Akshay Shah
@akshayshah.org
130 followers 340 following 110 posts
Field CTO @ Antithesis https://akshayshah.org
Posts Media Videos Starter Packs
akshayshah.org
My recollection is that the catchphrase was “move the code to the data,” which seems borne out by Section 3.4 (“Locality”) of Google’s MR paper. I think Hadoop also tried to schedule mappers onto nodes with the relevant data, but I could be wrong - you’d certainly know.
akshayshah.org
Datadog must have been polishing this piece of ragebait for years. It's art, really.

(But seriously, this API is like a cheese grater to the face.)
akshayshah.org
Vibe-writing in Excel? No thanks 🤮

Unreliable tools can be incredibly useful - but you need a bulletproof test for their output. And spreadsheets are famously hard to get right! 94% of operational spreadsheets contain errors, and humans are exceptionally bad at catching them.
akshayshah.org
Best convo on the internet right now @vicvijayakumar.com
Reposted by Akshay Shah
cynthiadunlop.bsky.social
How AWS S3 serves 1 petabyte per second on top of slow HDDs // Stanislav Kozlovski

Obscure feature + obscure feature + obscure feature = bug // Michael Gibson, @antithesis.com

%CPU utilization is a lie // @brendanlong.bsky.social
akshayshah.org
This is *so* clever! At least in my head, this rhymes with the MapReduce philosophy of moving compute to the data.
andypavlo.bsky.social
Our F3 files embed small WASM programs to decode data. If somebody creates a new encoding and the DBMS does not have native impl, it can still read data using WASM passing Arrow buffers. Our experiments show WASM is 15-20% slower than native. We use @spiraldb.com's Vortex encoding impls.
Overview of F3's decoding pipeline with WASM support.
akshayshah.org
I’d rather write statsd to the wire by hand than descend into the Lovecraftian horror of the OTel Go docs.
akshayshah.org
Poor Joe. I knew him before his radical sourdough ways.
akshayshah.org
Unexpected benefit of saving PDFs of all the papers and @oreilly.bsky.social books I’ve read over the years: I have the most awesome NotebookLM session for distributed systems.
Reposted by Akshay Shah
thenewstack.io
Find out why AI code gen delivers only partial wins today. Join us on Oct 9 for a live webinar with @antithesis.com's Will Wilson & @akshayshah.org. They'll share how to bridge the gap between AI generation and real engineering productivity.

Don't miss it ➡️ bit.ly/Antithesis-A...
akshayshah.org
It’ll be no surprise that a bug very similar to this one led to the creation of goleak!
akshayshah.org
Appreciate the shout-out for automaxprocs, but it’s not usually necessary anymore (finally). Go 1.25 does the right thing by default! go.dev/blog/contain...
Container-aware GOMAXPROCS - The Go Programming Language
New GOMAXPROCS defaults in Go 1.25 improve behavior in containers.
go.dev
akshayshah.org
…but Go 1.25 has finally shipped with cgroup-aware defaults! go.dev/blog/contain...

We’d been asking for this since we wrote automaxprocs.
Container-aware GOMAXPROCS - The Go Programming Language
New GOMAXPROCS defaults in Go 1.25 improve behavior in containers.
go.dev
akshayshah.org
That doesn’t take into account cgroup quotas. Exceeding quota makes your application VERY slow.
akshayshah.org
Shmargaret Shmitchell, indeed.
akshayshah.org
Is the “modern tooling” a pro or a con?
akshayshah.org
Kyle’s Maelstrom project also includes services that implement weaker consistency models than linearizability/strict serializable - and of course, Kyle makes sure that you see all these behaviors!

github.com/jepsen-io/ma...
github.com
akshayshah.org
To their credit, Oxide’s (and @bcantrill.bsky.social’s) personality shines through all their writing and speaking. An unadorned, “No, because it suggests that you don’t really want the job,” would be wildly off-brand.