Details
-
Bug
-
Resolution: Fixed
-
Test Blocker
-
7.2.0
-
Untriaged
-
0
-
Unknown
Description
First reported here: https://issues.couchbase.com/browse/MB-55268?focusedCommentId=660111&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-660111
We're seeing very high (over 100%) value for couch_docs_fragmentation in some of the perf tests.
I found a run that has cbcollects: http://perf.jenkins.couchbase.com/job/rhea-5node2/3193/consoleFull
The access phase shows couch_docs_fragmentation as 140%:
I checked the cbcollect for all 4 nodes looking at stat (db_data_size, db_file_size) - plotted as the first chart (left most). They look ok to me. DB size is around 1285GB Data size is 860GB. So the couch_docs_fragmentation should be 33%.
Since the computation happens on ns-server side, request the team to take a look as the source stats from kv/magma side look fine.
Attachments
Issue Links
- causes
-
MB-55780 Investigate: KV Compaction Time has Significantly Improved Between the 7.2.0-5186 and 7.2.0-5212
- Closed