Details
-
Bug
-
Resolution: Fixed
-
Critical
-
3.0, 3.0.1, 3.0.2, 3.0.3, 3.1.0, 4.0.0
-
Security Level: Public
-
Untriaged
-
Yes
-
Mar 9 - Mar 27
Description
It can happen in the following scenario.
- user starts upgrading 2.5 cluster to 3.* by rebalancing in a 3.* node n1
- n1 becomes an owner of a vbucket (say 0); at this point vbucket uuid is generated for the vbucket (vbuuid0)
- another rebalance is done (addition/removal, not important) and vbucket 0 moves from n1 to another node n2 (can be both 2.5 or 3.*), n1 becomes vbucket 0 replica node
- another rebalance is done and vbucket 0 moves back to n1
After this vbucket 0 on n1 will still have vbucket uuid vbuuid0. So xdcr might incorrectly reuse previous checkpoints and not replicate data that actually needs to be replicated.
Attachments
Issue Links
- blocks
-
MB-15886 3.1.1 Minor Release
- Resolved
For Gerrit Dashboard: MB-15837 | ||||||
---|---|---|---|---|---|---|
# | Subject | Branch | Project | Status | CR | V |
53784,4 | MB-15837 Generate a new vbucket UUID for TAP-based VBucket takeover. | 3.0.x | ep-engine | Status: MERGED | +2 | +1 |
53820,1 | Merge remote-tracking branch 'couchbase/3.0.x' into sherlock | sherlock | ep-engine | Status: ABANDONED | 0 | -1 |
53834,3 | MB-15837 Fix to tap_notify_set_vbucket_state unit test. | 3.0.x | ep-engine | Status: MERGED | +2 | +1 |
53838,1 | Merge remote-tracking branch 'couchbase/3.0.x' into sherlock | sherlock | ep-engine | Status: MERGED | +2 | +1 |
53841,1 | Merge remote-tracking branch 'couchbase/sherlock' | master | ep-engine | Status: MERGED | +2 | +1 |