Hey Livekit folks. We’ve been trying to track down the source of a highly noisy (600k+ errors / month) error

Error: ESRCH: no such process, read

with the

@livekit/agents-js

package. The error seems to start triggering after a while and slowly ramps up in volume until we restart our service. We see the same number of errors every minute until something causes the error count to increase and then it stabilizes again. This pattern repeats. We are self hosting and have seen this same error occur since before agents v1.0.15 and still occurs on v1.0.25. Note that in the attached error per minute graph everytime the errors drop back down to 0 is right after a deploy. I have a hunch that the root cause is in the supervised_proc.ts where you are calling

getchildMemoryUsageMB

inside of a

setInterval

. On line 215 there’s a call to

pidusage(pid)

which checks the process data. Since there’s a 5s interval this is most likely a race condition.

pidusage

can throw system errors including

ESRCH

and in the pidusage npm package they say to not wrap calls to

pidusage()

inside of `setInterval`: >

// Avoid using setInterval as they could overlap with asynchronous processing

I think at a minimum the call to

this.getChildMemoryUsageMB

needs error handling. Ideally a refactor to not wrap this call with a

setInterval

as well.

Sounds like you are on the right track... I don't see any GH issues related to ESRCH, and I don't think this is a known problem. You mention you are self-hosting, does the issue go away if you add error handling to the

getChildMemoryUsageMB

call? I agree a refactor would be best, but that does at least isolate the root cause