Sirupsen

turbopuffer is generally available. petabytes in prod.

turbopuffer.com/

Tak 🐡

now available: turbopuffer AWS regions ☁️

come find us in us-west-2 and ap-southeast-2

just received a redline in Word of a page on our website

i can't stop laughing

real alpha of a customer-obsessed remote team is that we have engineers in many cities our customers are in

our Paris engineer might need a sales quota soon, A+

We might have undersold the encryption here

tpuf is to my knowledge the only search engine where all at rest data can be encrypted with a different key per tenant, for hundreds of millions of tenants

maybe even only low latency data store?

continuing to ship full-text search features on our custom, object-storage first full-text search engine

search every byte 🐡💨

puffin' ❤️

❤️

Next step is surgery

California would soften me

(-22c)

Quick post about those 'latency numbers every programmer must know'. It became mostly an affirmation of
@sirupsen.bsky.social 's napkin-math repo (bookmark it).

so glad I got on this app

changelog for november & december 📉🔢💫

turbopuffer.com/docs/roadmap

Scandinavia discourages ambition (drew me away)

Canada tolerates ambition (good enough for me)

America encourages ambition

Play through a checkout flow of payment retries, abandoned checkout, fulfillment status changes, inventory exceptions, … Kind of annoying

S3 (Iceberg) Tables is everything I dreamt of, and more. I blogged some long-form thoughts: meltware.com/2024/12/04/s...

I think we're about to see an explosion of data tools (@materialize.com, @clickhouse.com, @duckdb.org, et al.) learn to write Iceberg tables via S3 table buckets.

#databs

Also not read your write, or can you do an LSN conditional query to PG?

Can Postgres and MySQL do that?

new super NVMe dense AWS SKU dropped 🤤

aws.amazon.com/ec2/instance...

But, that's a bit off. It's pretty scary to have your 64 core postgres writer maxing out all cores for an hour to build a 10M ANN index or however long it takes

I agree it's long-term a feature more than a product (we see ourselves as a search engine), but I do disagree that it's mega ez (if the vector indexes are > 10-100M, but once again, @glaubercosta.bsky.social probably also is implying individual indexes are small, and then yes, ez)

I’m guessing this is a theoretical number based on replica count and drive failures

Clone into closures · Issue #2407 · rust-lang/rfcs When working with move closures, and especially when spawning new threads, I often find myself having to clone a bunch of state (usually Arcs, but not always). This is either because I want that st...

github.com/rust-lang/rf... need this

Arc is a lifetime skill issue (so I use them a lot)

prod metrics > napkin math > RFCs

Not yet, not a bottleneck

Which other ones are you tracking?

lol we have a custom s3 client, we'll start testing tomorrow

We considered it, but it's just too janky. LIST has a high p90 in our experience, so now you have to start compacting, and that has a lot of edge-cases... So it felt like a web of pain for something that should NOT go wrong. So we waited and kept shipping on GCS.