The baby Steller's Jay is out barnstorming.

Since everyone forgot how to write json by hand after the AI meteor strike of 2021 caused permanent darkmode, Speedrun 1.55 can now generate the necessary json config to wire your GitHub markdown to your AWS accounts with just your Identity Center start url. github.com/No-Backspace...

Journeyed to Wisconsin for some cheese therapy. Shullsburg Creamery's muenster melts different and cures ails.

Finally maxed my ESM score.

Good news! That 50ms coldstart latency regression that appeared when using the AWS Javascript V3 SDK on Lambda with Node >= 22 is now fixed. github.com/aws/aws-sdk-... Update your dependency to v3.1011 to get back in the fast lane. Comparing my patched v3.1006 to v3.1011 shows no delta.

I ran my benchmark overnight to confirm there wasn't much improvement when turning on NODE_COMPILE_CACHE with Lambda Coldstarts. It's less than a 10ms improvement. There may be benefits with larger packages but for my setup (200K zip file) it isn't worth the trouble.

I experimented today with enabling NODE_COMPILE_CACHE to see if it would make Lambda coldstarts faster. I was able to create a .zip file with the cache from a previous run. I verified node used the cache successfully with NODE_DEBUG_NATIVE="COMPILE_CACHE". It didn't reduce the coldstart time 🫠.

LLRT v.8.1-beta has the fastest p50 InitDuration by 130ms. But if you measure E2E time from the client (what you should measure), it narrows to 79ms over AwsLite and 118ms over a modified AWS v3 JavaScript client on Node 24. This is likely due to the size of the LLRT binary being ~5MB.

Anecdotally I've noticed Lambda nodejs coldstart latencies trending down but visualizing the improvement using the daily maxday numbers is hard: maxday.github.io/lambda-perf/ I pulled the data and graphed it. Indeed, it's showing a 20-30ms improvement since November last year. Here's nodejs on arm.

Getting the client ip address when using a Lambda Function Url generally works, but sometimes you get a spoofed one. In this article I cover how to workaround the undocumented mangling Lambda does to the `x-forwarded-for` header. speedrun.nobackspacecrew.com/blog/2026/03...

I discovered that if you use Lambda furls, the x-forwarded-for header is truncated to the leftmost value. If you front your furl with CloudFront, the ip in that header is spoofable. CloudFront appends the viewer ip to the value of x-forwarded-for. Workaround is: github.com/aws-samples/...

TIL I was still using http 1.1 to connect to my api which supported http3. Because http3 is still kind of broken in Firefox: bugzilla.mozilla.org/show_bug.cgi... and when you enable http3 in CloudFront using the CDK, it disables http2. github.com/aws/aws-cdk/...

TIL you can get edge location in your CloudFront Functions. docs.aws.amazon.com/AmazonCloudF...

Hello darkness my old friend. I just updated to JQuery 4.0.0. Ready for 10 more years of $.

PSA: There is a performance penalty when using named exports with ESM import vs. require on built-in node modules. This is why the AWS JavaScript V3 SDK has 50 ms slower Lambda coldstarts on Node 22+. Some notes here: github.com/aws/aws-sdk-...

Look, the plating and bread choice leaves room for improvement. But the flavor of my latest Persian-American cuisine; the sumac avocado toast slaps.

After many exasperated console.log's I looked at the docs and discovered CloudFront functions don't run on responses with a status code >= 400. So my edge metrics only work on successful requests. docs.aws.amazon.com/AmazonCloudF...

Seeing some Dr. Suess bird in my latte this morning

Ode to 2025. My pithy year in review. "May your 2026 involve some hand-written code and may it compile without errors." speedrun.nobackspacecrew.com/blog/2026/01...

It's important to know someone in London. Eventually you'll need their help unsubscribing from Domino's UK marketing because the US is region blocked 😂.

I swear AWS had some hero image on a docs site with this hotel carpet vibe once.

Took a quick break from the curiosities of London to have a curry and a pint with Cloudflare's lord of logs and corduroy trousers @ankcorn.dev

A short story: 1. This showed up in the group chat 2. Replied with "Nicolas UNCAGED?" 3. Checked El goog to see if this pun was a world first 4. Alas

Boris is still at it and still gets it. Logging properly is on you and rn your: loggingsucks.com

The answer to why account based endpoints didn't reduce the blast radius of the DynamoDb Oct event was in DAT458. They don't point to cells yet, they still point to the regional endpoint🤦‍♂️. youtu.be/YZUNNzLDWb8?...

I'm relishing using AI to click the bait for me.

Clearly the old spice product marketing team was cut from the same cloth as me.

This is very clever. Tom Larkworthy shows how to unaggregate CloudWatch metrics and some neat insights you get when you do: observablehq.com/@tomlarkwort...

Not sure I want my company outlined in chalk but safety in numbers at re:invent.

I got Cloudflare Workers to act as an API Gateway and do granular throttling for a Lambda using github.com/mhart/aws4fe... 1. I really like 𝚠𝚛𝚊𝚗𝚐𝚕𝚎𝚛 𝚍𝚎𝚟 to rapidly iterate 2. Cloudflare Workers are way better than CloudFront Functions

There is not a deep enough place in hell for the continue reading button, the random playing video and the slop ads below the fold.

Been messing with Cloudflare Workers after reading @boristane.bsky.social boristane.com/blog/what-ar... Delighted to find how easy it is to implement granular api throttling: developers.cloudflare.com/workers/runt... Something I've wished I could do with Cloudfront Functions for some time.

If you use Discord and don't want everyone to know what you are doing, turn this off: Settings -> Activity Privacy.

Armored up and went ball hunting in the blackberries. Lightly shredded, but I can now play lacrosse, rugby, soccer, tee-ball and ultimate.

The Cognito M2M pricing was repealed! Now it's $2.25/1000 tokens regardless of number of clients. aws.amazon.com/about-aws/wh...

I've started tagging my CDK projects so I can find them on disk when I inevitably need to upgrade the Lambda runtime.

They briefly mention the account based endpoints in the writeup: aws.amazon.com/message/1019... But my question remains, were those endpoints also impacted? Were the teams that couldn't get to dynamo not using the account based endpoints?

Ha! Something good came of the AWS incident Monday. I purchased a day pass to SlingTV to watch the Mariners play only to find I bought the wrong color (bought orange instead of blue which had the game). There aren't refunds on day passes, but due to the outage, they refunded me. I'll allow it!

When the dust settles with the DynamoDB DNS issue, i'd be interested in understanding if it impacted the main DNS only: dynamodb.us-east-1.amazonaws.com or also the account id based DNS: <accountid>.ddb.us-east-1.amazonaws.com? The account id DNS should have reduced the blast radius in theory.

One nice thing about cycling is the tool roadkill. This is the second pair of lock-jaw pliers I've found in the street.

I wrote about migrating my API from Node 22 to LLRT and the dragons I battled on the way. speedrun.nobackspacecrew.com/blog/2025/10...

We did it fam. I successfully got my api working on LLRT. I had to switch to hono, drop jose, and modify some request parsing for form data, but IT WORKS! Seeing 188 ms Init Duration vs. 375 ms for Node 22. Warm requests with AWS service calls are like 60ms faster too.

These aren't called spaces because tabs are better. This conversation is complete.

Lambda perf hasn't been working since September 15th: maxday.github.io/lambda-perf/

It appears I was using `@initDuration+@billedDuration` and not `@initDuration + @duration` artificially inflating the baseline to 144ms from ~78ms. The delta between making a service call to STS from init vs. the handler is close to nothing when using llrt.

@jason.energy was a multimodal mc at #cascadiajs He ranged from concert in the park pose to Carole Baskin leopard print. Good times!

The next version of LLRT supports calling AWS services in init so you can utilize the extra CPU power. In testing, the initDuration+duration went from p50 of 144 ms to 76 ms! However, the E2E duration (from the client) stayed the same p50 of 268 ms vs p50 of 266 ms. This makes no sense to me.

My garbage disposal kept failing due to low voltage at the switch (36 volts instead of 120). Today after it failed for the third time and my normal voodoo didn't work, I bought a new switch. The culprit was smoked sugar ants in the contact.

Got my api running on llrt! They patched the iam client issue. Init Duration vs Node 22 went from 375 ms to 114 ms (3.3x improvement) and normal invokes went from 181 ms to 81 ms (2.2x improvement). All with much less memory; in Node I allocated 832 MB vs 256 MB for llrt.

PSA, the npm packages `debug` and `chalk` got compromised. www.aikido.dev/blog/npm-deb... This hacker news comment has a ripgrep command to make sure you don't have it: news.ycombinator.com/item?id=4517...