Jerome Gravel-Niquet @[email protected]

Finally. A developer focused S3 alternative (built on @flydotio). I've been waiting for something like this for ages. (wohoo!!) tigrisdata.com

Working on a post but I present a React table of 2 million rows, dynamically filtered in real time. tldr: the data structure backing the table is maintained via differential dataflow so we only need to process the single row that changed rather than the entire set when filtering

@chris_mccord @mrkurt @simonw Since the release, I've made yet-to-be-released changes to the subscriptions API and improved the docs greatly if any of you want to take another look! superfly.github.io/corrosion/api/…

@jeromegn @mrkurt @simonw It looks like a fantastic base to build an eventually consistent durable pubsub system on top of. Missed broadcasts are replayed thru regular inserts.

We open sourced Corrosion, the tool we built to replace Consul. It started as gossip based service discovery. If you want to win buzzword bingo: Corrosion is built with Rust, CRDTs, and sqlite. github.com/superfly/corro…

@louispilfold @bcjordan @mrkurt *single source of truth per unique primary keys constraint

@louispilfold @bcjordan @mrkurt It could be changed to use the node ID as the tie breaker, but that would mean a specific node might win all conflicts. To avoid these situations, it's best to have a single source of truth or data that changes slower than the cluster can synchronize.

@mrkurt @simonw @chris_mccord It's also what the templating feature uses to update produced files. More docs to come! The first integer per row / change is an internal rowid for the query result row. It's meant for updating state based on that number (e.g. for reactivity) without a making a new query.

@simonw Our proxy maintains in internal, in memory service state. We use the subscription to keep that in sync (roughly cache invalidation). @chris_mccord has some ideas for using it to do a pseudo-persistent, distributed pubsub as well.

@bcjordan @mrkurt We already are :) you can put any kind of data that fits in SQLite.

@mrkurt the developer urge to use this for distributed datastore/caching use cases beyond service discovery

@dovunderscore @tantaman @ccorcos Those numbers were for "v1" of what's current in production at Fly. cr-sqlite will be used for v2 which we intend on releasing open source in the coming months. It should be able to handle the same workload no problem!

@tantaman @ccorcos @jeromegn I'm not Jerome but we are planning to talk about the project a bunch soon :)

> we're approaching 600 nodes and ~40K changes per second Exciting to see my work deployed into production but always makes me nervous 😅

@dilrajio @strzibnyj @flydotio We do yes. You can disable it when you deploy iirc. I think they’re called auto start and auto stop or something like that. If you fly deploy -h it should list them. I assume this is a new v2 app and not a v1 migrated to v2 (which should have that turned off by default)

New router settings are too advanced for me

@fasterthanlime I need to read that whole series. I have NixOS going but barely understand what I’m doing.

@MichaelSharer @jacobmparis @kentcdodds Are you starting your app with `npm start`? If so, using `node` directly might work better. `npm start` itself uses ~50MB for the lifetime of the app.

@jacobmparis @kentcdodds Started with 256, bumped to 512. Still have it. It’s a Blues stack app. Im not too concerned given it is staging, just a little annoying getting emails daily

I have a lot of small projects on Fly that don't get or need much traffic - side projects - random experiments - and staging environments too Now they shut down when inactive, and spin back up on the first request with about a 5 second cold start jacobparis.com/guides/fly-aut…

@benbjohnson I’ve watched one like that before and I loved it! I don’t remember who the streamer was 🤔

Do you ever have an idea that's so nerdy that you're taken aback? I had an idea to do a research paper Twitch stream. Like, instead of a programming stream we'd read over and breakdown CS research papers.

@benbjohnson This got me wondering about what go does differently here. Cgo uses different threads to avoid blocking the runtime which is pretty much what we have to do too. There’s probably a more ergonomic way to do it, in our project, that doesn’t require manually sending a closure.

Does anyone like async Rust? I understand the perf benefits of the state machine approach but it seems like there's so many tradeoffs in complexity. Like, this thread about using SQLite with async seems insane. People sending closures over channels, etc github.com/rusqlite/rusql…

@joshmo_dev Such secure! We do tell you to note it down somewhere iirc.

I just spent 2 hours trying to figure out why I couldn't get a string connection to my db from fly.io. Turns out they only given you a string directly after provision and never again 😂😂😂 I need a break

Great detailed post from @benhoytnz on getting Go + SQLite running on Fly.io. Even includes load testing on our tiny 256MB VMs! 500+ req/sec is pretty good! 👌 benhoyt.com/writings/flyio/

@marbiano3 @flydotio What are you seeing? The current status page incident is considered resolved (we are monitoring) since yesterday.