Repository navigation
ANR only sends one event #11160
Description
Problem Statement
Currently the ANR feature only sends a single event and exits.
Solution Brainstorm
We chose to do this because:
- The ANR worker does not run a full Sentry client so there's no Dedupe integration
- If ANR has been detected, the event loop is getting blocked for long periods so we shouldn't expend more CPU time than we need to
Reacted by Ben Delaney and Adrien
Metadata
Metadata
Assignees
Projects
- Status
Activity
joaoviana commentedon Jun 5, 2024
Hey @timfish / Sentry team! Could you clarify this part please? "Currently the ANR feature only sends a single event and exits."
We run a multi-tenant distribution of an app that uses Sentry.
If an ANR event gets capture, and if it's only sent once, what happens for other users? Is it per
deploymentor persession?timfish commentedon Jun 5, 2024
It's currently one event per Node process. I can add an options to lift this limit but there's no easy way to have this per session.
We fetch context (scope, user, tags, etc) from the app thread but we can only do that if it becomes unblocked. If the event loops blocks for more than 5 seconds we give up fetching the extra context and send the event without it.
abustany commentedon Nov 21, 2024
We're using ANR (with a relatively low threshold of 500ms) to diagnose event loop lags in our NodeJS server, and often get an ANR event during startup while Node is compiling stuff... Which means we never get an ANR event afterwards during the actual event loop stall events 😅 So lifting this limitation (or alternatively adding an option to only start the pings after a given duration) would be super useful!
timfish commentedon Nov 22, 2024
Thanks for the ping. I'll look into this next week!