Details
-
Bug
-
Resolution: Fixed
-
Critical
-
2.2.0
-
Security Level: Public
-
Triaged
Description
This JIRA ticket is due to this - http://review.couchbase.org/#/c/30388/
Basically, customers are seeing beam.smp memory growth, in non-views, non-XDCR use cases. Alk's and Aaron's analysis showed many erlang processes (eprocesses) in couchdb code, due to having several eprocesses per vbucket, and this gets worse as customer was using multiple buckets.
Solution is to leverage erlang hibernation for the top memory hog eprocesses.
Attachments
Issue Links
- relates to
-
MB-9637 Rebalance-In 5 nodes on empty 18 buckets is very slow per bucket
- Closed
For Gerrit Dashboard: MB-9620 | ||||||
---|---|---|---|---|---|---|
# | Subject | Branch | Project | Status | CR | V |
30388,4 | MB-9620 Hibernate idle per-vbucket gen_servers | 2.1.1 | couchdb | Status: ABANDONED | +1 | +1 |
35933,5 | MB-9620: drop mccouch notifications on the floor | master | ns_server | Status: MERGED | +2 | +1 |
36025,2 | MB-9620: Revert "MB-6413: open all databases on startup" | master | couchdb | Status: MERGED | +2 | +1 |