Jim Calabro ♟️ @calabro.io · 24/09/2026I literally gasped out loud at how bad the macOS 27 buttons look 11421
Jim Calabro ♟️ @calabro.io · 13/08/2026@chrisshank.com thanks for the report, should be fixed! I made a little repro that was broken before, and I just shipped a quick fix, now it looks ok. let me know? 120
Jim Calabro ♟️ @calabro.io · 13/08/2026Do you like fast network replay? How does ~40 million events per second strike you? Replay the whole network in like 10 minutes In the best case scenario, running the jetstream server and the client on your own very powerful machine, this is roughly the peak I've achieved so far 1016125
Jim Calabro ♟️ @calabro.io · 13/08/2026Yep! You can do something like this from within the jetstream repo, check that ./cmd/client code out. The sample client is a _lot_ of code because it's got "one of everything", so tonight and tomorrow morning I'll be working on some simpler examples 110
Jim Calabro ♟️ @calabro.io · 14/07/2026ughhh it's too expensive but like, not _that_ expensive ughhh 250
Jim Calabro ♟️ @calabro.io · 10/07/2026no longer will this be an issue! it's now built on a cluster that can autoscale (still on-prem), but won't have to for a very long time because of how ridiculously efficient it is here's cpu usage on the old cluster vs. the new 2200
Jim Calabro ♟️ @calabro.io · 11/06/2026not in the bsky app right now, but I see that mu supports it 150
Jim Calabro ♟️ @calabro.io · 02/06/2026not to tease too much, but here's what I think is reasonably close to fastest possible with present mushroom PDS rate limits (which we should raise) 4,792,320 records in 30s 24,698,880 records in 120s (more pipes getting warmed up) 260
Jim Calabro ♟️ @calabro.io · 27/05/2026Our tempo (traces) DB started accumulating a bunch of disk and we only started getting alerted today. Since tempo stores its stuff in a columnar parquet format, I can just run duckdb against it to figure out what the biggest spans/services are 260
Jim Calabro ♟️ @calabro.io · 26/05/2026@evan.jarrett.net yo, I'm detecting a lot of commits from did:plc:wnbpdx4bn5uts5yd2o2g3gnm that are missing the prevData CID for the deletes that are being sent over the firehose FYI Might be worth taking a look at it? atproto.com/specs/sync this is needed for MST inversion in sync 1.1 470
Jim Calabro ♟️ @calabro.io · 24/05/2026heres those big spikes over the past 7 broken down by host. the big spikes are hygrophorous! odd! 130
Jim Calabro ♟️ @calabro.io · 24/05/2026this is...both interesting and concerning! hygrophorous is a problematic host that had full hardware failure a few weeks ago, and we migrated it to another host. looking in to that... top hosts by # of error logs over the past 7 days. @hailey.at and @mk.gg might want to take a peek at this? 280
Jim Calabro ♟️ @calabro.io · 24/05/2026this is pretty imprecise, but hopefully it's helpful! based on logs from the relay: github.com/bluesky-soci... here are some relay logs over the past 7 days: 180
Jim Calabro ♟️ @calabro.io · 15/04/2026coming soon...backed by spacetimedb...shockingly easy spacetimedb.com/docs/ 4271
Jim Calabro ♟️ @calabro.io · 10/04/2026in a rare C++ W, they got the naming of std::shared_lock and std::unique_lock right, all other choices are misleading 020
Jim Calabro ♟️ @calabro.io · 09/04/2026If you're writing C, C++, Rust, or Zig and not using tracy, you should take a look at it! Very useful github.com/wolfpld/tracy 4643
Jim Calabro ♟️ @calabro.io · 27/03/2026I take back everything I said about Tim Hortons because this stupid fucking tab won’t stay in its hole. Fuck this fucking place!!! Can’t even drink my damn coffee 2291
Jim Calabro ♟️ @calabro.io · 26/03/2026@gork generate a picture of me as a fashion model on the runway 120
Jim Calabro ♟️ @calabro.io · 20/03/2026I sent that from my claude one-shot vibe coded terminal client for bluesky, so I'm sure that ~I~ claude did something goofy lmao the tmux of bluesky 010
Jim Calabro ♟️ @calabro.io · 25/02/2026must be using jetstream1.us-east.bsky.network lol we're churning through the backlog now 010
Jim Calabro ♟️ @calabro.io · 25/02/2026This isn't even that big tbh Like, most of this is protobuf codegen changes 130
Jim Calabro ♟️ @calabro.io · 05/02/2026yep! fix is rolling out now, apologies. will take about 30 min. the bsky data plane needed to be more defensive of non-TID rkeys. we need a more robust strategy for handling them, but for now I'm just un-bricking it. TONS of stuff in our code assumes TIDs, and it's not a small change to unwind that 150
Jim Calabro ♟️ @calabro.io · 28/01/2026100% packet loss would be surprising (I see some 100% loss in the screenshot), but we have firewall rules to only allow X ICMP packets per second. So like, really fast ping commands like this will show a sizable amount of packet loss, and that's expected: 110
Jim Calabro ♟️ @calabro.io · 05/11/2025I like rust, zig, and go, and think that PLs these days are better than they ever have been 1270
