bugfix: batch processor doesn't work with uwsgi

[srikanthccv]

Description

The change introduced in #2242 didn't work for uwsgi since its worker process management is implemented in C and _at_fork_reinit is never invoked. We solve this by invoking it manually when we find process ids do not match.

Fixes #2266

Type of change

Please delete options that are not relevant.

• Bug fix (non-breaking change which fixes an issue)

[Cruuncher]

I need to use opentelemetry with uwsgi.

Are there plans for this to release? Can I work around this by monkey patching in this implementation of BatchSpanProcessor?

[srikanthccv]

Hi @Cruuncher ,

We have documented here how to use the uwsgi hooks here. I didn't get the time to test this change. I will revisit it this weekend.

[Cruuncher]

@srikanthccv Thank you you for pointing me to the relevant documentation! I've been really spinning my wheels trying to get this working, and googling doesn't really find the answers 😓

This appears to be working for us now

[srikanthccv]

cc @aabmass

[lzchen]

This fix would probably solve all cases in which os.register_at_fork isn't available when a new process is created correct? After this change, I think the only case we are missing is when os.register_at_fork is not available (like in Windows) and when a new process is not created (simply a new fork is created).

[srikanthccv]

and when a new process is not created (simply a new fork is created).

When does this happen?

One thing I am still thinking about is how do I do the same for PeriodicExportingMetricReader? Unlike tracing/logging where the events are received by processor, the PMR collects them with collect function. There is nothing like emit or on_start/on_end where I can check if the pids to see the fork hook is invoked and invoke if not.

[lzchen]

When does this happen?

Oh I guess it wouldn't happen for the cases we care about (only when child processes are spawned).

Different question: Is there still a need for os.register_at_fork if we can compare pids all the time?

[srikanthccv]

Is there still a need for os.register_at_fork if we can compare pids all the time?

It's fine to still have it. With the hook the reinit happens immediately but with pids records need to arrive at emit/on_end then it invokes the reinit (this code path very less likely compared to fork hook).

[lzchen]

One thing I am still thinking about is how do I do the same for PeriodicExportingMetricReader?

I am okay to discuss this in a separate issue/pr to unblock this.