Welcome to issue number 52 of Last Week in AWS.
Are you doing anything this Wednesday? Cancel it; you’re going to the San Francisco AWS Summit instead. Not only will I be live-tweeting the keynote, but I’m also giving one of the DevChat sessions, New to AWS? Getting Started by Counter Example. I’ll be discussing the “I-swear-I’m-not-making-this-up” serverless architecture that I use to build and publish this newsletter, and why virtually every technical decision I made was wrong.
There will also be a number of sessions that are actually useful. If you’re going to be around, hit reply and let me know; I’d love to meet some of you in person.
Autotrader talks about how they use a variety of Lambda functions to monitor their data lake. I suspect a half dozen monitoring startups just about keeled over in shock, but are quickly recovering and attempting to sell Autotrader a very expensive product instead.
My friend Kevin built a URL shortener with Lambda. JUST Lambda. No DynamoDB, no persistent datastore other than Lambda itself. It’s beautiful in its horrifying nature.
Two methods of analyzing a billion NYC taxi rides— one involving EC2, Hadoop, and a hilariously long setup process, the other leveraging EMR. Due to spot pricing, both analyses cost less than $3 an hour.
Nobody wants to be the first to migrate production to a new service, then speak publicly about it– except for Datree. They discuss in some detail the interesting parts about migrating to AWS ECS Fargate in production.
On the one hand, Serverless is interesting. On the other, I’m not particularly inclined to agree that SERVERLESS IS EATING THE STACK AND WE’RE ALL DOOMED!!1! until building this newsletter with a handful of Lambdas didn’t require two weeks of coding and a hilariously convoluted architecture diagram. Serverless is great, yes– but it’s not coming for your little dog too just yet.
Never in my life have I wanted to shout “ALEXA, BUY 40 MILLION SHARES OF BERKSHIRE HATHAWAY” in someone else’s office– but that just changed.
A quick PostgreSQL benchmark that races RDS, Aurora, and Google Cloud SQL. I’ll let the results speak for themselves.
Last week’s episode of Screaming in the Cloud featured Seth Vargo of GCP. Listen to his wise words in Turning Off Someone Else’s Site as a Service, or wherever fine snark is sold.
Choice Cuts From the AWS Blog
Alexa for Business now lets you create address books for easier calling from shared Echo devices – This is one PagerDuty integration away from “Alexa, get me the on-call responder” being the default way to report that the printer is out of paper.
Amazon DynamoDB Adds Support for Continuous Backups and Point-In-Time Recovery (PITR) – Two weeks ago the Pinboard service had tagging break for a few days (these things happen; I’m not blaming anyone), and my failure to code defensively enough resulted in having to do a lot of manual work to ship last week’s issue. THE NEXT DAY they release a feature that would have saved all of that work. I’ve already enabled it; I’m set to fight the last war! That said, as a cloud economist, let me issue one word of warning: It costs between 20-30 cents per gigabyte per month (region depending). For me, that works out to less than a penny against my AWS credit balance. For other workloads, that’s a couple of hundred bucks a month per terabyte. Billing alarms are your friend.
AWS Fargate Platform Version 1.1 Adds Support for Task Metadata, Container Health Checks, and Service Discovery – As Fargate moves closer and closer to becoming production ready for many of us, should we call it Neargate instead?
Enable Trusted Organization Access in AWS Organizations – This feels like some organizational units are about to learn some very tough lessons about how far their peers’ trust in them extends.
Longer Sessions For IAM Roles – You can now have IAM roles get session tokens that expire in 12 hours (note that you need to explicitly request this; it’s not a default). This is such welcome news that I can only surmise that there’s something equally terrible attached to it that I can’t see yet– my money’s on a particularly terrible pronunciation of “IAM.”
Making Easier to Track Your Amazon EBS Volume State – A more honest headline would be “your scripts can finally stop polling every 5 seconds asking if the task is done yet like a project manager on a time crunch.”
New Loss functions and automatic early stopping now in the Linear Learner Algorithm in Amazon SageMaker – I suddenly have a much greater degree of empathy for how my non-technical friends must feel when I go off on a cloud services rant. I understood roughly half of the words in this blog post.
Amazon ECS Service Discovery | AWS News Blog – ECS now has integrated service discovery. Good lord, even the demo architecture diagram is mind-twisting to me. That said, this is a big step forward for modernizing some legacy applications, without having to learn something like Consul.
Migrating from an In-House Deployment Agent to AWS CodeDeploy and AWS CodePipeline | AWS DevOps Blog – This is a well written guest blog post that can be thought of as “We used to use custom-built crap to solve global problems, now we use standardized services like sensible adults.” That pattern is incredibly common; nice to see a shop admitting that they’ve emerged from it.
Performing Unit Testing in an AWS CodeStar Project | AWS DevOps Blog – I feel sorry for the author of this excellent post. Unit testing is like flossing; the only time you claim you do it is when an authority figure asks.
All AWS Services GDPR ready | AWS Security Blog – I’ve read this entire thing four times now, looking for a caveat such as “except for EC2.” I’ve not found one.
How to Prepare for AWS’s Move to Its Own Certificate Authority | AWS Security Blog – AWS is moving to its own CA. Most of us won’t notice or care, but if you intentionally disable certificate transparency, and then wonder why Chrome breaks your site, well… you’ve been warned.
Introducing ec2details, which gives a lot of missing metadata for EC2 instances via API. Pricing, CPU comparisons… check this out if you’re tired of mangling the Bulk API that AWS makes available.
Two tools from two companies to solve the same problem in this issue: operational visibility into your environments. Chick-fil-a released bovine, while Trulia’s is called cidr-house-rules. I feel like product marketing teams held a conference or something on the best way to my heart being stupid puns. They’re not wrong.
…and that’s what happened Last Week in AWS.