Details
-
Bug
-
Resolution: Fixed
-
Major
-
6.5.0
-
Untriaged
-
No
-
Storage-Sprint-Feb-08-2019, Storage-Sprint-Mar-15-2019
Description
Plasma Index is a skip list with pages at the lowest level.
If a time series workload is used on plasma, where new keys only get appended to the tail of the skip list while documents from the start of the skip list just expire away, that puts a lot of load on the mvcc purger which is unable to keep up.
This causes the purge ratio to shoot up and data_size to explode when customers use expiry or TTL on the kv documents.
Current workarounds recommended:
The command to reduce the number of page deltas are as follows..
curl node.ip.address:9102/settings -u Administrator:password -X POST -d "{\"indexer.plasma.mainIndex.maxNumPageDeltas\":30}" |
curl node.ip.address:9102/settings -u Administrator:password -X POST -d "{\"indexer.plasma.backIndex.maxNumPageDeltas\":30}" |
Attachments
Issue Links
- duplicates
-
MB-40855 Long scan latencies when removing a subset of docs
- Closed