Good Morning!
Another week come and gone, and with it spooky season, but AWS’s releases heat up due to the horrifying desert conference at the end of this month.
Speaking of! If you’re going to be in Las Vegas during re:Invent, come to our drink-up and let us buy you drinks until the end of the night when we hilariously discover the bar doesn’t take AWS credits, nor is my account manager anywhere to be found. It’s my favorite part of re:Invent, which is admittedly like saying something is my favorite part of being waterboarded.
From the Community
A dive into what Sovereign Cloud actually means. Spoiler: it’s a compliance security blanket.
I got to opine again in The Register in AWS outage: Myths vs reality.
A fun foray into a pertinent question: why didn’t ngrok go down in last week’s AWS outage?
Podcasts
Last Week In AWS: DynamoDB Rises Like Expensive Phoenix
Screaming in the Cloud: Cyber Resilience Beyond Prevention with Anneka Gupta
Choice Cuts
Beyond pilots: A proven framework for scaling AI to production – AWS just invented management consulting and slapped "AI framework" on it. The Five V’s (or "two-and-a-half W’s") are literally just "have a plan, measure stuff, test it, deploy it, fund it," which is groundbreaking advice that’s been standard practice since several decades before S3 existed.
New Amazon CloudWatch metrics to monitor EC2 instances exceeding I/O performance – Finally, metrics that tell you when you’ve been hitting invisible performance ceilings. The fact these are free is AWS admitting they should’ve existed years ago. I suspect they didn’t because how many customers have needlessly over-provisioned EBS volumes when it’s the EC2 instance limit throttling them the entire time?
Processing Amazon S3 objects at scale with AWS Step Functions Distributed Map S3 prefix – "Do it your damned self with Step Functions" is apparently the backfill for the deprecated S3 Object Lambdas from this quarter’s deprecation batch.
What’s the difference between AWS ParallelCluster and AWS Parallel Computing Service? – Wait, these are two different things?! Who’s minding the store over there?
France Télévisions prepared for 2024 Olympic Games with AWS Countdown Premium – France Télévisions needed AWS hand-holding to broadcast the Olympics without melting down, which they’re now selling as a premium service called "Countdown." "Wait, hang on, the Olympics prints money; let’s slap a ‘Premium’ tier onto that!"
Amazon Kinesis Data Streams now supports 10x larger record sizes – They spent years telling you to chunk your data, now they’re charging you the same per-shard rate while you push 10x more through it. At least Lambda’s 6MiB limit means you’ll still hit arbitrary walls somewhere downstream.
Amazon DocumentDB (with MongoDB compatibility) announces upgraded query planner that can run queries up to 10x faster – Ten years after MongoDB figured out query optimization, Amazon Basics MongoDB finally catches up and calls it an innovation.
Announcing AWS X-Ray SDKs/Daemon End-of-Support and OpenTelemetry Migration – AWS is deprecating X-Ray SDKs for OpenTelemetry, which is actually the right move, because open standards beat vendor lock-in. But calling it a "migration" when you’re giving customers two years to rewrite their instrumentation? That’s just deprecation with better PR. Still, Google’d give you a month if you’re lucky.
Amazon S3 adds conditional write functionality to copy operations – Finally, a feature that doesn’t cost extra, presumably because AWS knows charging for basic write coordination would’ve triggered a congressional hearing. This should’ve shipped with S3 in 2006, but better late than never when you’re trying to prevent your customers from building their own locking mechanisms on top of DynamoDB. Which they’ve all already done. And are about to lose their minds when they see this feature.
Introducing the Capacity Reservation Topology API for AI, ML, and HPC instance types – AWS finally admits their capacity reservations were basically a black box, so now they’re giving you an API to see what you already paid for. Love that we need a separate API just to figure out if our expensive GPU clusters are sitting next to each other or in different zip codes. Good god, a topology API?
How to deploy a SQL Server Failover Cluster Instance across three Availability Zones using Storage Spaces Direct – AWS just published a 3,000-word tutorial on how to MacGyver SQL Server clustering across three AZs using Storage Spaces Direct because their actual managed services can’t do it. When your workaround documentation needs a table of contents, maybe it’s time to fix the product instead of teaching customers to build Rube Goldberg machines.
Reduce CAPTCHAs for AI agents browsing the web with Web Bot Auth (Preview) in Amazon Bedrock AgentCore Browser – "Wonder who that’s for?" says AWS as they walk past the sign reading "No AWS Allowed." It’s worth noting that this only works if website owners opt in, meaning you’re still one policy change away from your agent staring at fire hydrant photos.
Using Kubernetes Labels to Split and Track Application Costs on Amazon EKS – Finally, a way to track which team’s Kubernetes experiment is bankrupting you. The feature itself is genuinely useful: mapping K8s labels to cost allocation tags lets you blame the right people when the bill arrives. Just remember, accurate cost tracking doesn’t make the costs themselves any smaller.
Introducing AWS Lambda event source mapping tools in the AWS Serverless MCP Server – AWS shipped an AI assistant to help configure Lambda event source mappings because apparently the existing documentation wasn’t confusing enough. Clearly event-driven architecture has gotten so byzantine that they need LLMs to translate "I want to process Kafka events" into actual working configs. At least someone’s finally admitting this stuff is harder than it should be.
Split Cost Allocation Data for Amazon EKS supports Kubernetes labels – Now you can tag your way to understanding why your EKS bill looks like a phone number. Fifty custom labels should be just enough rope to either organize your costs brilliantly or hang yourself with tag sprawl. At least now you’ll know exactly which team to blame.
… and that’s what happened Last Week in AWS.
