Details
-
Bug
-
Resolution: Duplicate
-
Critical
-
7.1.0
-
None
-
Untriaged
-
1
-
Unknown
Description
Steps:
The panic/indexer restart is introduced in indexer:mergePartition function, just before sending MsgClustMgrMergePartition message to the manager.
1. Create 1 node cluster. n1ql, kv, index.
2. Create a partitioned index with 2 partitions - deferred build = true.
3. Add an indexer node.
4. Rebalance. One of the partitions will be moved to the new node. This will create a tombstone instance on the first node.
5. Remove the new node from the cluster and rebalance again.
6. During rebalance, the partition will move back to the first node. Merge partition will start, but due to panic, it will fail.
7. Indexer will be restarted and the message Storage corrupted and unrecoverable will be seen in the logs.
The fix for MB-51025 doesn't work for deferred indexes as the deferred indexes have state "INDEX_STATE_READY" which is not considered a valid recovery state.
More analysis is required to handle this scenario. CC Deepkaran Salooja
Attachments
Issue Links
- relates to
-
MB-51025 [System Test] Seeing "Storage corrupted and unrecoverable" error for one index partition
- Closed