One external SQS producer crashed our entire production API. Reserved concurrency could have safeguarded us at no cost. Here's how it works. ⚠️ Let's say you have one production AWS account. In your AWS account, there are two different kinds of resources:

I often work with clients on different operating systems. I need to verify that tools I offer them also work on Windows (I am on Mac) To test this quickly, I typically do the following: - set up an EC2 with the correct OS (Windows Server xx) - Connect via RDP and test it out

The most important command in the AWS CDK CLI isn't "deploy". It's "diff". I know this sounds obvious. But I see so many developers skipping this step. In a small side project, you might get away with it. You change a Lambda function, deploy it, and everything is fine.

AWS just invented a new pricing unit for Aurora DSQL. It’s called the DPU (Distributed Processing Unit). If you want to estimate your database costs, you need to understand this. Here is what we know so far: • It covers all request activities (reads, writes, queries)

Your developers are wasting 2 hours per service setting up alerts. You have 10 services. That's 20 hours solving the same problem. And you still don't know where half your alerts actually go. Your developers shouldn't be thinking about alert routing.

Giving developers permanent Admin access is a bad idea. But sometimes they really need it to fix things. I discovered TEAM (Temporary Elevated Access Management) some time back. It's a serverless React app from AWS that handles this perfectly: • Request temporary admin access

Your alert chaos isn't a configuration problem. It's an architecture problem. Different Slack apps per team. Email alerts nobody reads. Webhooks pointing to servers that don't exist anymore. PagerDuty configured by someone who left 2 years ago.

GitHub Actions announced that self-hosted runner now costs something. Honestly, I never understood why it didn't cost anything so far. Of course, I liked it.

Lambda Durable Functions are here! Step Functions, just in code without ASL - seems interesting! I think it adds a bit to the complexity of "which service should I actually take" in Step Function vs. Lambda But in the end, you just get more opportunities to choose your favourite ones.

The CloudWatch Logs Insights Console got an upgrade 💅🏽 You can now: - Query log log groups directly from the scope - Access facets & fields (I think they renamed it from indexed fields to facets 🤔) - Look at saved queries in a different pane (not mixed anymore with fields, samples, etc.)

CloudWatch alert setups break the moment you add a second AWS account. Suddenly every developer is fighting OAuth flows and Slack permissions just to get alerts working. Here's the problem: Every developer needs to connect their alarms to Slack or Teams. That means: - Setting up OAuth

CDK 4 TF is deprecated 🪦 It was a great idea, I liked the concepts. But tbh this is why I'm often afraid of using non-official IAC tools. (saying that as an avid sst user)

I've had my third (irl) conference talk in my life and I've got some feedback. I appreciate that! The talk was about "EDA for the real world". Let's look at the feedback: Quality of the session: 4.6 (n=14) What was liked most: Presentation style / Practical examples (n=13) Any Improvements:

I've been running centralized logging for our AWS accounts. CloudWatch log centralization has two gotchas that caught me off guard. Gotcha #1: Settings Don't Transfer 🚨 Log Group settings get stripped during the copy. Retention policies? Gone. Data protection policies? Gone.

Some re:invent launches I got from the main keynote (couldn't watch it all - more details tomorrow): AI - New chips - P6 family - AI Factories: Own datacenter for your AI workloads - AWS Trainium: Custom AI Chip -> Best Price Performance

AWS trains Fortune 500 companies with these workshops. You can access them for free. They're migrating from workshops.aws to the builder center: builder.aws.com/build/works... These are the same workshops used in official AWS Immersion Days at companies.

You're guessing in production because tutorials skip the boring parts. We bundled everything they don't teach: monitoring, costs, observability, real patterns. 4 days only. Back to full price Tuesday. 📙 The CloudWatch Book - 350+ pages, 6+ hours of video, production-ready code

Most RAG implementations suck because everyone skips the boring parts. We just shipped a related posts feature with Bedrock. - Total queries: 244 (against our knowledge base) - Average related posts: 1.4 - Posts with zero results: 30 That's not too good.

Honeycomb published a blog post called "The End of Observability as We Know It." I've been using observability tools for years now. Heck, I wrote a book about CloudWatch ⛅ And honestly? I like the take of this post. Let's see a typical daily observability issue: - You have 5 AWS accounts.

You can now stream AI responses through API Gateway natively! The main changes: - Integration timeout extended to 15 minutes - Larger payloads possible You can even stream your whole documents & media files from S3 without pre-signed links now 🤯

Public Lambda Roadmap is live 🔥 And OTEL is already in the researching state 👀 🔗 github.com/orgs/aws/pr...

We added semantic search to our blog in one afternoon. No vector database. No extra infrastructure. Just S3 and Bedrock. 100+ posts now have intelligent related content suggestions. The setup was surprisingly simple: 1️⃣ Upload Content to S3 Store your markdown files in S3.

"Based on this feedback, CodeCommit is returning to full General Availability, effective immediately." - that is a new one! Welcome Back CodeCommit 👨🏽‍💻

CloudWatch log centralization takes 3 steps. 𝗦𝘁𝗲𝗽 𝟭: 𝗗𝗲𝗹𝗲𝗴𝗮𝘁𝗲 𝗮𝗻 𝗔𝗱𝗺𝗶𝗻𝗶𝘀𝘁𝗿𝗮𝘁𝗼𝗿 Go to your management account. CloudWatch → Settings → Organization. Assign one delegated administrator. I recommend using your dedicated logging account.

AWS finally made centralized logging simple 🎉 Everybody says "you need a central logging account" but creating it was quite a hassle. Your options before: 🔹 Observability Access Manager (OAM) Works, but complicated setup. Requires understanding sinks, sources, and links.

Nice new launch - I think that means no more: - "Bucket with this name already exists" - "Couldn't delete bucket because it is not empty" The feedback loop is faster to have it BEFORE it actually starts deploying. So far it works on:

TIL: Step Functions Tasks can assume roles for cross-account access 🤯

What is your go-to Infrastructure-as-code tool nowadays? Please comment if it is other. I'm really curious.

As a YouTube Shorts viewer, I want to save Shorts to reuse the music in my own videos. I don't want to watch them in a saved folder called "Favorites." - Google 🙄

Typically, one of the first issues you will encounter in bad architected systems: 𝐓𝐢𝐦𝐞 𝐂𝐨𝐮𝐩𝐥𝐢𝐧𝐠 Imagine building a blogging platform. Your `create-post` API started out with just publishing a post. But after a few more cool ideas this API is now responsible for:

You typically only create metric alarms in CloudWatch but the real value often comes with composite alarms. Composite alarms let you monitor multiple metrics. For example you can observe: - Lambda errors - API Gateway 500 errors

One feature I really love about CDK ➡️ CDK Aspects You can apply certain configurations or modifications of one kind of a construct. Typical CDK aspects: - attach alarm actions to every CloudWatch alert - assign tags to every construct - set log retention And they are created SUPER easily!

I'm still using Leapp daily for all of my accounts. My typical setup looks like this: 🏦 AWS Organization -> e.g. AWS Fundamentals 🧾 AWS Management Account -> Paying Account 🧾 AWS Account -> Dev/Prod ∎ AWS Profile -> ck-dev / ck-prod (those are the ones used in my IAC)

Finally having a serverless SQL database at hand with no monthly costs within AWS is simply amazing. We are running a mini application on DSQL for quite some time to track our subscribers & sales I just calculated how much it would cost us monthly: $0.00287

First two YouTube videos are up 🙌🏽 We're not planning on building a huge channel now, but we focus more on adding videos to blog posts and socials now. So, we thought "why not reuse them on our own channel" 📹

We've updated pricing of our Shopify App 🔥 (not as straightforward as we thought - feedback incoming) We wanted to try it out a few more things: - Smallest price is MUCH bigger -> Target bigger stores - We DITCHED the Free Plan and have only a free trial now

I just integrated @posthog into our second @Shopify app - MergeIT ✅ Some events we capture: - pageviews / leaves - orders merged + source (extension, dashboard, suggestion page) - suggestions viewed - settings changed - subscription activated - subscription cancelled

The main difference you have in event-driven systems vs. traditional systems is how you think about incoming events. ⚡ Typically, your API is orchestrating other functions to do certain things. In an event-driven system you let your consumers decide what to do.

Finally having a serverless SQL database at hand with no monthly costs within AWS is simply amazing. We are running a mini application on DSQL for quite some time to track our subscribers & sales I just calculated how much it would cost us monthly: $0.00287

I think this is the first time I'm not at the AWS Summit Germany in 4-5 years 🥲 Have fun guys! See you at the Community Day 😉

DSQL is GA now! That was much faster than I expected tbh. I'm finishing up my blog post about the pricing and publish it soon as a post and newsletter. TBH - understanding DPUs and how it is exactly billed is not as straightforward as I would have thought 😬

Serverless isn't just Next.js API Routes or Vercel functions. There's an entire ecosystem of powerful "forgotten middle" serverless services: • CloudFront • API Gateway • Step Functions • SNS • SQS • EventBridge

Oh oh, a few more deprecations are coming in from AWS ☠️ - Timestream for LiveAnalytics - Pinpoint (this one existed for ages) - AWS IQ - AWS IoT Analytics - AWS IoT Events - AWS SimSpace Weaver - AWS Panorama - Amazon Inspector Classic - Amazon Connect Voice ID - AWS DMS Fleet Advisor

One thing that changed about cold starts in Lambda - they are 𝐧𝐨𝐭 𝐟𝐫𝐞𝐞 𝐚𝐧𝐲𝐦𝐨𝐫𝐞 💸 Cold starts typically do the following things: 💾 Download the code from S3 or image from ECR ⚒️ Bootstrap environment 🆕 Initiate extensions 📠 Execute static code

Your CloudWatch Costs could go down by itself with this launch 💸 AWS counts Logs coming from Lambda now as "vended logs" and they come with a volume tiered pricing model.

Yes 🎉 One more year of AWS Community Builder Program 😊

Missed my talk at the AWS Community Day in Prague? Check out my recent session at the @serverlessguru hackathon: EDA for the Real World. I'm covering a two-fold agenda: 1️⃣ (a bit) Theory & Architecture about EDA 2️⃣ Hands-on TypeScript & CDK Examples

After publishing our animations we are also publishing all of our infographics (a bit slimmed down) on our page! The first iteration includes all infographics with text. Currently, we have: - Lambda - VPC - S3 - EC2 - DynamoDB - CloudFront - API Gateway - Route53 - IAM - CloudWatch - ECS

Ouh interesting, @stripe launches Workflows - like Step Functions integrated in Stripe 👀 I'm pretty sure if implemented well this can replace quite some common tasks that you typically need to implement webhooks for. Will check it out in my next Stripe project 👨🏽‍💻 🔗 stripe.dev/blog/introd...

Step Functions is coming into @SST_dev 🎉 The implementation is similar (as far as I know) as the CDK implementation. That means no more ASL + replacing of dynamic data which I highly appreciate! I do like the CDK-way of building Step Functions purely in code.