You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
## Description
The OOM kill message re-computed the memory threshold at kill time via
`GetMemoryThreshold()`, which under `--enable-resource-isolation` could
read a different cgroup `memory.max` value than at init time (e.g. "max"
instead of digits), silently falling back to 0.95. Fix: have each
monitor pass the threshold it actually fired on through the callback
instead of letting NodeManager guess.
Now the OOM msg should be something like this
```
Memory on the node (IP: 10.0.0.1, ID: abc123) was 7.20GB / 8.00GB (0.900000);
OOM kill reason: Memory usage 7728742400B exceeded threshold of 6871947674B (85.9% of 8589934592B total);
Object store memory usage: [...];
Ray killed 1 worker(s) based on the killing policy: [...];
```
## Related issues
> Link related issues: "Fixesray-project#1234", "Closesray-project#1234", or "Related to
ray-project#1234".
## Additional information
> Optional: Add implementation details, API changes, usage examples,
screenshots, etc.
---------
Signed-off-by: You-Cheng Lin <mses010108@gmail.com>
Signed-off-by: You-Cheng Lin <106612301+owenowenisme@users.noreply.github.com>
Signed-off-by: phattruong <23120318@student.hcmus.edu.vn>
0 commit comments