sendmsg_x silently falls back to single-packet mode instead of batching when you use unconnected sockets too.
The fix is just to use a connected socket to talk to each WireGuard peer. kqueue then wakes us reliably when the socket is ready too.
Found the source of our mysterious packet loss issue on macOS and iOS: Apple's BSD sockets API fails to surface ENOBUFS for *unconnected sockets* - sockets which have no specific destination.
The datagrams (or batches in our case) are just silently dropped.
No backpressure.
🚨 Ongoing supply chain attack on Composer packages! We just found multiple laravel-lang/* packages compromised on Packagist (lang, http-statuses, attributes). Payload runs at autoload time. At least 50 package versions were compromised.
If you installed a compromised version, the malware already executed. Pin to a clean COMMIT (not version) and rotate secrets immediately.
If your lockfile already had an older commit from before today, you are safe. But you should not update at the moment.
I don't know who needs to hear this, but:
Replacing an explicit button or form in your app that works 100% of the time with an AI chat bot that works 50% of the time is a UX disaster.
Attempting to generate a reliable repro for a bug in some of our open source code (not sec-related).
It's failing with:
"API Error: Claude Code is unable to respond to this request, which appears to violate our Usage Policy"
@claudeai is this a known issue?
At @firezonehq we've done a TON of work recently to make Gateways more performant with lots of clients (>1000) connected.
Thanks to work by our Rust team on the STUN / TURN signaling code paths we've managed to increase client capacity more than 20x!
https://t.co/iQUyDbR8Av
@chris_mccord Interesting! How does spinning down the old VM work? Can it migrate processes over (e.g. connected Phoenix Channels)?
We ended up doing blue - green using containers with an iptables cutover but avoiding the reconnect storm would be super nice.
Amazing how even "small" packet pacing / loss problems can tank throughput, especially with any kind of latency.
Just fixed an issue in our eventloop to more fairly process rx packets while under high tx load - a three-line change.
Got a 3-4x throughput increase on a 100ms RTT simulated link: