secret Access Key) into a secure location. Possession of these keys allows the bearer to make calls to AWS so I take care not to leave them lying around.

Now I am ready to install the plugin! I go back to the WordPress Dashboard and click on Add New in the Plugins menu:

Then I click on Upload Plugin and locate the ZIP file that I downloaded from the WordPress Plugins site. After I find it I click on Install Now to proceed:

WordPress uploads and installs the plugin. Now I click on Activate Plugin to move ahead:

With the plugin installed, I click on Settings to set it up:

I enter my keys and click on Save Changes:

The General settings let me control the sample rate, voice, player position, the default setting for new posts, and the autoplay option. I can leave all of the settings as-is to get started:

The Cloud Storage settings let me store audio in S3 and to use CloudFront to distribute the audio:

The Amazon Pollycast settings give me control over the iTunes parameters that are included in the generated RSS feed:

Finally, the Bulk Update button lets me regenerate all of the audio files after I change any of the other settings:

With the plugin installed and configured, I can create a new post. As you can see, the plugin can be enabled and customized for each post:

I can see how much it will cost to convert to audio with a click:

When I click on Publish, the plugin breaks the text into multiple blocks on sentence boundaries, calls the Polly SynthesizeSpeech API for each block, and accumulates the resulting audio in a single MP3 file. The published blog post references the file using the <audio> tag. Here’s the post:

I can’t seem to use an <audio> tag in this post, but you can download and play the MP3 file yourself if you’d like.

The Pollycast feature generates an RSS file with links to an MP3 file for each post:

The plugin will make calls to Amazon Polly each time the post is saved or updated. Pricing is based on the number of characters in the speech requests, as described on the Polly Pricing page. Also, the AWS Free Tier lets you process up to 5 million characters per month at no charge, for a period of one year that starts when you make your first call to Polly.

Going Further
The plugin is available on GitHub in source code form and we are looking forward to your pull requests! Here are a couple of ideas to get you started:

Voice Per Author – Allow selection of a distinct Polly voice for each author.

Quoted Text – For blogs that make frequent use of embedded quotes, use a distinct voice for the quotes.

Translation – Use Amazon Translate to translate the texts into another language, and then use Polly to generate audio in that language.

Other Blogging Engines – Build a similar plugin for your favorite blogging engine.

SSML Support – Figure out an interesting way to use Polly’s SSML tags to add additional character to the audio.

Let me know what you come up with!



Backblaze Hard Drive Stats for 2017

Post Syndicated from Andy Klein original https://www.backblaze.com/blog/hard-drive-stats-for-2017/

Backbalze Drive Stats 2017 Review

Beginning in April 2013, Backblaze has recorded and saved daily hard drive statistics from the drives in our data centers. Each entry consists of the date, manufacturer, model, serial number, status (operational or failed), and all of the SMART attributes reported by that drive. As of the end of 2017, there are about 88 million entries totaling 23 GB of data. You can download this data from our website if you want to do your own research, but for starters here’s what we found.


At the end of 2017 we had 93,240 spinning hard drives. Of that number, there were 1,935 boot drives and 91,305 data drives. This post looks at the hard drive statistics of the data drives we monitor. We’ll review the stats for Q4 2017, all of 2017, and the lifetime statistics for all of the drives Backblaze has used in our cloud storage data centers since we started keeping track. Along the way we’ll share observations and insights on the data presented and we look forward to you doing the same in the comments.

Hard Drive Reliability Statistics for Q4 2017

At the end of Q4 2017 Backblaze was monitoring 91,305 hard drives used to store data. For our evaluation we remove from consideration those drives which were used for testing purposes and those drive models for which we did not have at least 45 drives (read why after the chart). This leaves us with 91,243 hard drives. The table below is for the period of Q4 2017.

Hard Drive Annualized Failure Rates for Q4 2017

A few things to remember when viewing this chart:

  • The failure rate listed is for just Q4 2017. If a drive model has a failure rate of 0%, it means there were no drive failures of that model during Q4 2017.
  • There were 62 drives (91,305 minus 91,243) that were not included in the list above because we did not have at least 45 of a given drive model. The most common reason we would have fewer than 45 drives of one model is that we needed to replace a failed drive and we had to purchase a different model as a replacement because the original model was no longer available. We use 45 drives of the same model as the minimum number to qualify for reporting quarterly, yearly, and lifetime drive statistics.
  • Quarterly failure rates can be volatile, especially for models that have a small number of drives and/or a small number of drive days. For example, the Seagate 4 TB drive, model ST4000DM005, has a annualized failure rate of 29.08%, but that is based on only 1,255 drive days and 1 (one) drive failure.
  • AFR stands for Annualized Failure Rate, which is the projected failure rate for a year based on the data from this quarter only.

Bulking Up and Adding On Storage

Looking back over 2017, we not only added new drives, we “bulked up” by swapping out functional and smaller 2, 3, and 4TB drives with larger 8, 10, and 12TB drives. The changes in drive quantity by quarter are shown in the chart below:

Backblaze Drive Population by Drive Size

For 2017 we added 25,746 new drives, and lost 6,442 drives to retirement for a net of 19,304 drives. When you look at storage space, we added 230 petabytes and retired 19 petabytes, netting us an additional 211 petabytes of storage in our data center in 2017.

2017 Hard Drive Failure Stats

Below are the lifetime hard drive failure statistics for the hard drive models that were operational at the end of Q4 2017. As with the quarterly results above, we have removed any non-production drives and any models that had fewer than 45 drives.

Hard Drive Annualized Failure Rates

The chart above gives us the lifetime view of the various drive models in our data center. The Q4 2017 chart at the beginning of the post gives us a snapshot of the most recent quarter of the same models.

Let’s take a look at the same models over time, in our case over the past 3 years (2015 through 2017), by looking at the annual failure rates for each of those years.

Annual Hard Drive Failure Rates by Year

The failure rate for each year is calculated for just that year. In looking at the results the following observations can be made:

  • The failure rates for both of the 6 TB models, Seagate and WDC, have decreased over the years while the number of drives has stayed fairly consistent from year to year.
  • While it looks like the failure rates for the 3 TB WDC drives have also decreased, you’ll notice that we migrated out nearly 1,000 of these WDC drives in 2017. While the remaining 180 WDC 3 TB drives are performing very well, decreasing the data set that dramatically makes trend analysis suspect.
  • The Toshiba 5 TB model and the HGST 8 TB model had zero failures over the last year. That’s impressive, but with only 45 drives in use for each model, not statistically useful.
  • The HGST/Hitachi 4 TB models delivered sub 1.0% failure rates for each of the three years. Amazing.

A Few More Numbers

To save you countless hours of looking, we’ve culled through the data to uncover the following tidbits regarding our ever changing hard drive farm.

  • 116,833 — The number of hard drives for which we have data from April 2013 through the end of December 2017. Currently there are 91,305 drives (data drives) in operation. This means 25,528 drives have either failed or been removed from service due for some other reason — typically migration.
  • 29,844 — The number of hard drives that were installed in 2017. This includes new drives, migrations, and failure replacements.
  • 81.76 — The number of hard drives that were installed each day in 2017. This includes new drives, migrations, and failure replacements.
  • 95,638 — The number of drives installed since we started keeping records in April 2013 through the end of December 2017.
  • 55.41 — The average number of hard drives installed per day from April 2013 to the end of December 2017. The installations can be new drives, migration replacements, or failure replacements.
  • 1,508 — The number of hard drives that were replaced as failed in 2017.
  • 4.13 — The average number of hard drives that have failed each day in 2017.
  • 6,795 — The number of hard drives that have failed from April 2013 until the end of December 2017.
  • 3.94 — The average number of hard drives that have failed each day from April 2013 until the end of December 2017.

Can’t Get Enough Hard Drive Stats?

We’ll be presenting the webinar “Backblaze Hard Drive Stats for 2017” on Thursday February 9, 2017 at 10:00 Pacific time. The webinar will dig deeper into the quarterly, yearly, and lifetime hard drive stats and include the annual and lifetime stats by drive size and manufacturer. You will need to subscribe to the Backblaze BrightTALK channel to view the webinar. Sign up today.

As a reminder, the complete data set used to create the information used in this review is available on our Hard Drive Test Data page. You can download and use this data for free for your own purpose. All we ask are three things: 1) you cite Backblaze as the source if you use the data, 2) you accept that you are solely responsible for how you use the data, and 3) you do not sell this data to anyone — it is free.

Good luck and let us know if you find anything interesting.

The post Backblaze Hard Drive Stats for 2017 appeared first on Backblaze Blog | Cloud Storage & Cloud Backup.

Top 8 Best Practices for High-Performance ETL Processing Using Amazon Redshift

Post Syndicated from Thiyagarajan Arumugam original https://aws.amazon.com/blogs/big-data/top-8-best-practices-for-high-performance-etl-processing-using-amazon-redshift/

An ETL (Extract, Transform, Load) process enables you to load data from source systems into your data warehouse. This is typically executed as a batch or near-real-time ingest process to keep the data warehouse current and provide up-to-date analytical data to end users.

Amazon Redshift is a fast, petabyte-scale data warehouse that enables you easily to make data-driven decisions. With Amazon Redshift, you can get insights into your big data in a cost-effective fashion using standard SQL. You can set up any type of data model, from star and snowflake schemas, to simple de-normalized tables for running any analytical queries.

To operate a robust ETL platform and deliver data to Amazon Redshift in a timely manner, design your ETL processes to take account of Amazon Redshift’s architecture. When migrating from a legacy data warehouse to Amazon Redshift, it is tempting to adopt a lift-and-shift approach, but this can result in performance and scale issues long term. This post guides you through the following best practices for ensuring optimal, consistent runtimes for your ETL processes:

  • COPY data from multiple, evenly sized files.
  • Use workload management to improve ETL runtimes.
  • Perform table maintenance regularly.
  • Perform multiple steps in a single transaction.
  • Loading data in bulk.
  • Use UNLOAD to extract large result sets.
  • Use Amazon Redshift Spectrum for ad hoc ETL processing.
  • Monitor daily ETL health using diagnostic queries.

1. COPY data from multiple, evenly sized files

Amazon Redshift is an MPP (massively parallel processing) database, where all the compute nodes divide and parallelize the work of ingesting data. Each node is further subdivided into slices, with each slice having one or more dedicated cores, equally dividing the processing capacity. The number of slices per node depends on the node type of the cluster. For example, each DS2.XLARGE compute node has two slices, whereas each DS2.8XLARGE compute node has 16 slices.

When you load data into Amazon Redshift, you should aim to have each slice do an equal amount of work. When you load the data from a single large file or from files split into uneven sizes, some slices do more work than others. As a result, the process runs only as fast as the slowest, or most heavily loaded, slice. In the example shown below, a single large file is loaded into a two-node cluster, resulting in only one of the nodes, “Compute-0”, performing all the data ingestion:

When splitting your data files, ensure that they are of approximately equal size – between 1 MB and 1 GB after compression. The number of files should be a multiple of the number of slices in your cluster. Also, I strongly recommend that you individually compress the load files using gzip, lzop, or bzip2 to efficiently load large datasets.

When loading multiple files into a single table, use a single COPY command for the table, rather than multiple COPY commands. Amazon Redshift automatically parallelizes the data ingestion. Using a single COPY command to bulk load data into a table ensures optimal use of cluster resources, and quickest possible throughput.

2. Use workload management to improve ETL runtimes

Use Amazon Redshift’s workload management (WLM) to define multiple queues dedicated to different workloads (for example, ETL versus reporting) and to manage the runtimes of queries. As you migrate more workloads into Amazon Redshift, your ETL runtimes can become inconsistent if WLM is not appropriately set up.

I recommend limiting the overall concurrency of WLM across all queues to around 15 or less. This WLM guide helps you organize and monitor the different queues for your Amazon Redshift cluster.

When managing different workloads on your Amazon Redshift cluster, consider the following for the queue setup:

  • Create a queue dedicated to your ETL processes. Configure this queue with a small number of slots (5 or fewer). Amazon Redshift is designed for analytics queries, rather than transaction processing. The cost of COMMIT is relatively high, and excessive use of COMMIT can result in queries waiting for access to the commit queue. Because ETL is a commit-intensive process, having a separate queue with a small number of slots helps mitigate this issue.
  • Claim extra memory available in a queue. When executing an ETL query, you can take advantage of the wlm_query_slot_count to claim the extra memory available in a particular queue. For example, a typical ETL process might involve COPYing raw data into a staging table so that downstream ETL jobs can run transformations that calculate daily, weekly, and monthly aggregates. To speed up the COPY process (so that the downstream tasks can start in parallel sooner), the wlm_query_slot_count can be increased for this step.
  • Create a separate queue for reporting queries. Configure query monitoring rules on this queue to further manage long-running and expensive queries.
  • Take advantage of the dynamic memory parameters. They swap the memory from your ETL to your reporting queue after the ETL job has completed.

3. Perform table maintenance regularly

Amazon Redshift is a columnar database, which enables fast transformations for aggregating data. Performing regular table maintenance ensures that transformation ETLs are predictable and performant. To get the best performance from your Amazon Redshift database, you must ensure that database tables regularly are VACUUMed and ANALYZEd. The Analyze & Vacuum schema utility helps you automate the table maintenance task and have VACUUM & ANALYZE executed in a regular fashion.

  • Use VACUUM to sort tables and remove deleted blocks

During a typical ETL refresh process, tables receive new incoming records using COPY, and unneeded data (cold data) is removed using DELETE. New rows are added to the unsorted region in a table. Deleted rows are simply marked for deletion.

DELETE does not automatically reclaim the space occupied by the deleted rows. Adding and removing large numbers of rows can therefore cause the unsorted region and the number of deleted blocks to grow. This can degrade the performance of queries executed against these tables.

After an ETL process completes, perform VACUUM to ensure that user queries execute in a consistent manner. The complete list of tables that need VACUUMing can be found using the Amazon Redshift Util’s table_info script.

Use the following approaches to ensure that VACCUM is completed in a timely manner:

  • Use wlm_query_slot_count to claim all the memory allocated in the ETL WLM queue during the VACUUM process.
  • DROP or TRUNCATE intermediate or staging tables, thereby eliminating the need to VACUUM them.
  • If your table has a compound sort key with only one sort column, try to load your data in sort key order. This helps reduce or eliminate the need to VACUUM the table.
  • Consider using time series This helps reduce the amount of data you need to VACUUM.
  • Use ANALYZE to update database statistics

Amazon Redshift uses a cost-based query planner and optimizer using statistics about tables to make good decisions about the query plan for the SQL statements. Regular statistics collection after the ETL completion ensures that user queries run fast, and that daily ETL processes are performant. The Amazon Redshift utility table_info script provides insights into the freshness of the statistics. Keeping the statistics off (pct_stats_off) less than 20% ensures effective query plans for the SQL queries.

4. Perform multiple steps in a single transaction

ETL transformation logic often spans multiple steps. Because commits in Amazon Redshift are expensive, if each ETL step performs a commit, multiple concurrent ETL processes can take a long time to execute.

To minimize the number of commits in a process, the steps in an ETL script should be surrounded by a BEGIN…END statement so that a single commit is performed only after all the transformation logic has been executed. For example, here is an example multi-step ETL script that performs one commit at the end:

CREATE temporary staging_table;
INSERT INTO staging_table SELECT .. FROM source (transformation logic);
DELETE FROM daily_table WHERE dataset_date =?;
INSERT INTO daily_table SELECT .. FROM staging_table (daily aggregate);
DELETE FROM weekly_table WHERE weekending_date=?;
INSERT INTO weekly_table SELECT .. FROM staging_table(weekly aggregate);

5. Loading data in bulk

Amazon Redshift is designed to store and query petabyte-scale datasets. Using Amazon S3 you can stage and accumulate data from multiple source systems before executing a bulk COPY operation. The following methods allow efficient and fast transfer of these bulk datasets into Amazon Redshift:

  • Use a manifest file to ingest large datasets that span multiple files. The manifest file is a JSON file that lists all the files to be loaded into Amazon Redshift. Using a manifest file ensures that Amazon Redshift has a consistent view of the data to be loaded from S3, while also ensuring that duplicate files do not result in the same data being loaded more than one time.
  • Use temporary staging tables to hold the data for transformation. These tables are automatically dropped after the ETL session is complete. Temporary tables can be created using the CREATE TEMPORARY TABLE syntax, or by issuing a SELECT … INTO #TEMP_TABLE query. Explicitly specifying the CREATE TEMPORARY TABLE statement allows you to control the DISTRIBUTION KEY, SORT KEY, and compression settings to further improve performance.
  • User ALTER table APPEND to swap data from the staging tables to the target table. Data in the source table is moved to matching columns in the target table. Column order doesn’t matter. After data is successfully appended to the target table, the source table is empty. ALTER TABLE APPEND is much faster than a similar CREATE TABLE AS or INSERT INTO operation because it doesn’t involve copying or moving data.

6. Use UNLOAD to extract large result sets

Fetching a large number of rows using SELECT is expensive and takes a long time. When a large amount of data is fetched from the Amazon Redshift cluster, the leader node has to hold the data temporarily until the fetches are complete. Further, data is streamed out sequentially, which results in longer elapsed time. As a result, the leader node can become hot, which not only affects the SELECT that is being executed, but also throttles resources for creating execution plans and managing the overall cluster resources. Here is an example of a large SELECT statement. Notice that the leader node is doing most of the work to stream out the rows:

Use UNLOAD to extract large results sets directly to S3. After it’s in S3, the data can be shared with multiple downstream systems. By default, UNLOAD writes data in parallel to multiple files according to the number of slices in the cluster. All the compute nodes participate to quickly offload the data into S3.

If you are extracting data for use with Amazon Redshift Spectrum, you should make use of the MAXFILESIZE parameter to and keep files are 150 MB. Similar to item 1 above, having many evenly sized files ensures that Redshift Spectrum can do the maximum amount of work in parallel.

7. Use Redshift Spectrum for ad hoc ETL processing

Events such as data backfill, promotional activity, and special calendar days can trigger additional data volumes that affect the data refresh times in your Amazon Redshift cluster. To help address these spikes in data volumes and throughput, I recommend staging data in S3. After data is organized in S3, Redshift Spectrum enables you to query it directly using standard SQL. In this way, you gain the benefits of additional capacity without having to resize your cluster.

For tips on getting started with and optimizing the use of Redshift Spectrum, see the previous post, 10 Best Practices for Amazon Redshift Spectrum.

8. Monitor daily ETL health using diagnostic queries

Monitoring the health of your ETL processes on a regular basis helps identify the early onset of performance issues before they have a significant impact on your cluster. The following monitoring scripts can be used to provide insights into the health of your ETL processes:

ScriptUse when…Solution
commit_stats.sql – Commit queue statistics from past days, showing largest queue length and queue time firstDML statements such as INSERT/UPDATE/COPY/DELETE operations take several times longer to execute when multiple of these operations are in progressSet up separate WLM queues for the ETL process and limit the concurrency to < 5.
copy_performance.sql –  Copy command statistics for the past days Daily COPY operations take longer to execute• Follow the best practices for the COPY command.
• Analyze data growth with the incoming datasets and consider cluster resize to meet the expected SLA.
table_info.sql – Table skew and unsorted statistics along with storage and key informationTransformation steps take longer to execute• Set up regular VACCUM jobs to address unsorted rows and claim the deleted blocks so that transformation SQL execute optimally.
• Consider a table redesign to avoid data skewness.
v_check_transaction_locks.sql – Monitor transaction locksINSERT/UPDATE/COPY/DELETE operations on particular tables do not respond back in timely manner, compared to when run after the ETLMultiple DML statements are operating on the same target table at the same moment from different transactions. Set up ETL job dependency so that they execute serially for the same target table.
v_get_schema_priv_by_user.sql – Get the schema that the user has access toReporting users can view intermediate tablesSet up separate database groups for reporting and ETL users, and grants access to objects using GRANT.
v_generate_tbl_ddl.sql – Get the table DDLYou need to create an empty table with same structure as target table for data backfillGenerate DDL using this script for data backfill.
v_space_used_per_tbl.sql – monitor space used by individual tablesAmazon Redshift data warehouse space growth is trending upwards more than normal

Analyze the individual tables that are growing at higher rate than normal. Consider data archival using UNLOAD to S3 and Redshift Spectrum for later analysis.

Use unscanned_table_summary.sql to find unused table and archive or drop them.

top_queries.sql – Return the top 50 time consuming statements aggregated by its textETL transformations are taking longer to executeAnalyze the top transformation SQL and use EXPLAIN to find opportunities for tuning the query plan.

There are several other useful scripts available in the amazon-redshift-utils repository. The AWS Lambda Utility Runner runs a subset of these scripts on a scheduled basis, allowing you to automate much of monitoring of your ETL processes.

Example ETL process

The following ETL process reinforces some of the best practices discussed in this post. Consider the following four-step daily ETL workflow where data from an RDBMS source system is staged in S3 and then loaded into Amazon Redshift. Amazon Redshift is used to calculate daily, weekly, and monthly aggregations, which are then unloaded to S3, where they can be further processed and made available for end-user reporting using a number of different tools, including Redshift Spectrum and Amazon Athena.

Step 1:  Extract from the RDBMS source to a S3 bucket

In this ETL process, the data extract job fetches change data every 1 hour and it is staged into multiple hourly files. For example, the staged S3 folder looks like the following:

 [[email protected] ~]$ aws s3 ls s3://<<S3 Bucket>>/batch/2017/07/02/
2017-07-02 01:59:58   81900220 20170702T01.export.gz
2017-07-02 02:59:56   84926844 20170702T02.export.gz
2017-07-02 03:59:54   78990356 20170702T03.export.gz
2017-07-02 22:00:03   75966745 20170702T21.export.gz
2017-07-02 23:00:02   89199874 20170702T22.export.gz
2017-07-02 00:59:59   71161715 20170702T23.export.gz

Organizing the data into multiple, evenly sized files enables the COPY command to ingest this data using all available resources in the Amazon Redshift cluster. Further, the files are compressed (gzipped) to further reduce COPY times.

Step 2: Stage data to the Amazon Redshift table for cleansing

Ingesting the data can be accomplished using a JSON-based manifest file. Using the manifest file ensures that S3 eventual consistency issues can be eliminated and also provides an opportunity to dedupe any files if needed. A sample manifest20170702.json file looks like the following:

  "entries": [
    {"url":" s3://<<S3 Bucket>>/batch/2017/07/02/20170702T01.export.gz", "mandatory":true},
    {"url":" s3://<<S3 Bucket>>/batch/2017/07/02/20170702T02.export.gz", "mandatory":true},
    {"url":" s3://<<S3 Bucket>>/batch/2017/07/02/20170702T23.export.gz", "mandatory":true}

The data can be ingested using the following command:

SET wlm_query_slot_count TO <<max available concurrency in the ETL queue>>;
COPY stage_tbl FROM 's3:// <<S3 Bucket>>/batch/manifest20170702.json' iam_role 'arn:aws:iam::0123456789012:role/MyRedshiftRole' manifest;

Because the downstream ETL processes depend on this COPY command to complete, the wlm_query_slot_count is used to claim all the memory available to the queue. This helps the COPY command complete as quickly as possible.

Step 3: Transform data to create daily, weekly, and monthly datasets and load into target tables

Data is staged in the “stage_tbl” from where it can be transformed into the daily, weekly, and monthly aggregates and loaded into target tables. The following job illustrates a typical weekly process:

INSERT into ETL_LOG (..) values (..);
DELETE from weekly_tbl where dataset_week = <<current week>>;
INSERT into weekly_tbl (..)
  SELECT date_trunc('week', dataset_day) AS week_begin_dataset_date, SUM(C1) AS C1, SUM(C2) AS C2
	FROM   stage_tbl
GROUP BY date_trunc('week', dataset_day);
INSERT into AUDIT_LOG values (..);

As shown above, multiple steps are combined into one transaction to perform a single commit, reducing contention on the commit queue.

Step 4: Unload the daily dataset to populate the S3 data lake bucket

The transformed results are now unloaded into another S3 bucket, where they can be further processed and made available for end-user reporting using a number of different tools, including Redshift Spectrum and Amazon Athena.

unload ('SELECT * FROM weekly_tbl WHERE dataset_week = <<current week>>’) TO 's3:// <<S3 Bucket>>/datalake/weekly/20170526/' iam_role 'arn:aws:iam::0123456789012:role/MyRedshiftRole';


Amazon Redshift lets you easily operate petabyte-scale data warehouses on the cloud. This post summarized the best practices for operating scalable ETL natively within Amazon Redshift. I demonstrated efficient ways to ingest and transform data, along with close monitoring. I also demonstrated the best practices being used in a typical sample ETL workload to transform the data into Amazon Redshift.

If you have questions or suggestions, please comment below.


About the Author

Thiyagarajan Arumugam is a Big Data Solutions Architect at Amazon Web Services and designs customer architectures to process data at scale. Prior to AWS, he built data warehouse solutions at Amazon.com. In his free time, he enjoys all outdoor sports and practices the Indian classical drum mridangam.


Raspberry Pi Spy’s Alexa Skill

Post Syndicated from Alex Bate original https://www.raspberrypi.org/blog/pi-spy-alexa-skill/

With Raspberry Pi projects using home assistant services such as Amazon Alexa and Google Home becoming more and more popular, we invited Raspberry Pi maker Matt ‘Raspberry Pi Spy‘ Hawkins to write a guest post about his latest project, the Pi Spy Alexa Skill.

Pi Spy Alexa Skill Raspberry Pi

Pi Spy Skill

The Alexa system uses Skills to provide voice-activated functionality, and it allows you to create new Skills to add extra features. With the Pi Spy Skill, you can ask Alexa what function each pin on the Raspberry Pi’s GPIO header provides, for example by using the phrase “Alexa, ask Pi Spy what is Pin 2.” In response to a phrase such as “Alexa, ask Pi Spy where is GPIO 8”, Alexa can now also tell you on which pin you can find a specific GPIO reference number.

This information is already available in various forms, but I thought it would be useful to retrieve it when I was busy soldering or building circuits and had no hands free.

Creating an Alexa Skill

There is a learning curve to creating a new Skill, and in some regards it was similar to mobile app development.

A Skill consists of two parts: the first is created within the Amazon Developer Console and defines the structure of the voice commands Alexa should recognise. The second part is a webservice that can receive data extracted from the voice commands and provide a response back to the device. You can create the webservice on a webserver, internet-connected device, or cloud service.

I decided to use Amazon’s AWS Lambda service. Once set up, this allows you to write code without having to worry about the server it is running on. It also supports Python, so it fit in nicely with most of my other projects.

To get started, I logged into the Amazon Developer Console with my personal Amazon account and navigated to the Alexa section. I created a new Skill named Pi Spy. Within a Skill, you define an Intent Schema and some Sample Utterances. The schema defines individual intents, and the utterances define how these are invoked by the user.

Here is how my ExaminePin intent is defined in the schema:

Pi Spy Alexa Skill Raspberry Pi

Example utterances then attempt to capture the different phrases the user might speak to their device.

Pi Spy Alexa Skill Raspberry Pi

Whenever Alexa matches a spoken phrase to an utterance, it passes the name of the intent and the variable PinID to the webservice.

In the test section, you can check what JSON data will be generated and passed to your webservice in response to specific phrases. This allows you to verify that the webservices’ responses are correct.

Pi Spy Alexa Skill Raspberry Pi

Over on the AWS Services site, I created a Lambda function based on one of the provided examples to receive the incoming requests. Here is the section of that code which deals with the ExaminePin intent:

Pi Spy Alexa Skill Raspberry Pi

For this intent, I used a Python dictionary to match the incoming pin number to its description. Another Python function deals with the GPIO queries. A URL to this Lambda function was added to the Skill as its ‘endpoint’.

As with the Skill, the Python code can be tested to iron out any syntax errors or logic problems.

With suitable configuration, it would be possible to create the webservice on a Pi, and that is something I’m currently working on. This approach is particularly interesting, as the Pi can then be used to control local hardware devices such as cameras, lights, or pet feeders.


My Alexa Skill is currently only available to UK users. I’m hoping Amazon will choose to copy it to the US service, but I think that is down to its perceived popularity, or it may be done in bulk based on release date. In the next update, I’ll be adding an American English version to help speed up this process.

The post Raspberry Pi Spy’s Alexa Skill appeared first on Raspberry Pi.

Three new stable kernels

Post Syndicated from jake original https://lwn.net/Articles/743246/rss

Greg Kroah-Hartman has announced the release of the 4.14.12, 4.9.75, and 4.4.110 stable kernels. The bulk of the
changes are either to fix the mitigations for Meltdown/Spectre (in 4.14.12) or to backport
those mitigations (in the two older kernels). There are apparently known (or
suspected) problems with
each of the releases, which Kroah-Hartman is hoping to get shaken out in
the near term. For example, the 4.4.110 announcement warns: “But be
careful, there have been some reports of problems with this
release during the -rc review cycle. Hopefully all of those issues are
now resolved.

So please test, as of right now, it should be ‘bug compatible’ with the
‘enterprise’ kernel releases with regards to the Meltdown bug and proper
support on all virtual platforms (meaning there is still a vdso issue
that might trip up some old binaries, again, please test!)”

New- AWS IoT Device Management

Post Syndicated from Jeff Barr original https://aws.amazon.com/blogs/aws/aws-iot-device-management/

AWS IoT and AWS Greengrass give you a solid foundation and programming environment for your IoT devices and applications.

The nature of IoT means that an at-scale device deployment often encompasses millions or even tens of millions of devices deployed at hundreds or thousands of locations. At that scale, treating each device individually is impossible. You need to be able to set up, monitor, update, and eventually retire devices in bulk, collective fashion while also retaining the flexibility to accommodate varying deployment configurations, device models, and so forth.

New AWS IoT Device Management
Today we are launching AWS IoT Device Management to help address this challenge. It will help you through each phase of the device lifecycle, from manufacturing to retirement. Here’s what you get:

Onboarding – Starting with devices in their as-manufactured state, you can control the provisioning workflow. You can use IoT Device Management templates to quickly onboard entire fleets of devices with a few clicks. The templates can include information about device certificates and access policies.

Organization – In order to deal with massive numbers of devices, AWS IoT Device Management extends the existing IoT Device Registry and allows you to create a hierarchical model of your fleet and to set policies on a hierarchical basis. You can drill-down through the hierarchy in order to locate individual devices. You can also query your fleet on attributes such as device type or firmware version.

Monitoring – Telemetry from the devices is used to gather real-time connection, authentication, and status metrics, which are published to Amazon CloudWatch. You can examine the metrics and locate outliers for further investigation. IoT Device Management lets you configure the log level for each device group, and you can also publish change events for the Registry and Jobs for monitoring purposes.

Remote ManagementAWS IoT Device Management lets you remotely manage your devices. You can push new software and firmware to them, reset to factory defaults, reboot, and set up bulk updates at the desired velocity.

Exploring AWS IoT Device Management
The AWS IoT Device Management Console took me on a tour and pointed out how to access each of the features of the service:

I already have a large set of devices (pressure gauges):

These gauges were created using the new template-driven bulk registration feature. Here’s how I create a template:

The gauges are organized into groups (by US state in this case):

Here are the gauges in Colorado:

AWS IoT group policies allow you to control access to specific IoT resources and actions for all members of a group. The policies are structured very much like IAM policies, and can be created in the console:

Jobs are used to selectively update devices. Here’s how I create one:

As indicated by the Job type above, jobs can run either once or continuously. Here’s how I choose the devices to be updated:

I can create custom authorizers that make use of a Lambda function:

I’ve shown you a medium-sized subset of AWS IoT Device Management in this post. Check it out for yourself to learn more!



NetNeutrality vs. Verizon censoring Naral

Post Syndicated from Robert Graham original http://blog.erratasec.com/2017/11/netneutrality-vs-verizon-censoring-naral.html

People keep retweeting this ACLU graphic in support of net neutrality. It’s wrong. In this post, I debunk the second item. I debunk other items in other posts [1] [4].

Firstly, it’s not a NetNeutrality issue (which applies only to the Internet), but an issue with text-messages. In other words, it’s something that will continue to happen even with NetNeutrality rules. People relate this to NetNeutrality as an analogy, not because it actually is such an issue.

Secondly, it’s an edge/content issue, not a transit issue. The details in this case is that Verizon provides a program for sending bulk messages to its customers from the edge of the network. Verizon isn’t censoring text messages in transit, but from the edge. You can send a text message to your friend on the Verizon network, and it won’t be censored. Thus the analogy is incorrect — the correct analogy would be with content providers like Twitter and Facebook, not ISPs like Comcast.

Like all cell phone vendors, Verizon polices this content, canceling accounts that abuse the system, like spammers. We all agree such censorship is a good thing, and that such censorship of content providers is not remotely a NetNeutrality issue. Content providers do this not because they disapprove of the content of spam such much as the distaste their customers have for spam.
Content providers that are political, rather than neutral to politics is indeed worrisome. It’s not a NetNeutrality issue per se, but it is a general “neutrality” issue. We free-speech activists want all content providers (Twitter, Facebook, Verizon mass-texting programs) to be free of political censorship — though we don’t want government to mandate such neutrality.
But even here, Verizon may be off the hook. They appear not be to be censoring one political view over another, but the controversial/unsavory way Naral expresses its views. Presumably, Verizon would be okay with less controversial political content.

In other words, as Verizon expresses it’s principles, it wants to block content that drivers away customers, but is otherwise neutral to the content. While this may unfairly target controversial political content, it’s at least basically neutral.

So in conclusion, while activists portray this as a NetNeutrality issue, it isn’t. It’s not even close.

Capturing Custom, High-Resolution Metrics from Containers Using AWS Step Functions and AWS Lambda

Post Syndicated from Nathan Taber original https://aws.amazon.com/blogs/compute/capturing-custom-high-resolution-metrics-from-containers-using-aws-step-functions-and-aws-lambda/

Contributed by Trevor Sullivan, AWS Solutions Architect

When you deploy containers with Amazon ECS, are you gathering all of the key metrics so that you can correctly monitor the overall health of your ECS cluster?

By default, ECS writes metrics to Amazon CloudWatch in 5-minute increments. For complex or large services, this may not be sufficient to make scaling decisions quickly. You may want to respond immediately to changes in workload or to identify application performance problems. Last July, CloudWatch announced support for high-resolution metrics, up to a per-second basis.

These high-resolution metrics can be used to give you a clearer picture of the load and performance for your applications, containers, clusters, and hosts. In this post, I discuss how you can use AWS Step Functions, along with AWS Lambda, to cost effectively record high-resolution metrics into CloudWatch. You implement this solution using a serverless architecture, which keeps your costs low and makes it easier to troubleshoot the solution.

To show how this works, you retrieve some useful metric data from an ECS cluster running in the same AWS account and region (Oregon, us-west-2) as the Step Functions state machine and Lambda function. However, you can use this architecture to retrieve any custom application metrics from any resource in any AWS account and region.

Why Step Functions?

Step Functions enables you to orchestrate multi-step tasks in the AWS Cloud that run for any period of time, up to a year. Effectively, you’re building a blueprint for an end-to-end process. After it’s built, you can execute the process as many times as you want.

For this architecture, you gather metrics from an ECS cluster, every five seconds, and then write the metric data to CloudWatch. After your ECS cluster metrics are stored in CloudWatch, you can create CloudWatch alarms to notify you. An alarm can also trigger an automated remediation activity such as scaling ECS services, when a metric exceeds a threshold defined by you.

When you build a Step Functions state machine, you define the different states inside it as JSON objects. The bulk of the work in Step Functions is handled by the common task state, which invokes Lambda functions or Step Functions activities. There is also a built-in library of other useful states that allow you to control the execution flow of your program.

One of the most useful state types in Step Functions is the parallel state. Each parallel state in your state machine can have one or more branches, each of which is executed in parallel. Another useful state type is the wait state, which waits for a period of time before moving to the next state.

In this walkthrough, you combine these three states (parallel, wait, and task) to create a state machine that triggers a Lambda function, which then gathers metrics from your ECS cluster.

Step Functions pricing

This state machine is executed every minute, resulting in 60 executions per hour, and 1,440 executions per day. Step Functions is billed per state transition, including the Start and End state transitions, and giving you approximately 37,440 state transitions per day. To reach this number, I’m using this estimated math:

26 state transitions per-execution x 60 minutes x 24 hours

Based on current pricing, at $0.000025 per state transition, the daily cost of this metric gathering state machine would be $0.936.

Step Functions offers an indefinite 4,000 free state transitions every month. This benefit is available to all customers, not just customers who are still under the 12-month AWS Free Tier. For more information and cost example scenarios, see Step Functions pricing.

Why Lambda?

The goal is to capture metrics from an ECS cluster, and write the metric data to CloudWatch. This is a straightforward, short-running process that makes Lambda the perfect place to run your code. Lambda is one of the key services that makes up “Serverless” application architectures. It enables you to consume compute capacity only when your code is actually executing.

The process of gathering metric data from ECS and writing it to CloudWatch takes a short period of time. In fact, my average Lambda function execution time, while developing this post, is only about 250 milliseconds on average. For every five-second interval that occurs, I’m only using 1/20th of the compute time that I’d otherwise be paying for.

Lambda pricing

For billing purposes, Lambda execution time is rounded up to the nearest 100-ms interval. In general, based on the metrics that I observed during development, a 250-ms runtime would be billed at 300 ms. Here, I calculate the cost of this Lambda function executing on a daily basis.

Assuming 31 days in each month, there would be 535,680 five-second intervals (31 days x 24 hours x 60 minutes x 12 five-second intervals = 535,680). The Lambda function is invoked every five-second interval, by the Step Functions state machine, and runs for a 300-ms period. At current Lambda pricing, for a 128-MB function, you would be paying approximately the following:

Total compute

Total executions = 535,680
Total compute = total executions x (3 x $0.000000208 per 100 ms) = $0.334 per day

Total requests

Total requests = (535,680 / 1000000) * $0.20 per million requests = $0.11 per day

Total Lambda Cost

$0.11 requests + $0.334 compute time = $0.444 per day

Similar to Step Functions, Lambda offers an indefinite free tier. For more information, see Lambda Pricing.


In the following sections, I step through the process of configuring the solution just discussed. If you follow along, at a high level, you will:

  • Configure an IAM role and policy
  • Create a Step Functions state machine to control metric gathering execution
  • Create a metric-gathering Lambda function
  • Configure a CloudWatch Events rule to trigger the state machine
  • Validate the solution


You should already have an AWS account with a running ECS cluster. If you don’t have one running, you can easily deploy a Docker container on an ECS cluster using the AWS Management Console. In the example produced for this post, I use an ECS cluster running Windows Server (currently in beta), but either a Linux or Windows Server cluster works.

Create an IAM role and policy

First, create an IAM role and policy that enables Step Functions, Lambda, and CloudWatch to communicate with each other.

  • The CloudWatch Events rule needs permissions to trigger the Step Functions state machine.
  • The Step Functions state machine needs permissions to trigger the Lambda function.
  • The Lambda function needs permissions to query ECS and then write to CloudWatch Logs and metrics.

When you create the state machine, Lambda function, and CloudWatch Events rule, you assign this role to each of those resources. Upon execution, each of these resources assumes the specified role and executes using the role’s permissions.

  1. Open the IAM console.
  2. Choose Roles, create New Role.
  3. For Role Name, enter WriteMetricFromStepFunction.
  4. Choose Save.

Create the IAM role trust relationship
The trust relationship (also known as the assume role policy document) for your IAM role looks like the following JSON document. As you can see from the document, your IAM role needs to trust the Lambda, CloudWatch Events, and Step Functions services. By configuring your role to trust these services, they can assume this role and inherit the role permissions.

  1. Open the IAM console.
  2. Choose Roles and select the IAM role previously created.
  3. Choose Trust RelationshipsEdit Trust Relationships.
  4. Enter the following trust policy text and choose Save.
  "Version": "2012-10-17",
  "Statement": [
      "Effect": "Allow",
      "Principal": {
        "Service": "lambda.amazonaws.com"
      "Action": "sts:AssumeRole"
      "Effect": "Allow",
      "Principal": {
        "Service": "events.amazonaws.com"
      "Action": "sts:AssumeRole"
      "Effect": "Allow",
      "Principal": {
        "Service": "states.us-west-2.amazonaws.com"
      "Action": "sts:AssumeRole"

Create an IAM policy

After you’ve finished configuring your role’s trust relationship, grant the role access to the other AWS resources that make up the solution.

The IAM policy is what gives your IAM role permissions to access various resources. You must whitelist explicitly the specific resources to which your role has access, because the default IAM behavior is to deny access to any AWS resources.

I’ve tried to keep this policy document as generic as possible, without allowing permissions to be too open. If the name of your ECS cluster is different than the one in the example policy below, make sure that you update the policy document before attaching it to your IAM role. You can attach this policy as an inline policy, instead of creating the policy separately first. However, either approach is valid.

  1. Open the IAM console.
  2. Select the IAM role, and choose Permissions.
  3. Choose Add in-line policy.
  4. Choose Custom Policy and then enter the following policy. The inline policy name does not matter.
    "Version": "2012-10-17",
    "Statement": [
            "Effect": "Allow",
            "Action": [ "logs:*" ],
            "Resource": "*"
            "Effect": "Allow",
            "Action": [ "cloudwatch:PutMetricData" ],
            "Resource": "*"
            "Effect": "Allow",
            "Action": [ "states:StartExecution" ],
            "Resource": [
            "Effect": "Allow",
            "Action": [ "lambda:InvokeFunction" ],
            "Resource": "arn:aws:lambda:*:*:function:WriteMetricFromStepFunction"
            "Effect": "Allow",
            "Action": [ "ecs:Describe*" ],
            "Resource": "arn:aws:ecs:*:*:cluster/ECSEsgaroth"

Create a Step Functions state machine

In this section, you create a Step Functions state machine that invokes the metric-gathering Lambda function every five (5) seconds, for a one-minute period. If you divide a minute (60) seconds into equal parts of five-second intervals, you get 12. Based on this math, you create 12 branches, in a single parallel state, in the state machine. Each branch triggers the metric-gathering Lambda function at a different five-second marker, throughout the one-minute period. After all of the parallel branches finish executing, the Step Functions execution completes and another begins.

Follow these steps to create your Step Functions state machine:

  1. Open the Step Functions console.
  2. Choose DashboardCreate State Machine.
  3. For State Machine Name, enter WriteMetricFromStepFunction.
  4. Enter the state machine code below into the editor. Make sure that you insert your own AWS account ID for every instance of “676655494xxx”
  5. Choose Create State Machine.
  6. Select the WriteMetricFromStepFunction IAM role that you previously created.
    "Comment": "Writes ECS metrics to CloudWatch every five seconds, for a one-minute period.",
    "StartAt": "ParallelMetric",
    "States": {
      "ParallelMetric": {
        "Type": "Parallel",
        "Branches": [
            "StartAt": "WriteMetricLambda",
            "States": {
             	"WriteMetricLambda": {
                  "Type": "Task",
				  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "WaitFive",
            "States": {
            	"WaitFive": {
            		"Type": "Wait",
            		"Seconds": 5,
            		"Next": "WriteMetricLambdaFive"
             	"WriteMetricLambdaFive": {
                  "Type": "Task",
				  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "WaitTen",
            "States": {
            	"WaitTen": {
            		"Type": "Wait",
            		"Seconds": 10,
            		"Next": "WriteMetricLambda10"
             	"WriteMetricLambda10": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "WaitFifteen",
            "States": {
            	"WaitFifteen": {
            		"Type": "Wait",
            		"Seconds": 15,
            		"Next": "WriteMetricLambda15"
             	"WriteMetricLambda15": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "Wait20",
            "States": {
            	"Wait20": {
            		"Type": "Wait",
            		"Seconds": 20,
            		"Next": "WriteMetricLambda20"
             	"WriteMetricLambda20": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "Wait25",
            "States": {
            	"Wait25": {
            		"Type": "Wait",
            		"Seconds": 25,
            		"Next": "WriteMetricLambda25"
             	"WriteMetricLambda25": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "Wait30",
            "States": {
            	"Wait30": {
            		"Type": "Wait",
            		"Seconds": 30,
            		"Next": "WriteMetricLambda30"
             	"WriteMetricLambda30": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "Wait35",
            "States": {
            	"Wait35": {
            		"Type": "Wait",
            		"Seconds": 35,
            		"Next": "WriteMetricLambda35"
             	"WriteMetricLambda35": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "Wait40",
            "States": {
            	"Wait40": {
            		"Type": "Wait",
            		"Seconds": 40,
            		"Next": "WriteMetricLambda40"
             	"WriteMetricLambda40": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "Wait45",
            "States": {
            	"Wait45": {
            		"Type": "Wait",
            		"Seconds": 45,
            		"Next": "WriteMetricLambda45"
             	"WriteMetricLambda45": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "Wait50",
            "States": {
            	"Wait50": {
            		"Type": "Wait",
            		"Seconds": 50,
            		"Next": "WriteMetricLambda50"
             	"WriteMetricLambda50": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
            "StartAt": "Wait55",
            "States": {
            	"Wait55": {
            		"Type": "Wait",
            		"Seconds": 55,
            		"Next": "WriteMetricLambda55"
             	"WriteMetricLambda55": {
                  "Type": "Task",
                  "Resource": "arn:aws:lambda:us-west-2:676655494xxx:function:WriteMetricFromStepFunction",
                  "End": true
        "End": true

Now you’ve got a shiny new Step Functions state machine! However, you might ask yourself, “After the state machine has been created, how does it get executed?” Before I answer that question, create the Lambda function that writes the custom metric, and then you get the end-to-end process moving.

Create a Lambda function

The meaty part of the solution is a Lambda function, written to consume the Python 3.6 runtime, that retrieves metric values from ECS, and then writes them to CloudWatch. This Lambda function is what the Step Functions state machine is triggering every five seconds, via the Task states. Key points to remember:

The Lambda function needs permission to:

  • Write CloudWatch metrics (PutMetricData API).
  • Retrieve metrics from ECS clusters (DescribeCluster API).
  • Write StdOut to CloudWatch Logs.

Boto3, the AWS SDK for Python, is included in the Lambda execution environment for Python 2.x and 3.x.

Because Lambda includes the AWS SDK, you don’t have to worry about packaging it up and uploading it to Lambda. You can focus on writing code and automatically take a dependency on boto3.

As for permissions, you’ve already created the IAM role and attached a policy to it that enables your Lambda function to access the necessary API actions. When you create your Lambda function, make sure that you select the correct IAM role, to ensure it is invoked with the correct permissions.

The following Lambda function code is generic. So how does the Lambda function know which ECS cluster to gather metrics for? Your Step Functions state machine automatically passes in its state to the Lambda function. When you create your CloudWatch Events rule, you specify a simple JSON object that passes the desired ECS cluster name into your Step Functions state machine, which then passes it to the Lambda function.

Use the following property values as you create your Lambda function:

Function Name: WriteMetricFromStepFunction
Description: This Lambda function retrieves metric values from an ECS cluster and writes them to Amazon CloudWatch.
Runtime: Python3.6
Memory: 128 MB
IAM Role: WriteMetricFromStepFunction

import boto3

def handler(event, context):
    cw = boto3.client('cloudwatch')
    ecs = boto3.client('ecs')
    print('Got boto3 client objects')
    Dimension = {
        'Name': 'ClusterName',
        'Value': event['ECSClusterName']

    cluster = get_ecs_cluster(ecs, Dimension['Value'])
    cw_args = {
       'Namespace': 'ECS',
       'MetricData': [
               'MetricName': 'RunningTask',
               'Dimensions': [ Dimension ],
               'Value': cluster['runningTasksCount'],
               'Unit': 'Count',
               'StorageResolution': 1
               'MetricName': 'PendingTask',
               'Dimensions': [ Dimension ],
               'Value': cluster['pendingTasksCount'],
               'Unit': 'Count',
               'StorageResolution': 1
               'MetricName': 'ActiveServices',
               'Dimensions': [ Dimension ],
               'Value': cluster['activeServicesCount'],
               'Unit': 'Count',
               'StorageResolution': 1
               'MetricName': 'RegisteredContainerInstances',
               'Dimensions': [ Dimension ],
               'Value': cluster['registeredContainerInstancesCount'],
               'Unit': 'Count',
               'StorageResolution': 1
    print('Finished writing metric data')
def get_ecs_cluster(client, cluster_name):
    cluster = client.describe_clusters(clusters = [ cluster_name ])
    print('Retrieved cluster details from ECS')
    return cluster['clusters'][0]

Create the CloudWatch Events rule

Now you’ve created an IAM role and policy, Step Functions state machine, and Lambda function. How do these components actually start communicating with each other? The final step in this process is to set up a CloudWatch Events rule that triggers your metric-gathering Step Functions state machine every minute. You have two choices for your CloudWatch Events rule expression: rate or cron. In this example, use the cron expression.

A couple key learning points from creating the CloudWatch Events rule:

  • You can specify one or more targets, of different types (for example, Lambda function, Step Functions state machine, SNS topic, and so on).
  • You’re required to specify an IAM role with permissions to trigger your target.
    NOTE: This applies only to certain types of targets, including Step Functions state machines.
  • Each target that supports IAM roles can be triggered using a different IAM role, in the same CloudWatch Events rule.
  • Optional: You can provide custom JSON that is passed to your target Step Functions state machine as input.

Follow these steps to create the CloudWatch Events rule:

  1. Open the CloudWatch console.
  2. Choose Events, RulesCreate Rule.
  3. Select Schedule, Cron Expression, and then enter the following rule:
    0/1 * * * ? *
  4. Choose Add Target, Step Functions State MachineWriteMetricFromStepFunction.
  5. For Configure Input, select Constant (JSON Text).
  6. Enter the following JSON input, which is passed to Step Functions, while changing the cluster name accordingly:
    { "ECSClusterName": "ECSEsgaroth" }
  7. Choose Use Existing Role, WriteMetricFromStepFunction (the IAM role that you previously created).

After you’ve completed with these steps, your screen should look similar to this:

Validate the solution

Now that you have finished implementing the solution to gather high-resolution metrics from ECS, validate that it’s working properly.

  1. Open the CloudWatch console.
  2. Choose Metrics.
  3. Choose custom and select the ECS namespace.
  4. Choose the ClusterName metric dimension.

You should see your metrics listed below.

Troubleshoot configuration issues

If you aren’t receiving the expected ECS cluster metrics in CloudWatch, check for the following common configuration issues. Review the earlier procedures to make sure that the resources were properly configured.

  • The IAM role’s trust relationship is incorrectly configured.
    Make sure that the IAM role trusts Lambda, CloudWatch Events, and Step Functions in the correct region.
  • The IAM role does not have the correct policies attached to it.
    Make sure that you have copied the IAM policy correctly as an inline policy on the IAM role.
  • The CloudWatch Events rule is not triggering new Step Functions executions.
    Make sure that the target configuration on the rule has the correct Step Functions state machine and IAM role selected.
  • The Step Functions state machine is being executed, but failing part way through.
    Examine the detailed error message on the failed state within the failed Step Functions execution. It’s possible that the
  • IAM role does not have permissions to trigger the target Lambda function, that the target Lambda function may not exist, or that the Lambda function failed to complete successfully due to invalid permissions.
    Although the above list covers several different potential configuration issues, it is not comprehensive. Make sure that you understand how each service is connected to each other, how permissions are granted through IAM policies, and how IAM trust relationships work.


In this post, you implemented a Serverless solution to gather and record high-resolution application metrics from containers running on Amazon ECS into CloudWatch. The solution consists of a Step Functions state machine, Lambda function, CloudWatch Events rule, and an IAM role and policy. The data that you gather from this solution helps you rapidly identify issues with an ECS cluster.

To gather high-resolution metrics from any service, modify your Lambda function to gather the correct metrics from your target. If you prefer not to use Python, you can implement a Lambda function using one of the other supported runtimes, including Node.js, Java, or .NET Core. However, this post should give you the fundamental basics about capturing high-resolution metrics in CloudWatch.

If you found this post useful, or have questions, please comment below.

EC2 Convertible Reserved Instance Update – New 1-Year CRI, Merges & Splits

Post Syndicated from Jeff Barr original https://aws.amazon.com/blogs/aws/ec2-convertible-reserved-instance-update-new-1-year-cri-merges-splits/

We launched Convertible Reserved Instances for EC2 just about a year ago. The Convertible RIs give you a significant discount (typically 54% when compared to On-Demand) and allow you to change the instance family and other parameters associated with the RI if your needs change.

Today we are introducing Convertible RIs with a 1-year term, complementing the existing 3-year term. We are also making the Convertible Reserved Instance model more flexible by allowing you to exchange portions of your RIs and to perform bulk exchanges.

New 1-Year Convertible RIs
Convertible Reserved Instances with a 1-year term are now available. This will give you more options and more flexibility; you can now purchase a mix of 1-year and 3-year Convertible Reserved Instances (CRIs) in accord with your needs. Startups with financial constraints will find this option attractive, as will other ventures that may not be in a position to make a commitment that runs for longer than one year.

Merging and Splitting Convertible RIs
Let’s say that you start running your web and application servers on M4 instances and uses Convertible RIs to save money. Later, after a tuning exercise you move your application servers to C4 instances. With today’s launch you can exchange a portion of your M4 Convertible RIs for C4 Convertible RIs. You can also merge two or more CRIs (perhaps for smaller instances) and obtain one for a larger instance.

The exchange model for Convertible Reserved Instances is based on splitting, exchanging, and merging. Let’s say I own a 3-year Partial Upfront CRI for four t2.micro instances:

My application has changed and now I want to use a pair of t2.micro instances and a single r4.xlarge. The first step is to split this CRI into the part that I want to keep and the part that I want to exchange. I select it and click on Modify Reserved Instances. Then I create my desired configuration and click on Continue:

I review the request and click on Submit Modifications:

The state of the CRI changes to indicate that it is being modified. After a moment or two it will be marked as retired, replaced by a pair that are active:

Now I can exchange one of the 2-instance CRIs. I select it, click on Exchange Reserved Instance, and enter the desired configuration for my new CRI:

I click on Find Offering to see my options, and choose the desired one, an r4.xlarge Partial Upfront. As you can see, the console “does the math” takes the remaining upfront value ($139.995 in this case) of the unneeded CRIs into account when computing the upfront payment:

When I am ready to move forward I click on Exchange. This initiates the exchange process and lets me know that it may take a few minutes to complete.

I can also merge two or more Convertible Reserved Instances together and then use them as the starting point for an exchange. To do this I simply select the existing CRIs, click on Action, and choose Exchange Reserved Instances. I can see the total remaining upfront value of the selected CRIs and proceed accordingly:

You can merge CRIs that have different start dates and/or term lengths. The merged CRI will have the expiry date of the RI that is furthest from the date of exchange. Merging CRIs with different term lengths always produces a 3-year CRI.

You can also perform the split, exchange, and merge operations using the AWS Command Line Interface (CLI) and the EC2 APIs.

Available Now
All of the functions and the 1-year CRIs described in this post are available now and you can start using them today.