Details
-
Bug
-
Resolution: Fixed
-
Major
-
5.5.0
Description
As observed in CBSE-4539, ep-engine has not sent temporary backoffs (tempOOM) back to clients when the DWQ exceeds a fixed number of items (1M) since 3.0.1 - see http://review.couchbase.org/#/c/46614/.
Instead, it only sends tempOOM when the mem_used of the bucket exceeds a given percentage of the bucket quota - default 99%:
"replication_throttle_threshold": {
|
"default": "99",
|
Given that you'll only normally reach 99% usage after ~all items have been evicted and all the memory is being used for the checkpoints (in a full-eviction bucket), this isn't a very effective limit - it kicks in far too late.
We should investigate an alternative mechanism. Possible ideas:
- Trigger when checkpoint memory usage (CheckpointManager::getMemoryUsage) exceeds a percentage of the bucket quota.
Attachments
Issue Links
- is duplicated by
-
MB-38855 [Magma] Memcached is OOM killed by kernel.
- Closed
-
MB-39300 Cap checkpoint memory usage to % of quota
- Closed
-
MB-40644 Magma S0 tests on 7.0 hit out of memory issue
- Closed
- relates to
-
MB-46827 Limit the Checkpoint memory usage
- Closed
-
MB-26494 Cursor Dropping - Improve Trigger Policy
- Closed
-
MB-29227 Couchbase Server can get into a livelock state due to high checkpoint memory usage
- Closed