-
Type: Bug
-
Resolution: Fixed
-
Priority: Major - P3
-
Affects Version/s: 6.0.3, 6.0.4
-
Component/s: None
-
Fully Compatible
-
ALL
-
Sharding EMEA 2023-02-06
-
59
During the balancer round the FCV is checked twice in the following order:
A) On the selectChunksToMove phase: we check if gBalanceAccordingToDataSize flag is enabled, which is enabled on FCV6.0 and disabled on FCV5.0. If the flag is enabled, we don't store the max key value of the chunk to migrate since we want to call moveRange intentionally without the max key in order to leave the decision of how much data migrate in hands of the shard.
B) On the moveChunks phase: we check if FCV is 6.0 or not to either send a moveRange or a moveChunk command to the shard that owns the chunk.
In case of having a downgrade FCV in between A and B, the moveChunk command request will be scheduled without a max key value and this invariant will be hit.
- is caused by
-
SERVER-65816 Change balancer policy to balance on data size rather than number of chunks
- Closed