getsentry
Profiling for eventlet
Problem Statement
The flamegraph looks very incomplete if eventlet is used. User turns off eventlet, then finds the flamegraph has a lot more information.
The problem may be:
Due to the async nature of the eventlet, frames do not usually stay on the stack for long. Combined with some biases in when the profiler samples, the profiles can look like that screenshot where it's only able to capture the frames related to eventlet.
Solution Brainstorm
Support for Eventlet.
yea we're aware of the shortcomings here, but this is basically uncharted territory and I think @Zylphrex has already given some thought into making the async user story better here.
Closing in favor of https://github.com/getsentry/sentry-python/issues/1949
