Reducing latency with esoteric Linux process flags
recall.aiIs this exactly the same bug as this very recent post?
Yes, it is. But clearly in a different codebase. Crazy that two teams were using this same esoteric, broken prctl and diagnosing the problem it caused in such short succession.
Haha, yea. When I was writing this blog post (we had already pushed to production) I came across that post when searching for sources to confirm my findings. Antonio and I were so surprised at the coincidence.
What? The headline does not match the article content at all.
There was no reduced latency. The article title should be something along the line of:
"Bad architecture resulted in a hard-to-troubleshoot bug"
I suppose the latency was reduced because the process wasn't being terminated by the kernel anymore.
Really, the title should say "PR_SET_PDEATHSIG doesn't mean what you think it means."
The man page for PR_SET_PDEATHSIG: "The parent-death signal is sent upon subsequent termination of the parent thread and also upon termination of each subreaper process"
I was hoping it was something about reducing process latency by using some obscure IOCTL call or switching the scheduler.
I think the latency reduced was the 12 seconds to start playing the video via the ai agent. However I was also expecting some mad kernel microsecond based stuff also.