Details
-
Bug
-
Resolution: Fixed
-
Critical
-
master
-
None
-
Untriaged
-
1
-
Unknown
-
KV-Engine 2021-March
Description
Spotted whilst test MB-45144 the following sequence results in an increase of the counter when it should result in no change.
- store(k1, "value")
- flush
- this is an insert, so counter increases e.g. size:32 (key+value+meta)
- delete(k1)
- flush
- this is a delete (but really just a new write of k1 but with no value) so counter decreases, as we account tombstones (key+meta remains) size:18
Next another delete, this requires a store/delete before flush, de-dup means we only flush the delete, but a store is required so that the delete doesn't fail with "no-key".
- store(k1, "value")
- delete(k1)
- flush
So we only now flush a delete, the disk is logically unchanged but the stat code treats this as a insert and size goes up, i.e. size becomes 18 + 18
Attachments
Issue Links
- blocks
-
MB-44826 [UI][stats] Total disk usage in bucket details is not matching with what is shown at scopes/collections level.
- Closed
-
MB-45469 CouchKVStore::compactDB: exception while performing compaction for vb:752 - Details: cb::couchstore::replay() - precommit hook Failed: error reading file
- Closed