Arne Welzel

As discussed a bit with @J-Gras: It would be cool if we had telemetry of the number of publishes and the size of the resulting serialized message. In a Prometheus...

> If the unit is public we can never make this optimization since hooks might be implemented in C++ which we cannot inspect. If the field is anonymous, one can't...

> Do you share that quick analysis or might there be something else involved? I think it's this call to fnv1a32 causing pool-related differences as it's now computing a different...

> Too many open files (src/kqueue.cpp:35) Hmm - what's your `ulimit -n`. I'm not sure sure there's all that much we can do here. It's unfortunate libzmq aborts on this...

> In Zeek 7.1.0, it used to produce the following record in http.log: > In Zeek 7.2.0 rc1, these lines are not produced. > I believe the fact that we're...

> Was there a reason we did not want to do that? Unclear if it's actually interesting/valuable. And needs a new event `analyzer_attached(tag: AllAnalyzers::Tag, info: AnalyzerAttachedInfo)`, but that might be...

> If one sets dpd_buffer_size to, e.g., 2048 - http.log will show up again, as before. Not adding much to the discussion - feel free to ignore: To me this...

I now have btest based reproducer using Python. The crux is actually sending the pong replies: It's the reply thread (sending pongs) that detects the closed socket and raises the...

> Hmm... need to stare a bit. Upstream issue: https://github.com/machinezone/IXWebSocket/issues/551

Appeared here again: https://cirrus-ci.com/task/4729416351219712?logs=test#L20