Details
-
Improvement
-
Resolution: Fixed
-
Major
-
7.6.0
-
0
Description
ns_server uses the ep_temp_oom/ep_oom counters for differentiating between recoverable (aka Temporary) and unrecoverable (aka Hard) OOM states.
The Hard side of the thing is just a mis-interpretation.
In memcached we have:
- ep_temp_oom: Memory pressure (defined by some internal thresholds), but memory still below the Bucket Quota
- ep_oom: Memory has reached the Bucket Quota
That has nothing to do with whether the OOM state is recoverable or unrecoverable. Actually, the logic in memcached always tries to recover from OOM state, so there's no really concept of "unrecoverable OOM" in memcached except for specific scenarios like value-eviction + doc metadata eats the entire quota, or ephemeral/fail_new_data buckets.
The mis-interpretation above generates misleading "unrecoverable OOM" alerts.
Could we improve that?
Thanks
Attachments
Issue Links
- is parent task of
-
DOC-12070 Doc: Misleading "Hard OOM" alerts on WebUI
- In Progress