Details
-
Bug
-
Resolution: Fixed
-
Critical
-
4.0.0
-
Security Level: Public
-
Untriaged
-
No
Description
Found via code review during MB-15174.
If MAINT_STREAM is backlogged, there can be many TS in timekeeper queue waiting to be flushed(lets say T1-T10). Lets say there is an index being built in parallel and it reaches its build completion. Index will move to catchup and the corresponding mutations will come in MAINT_STREAM for the new index from T11. But right now the merge can happen anywhere between T1-T10, which is not correct.
Indexer needs to make sure the merge happens at or after T11.
Attachments
For Gerrit Dashboard: MB-15196 | ||||||
---|---|---|---|---|---|---|
# | Subject | Branch | Project | Status | CR | V |
51955,3 | MB-15196 Consider in progress flush during stream merge | unstable | indexing | Status: MERGED | +2 | +1 |
52770,9 | MB-15196: projector.AddInstances() returns current timestamp. | unstable | indexing | Status: MERGED | +2 | +1 |
53116,8 | MB-15196 Use CurrentTs From AddInstance Response for Merge | unstable | indexing | Status: MERGED | +2 | +1 |