Details
-
Bug
-
Resolution: Fixed
-
Critical
-
7.6.0
-
Untriaged
-
0
-
Unknown
Description
Consider a cluster with the following layout after rebalance -
n1: s0-0 - idx1, idx2 (new), idx3 (new)
n2: s0-1 - idx1 (shard repair by copying s0-0 from n1)
n3: s0-2 - idx2 (shard repair by copying s0-0 from n1)
In this case, rebalance can fail when we try to create idx2 and idx3 on s0 as it will locked for copy and will get unlocked once the shard token moves to Ready state. With emptyNodeBatching enabled, transition to ShardTokenReady happens only after DCP rebalance is done. Therefore, DCP rebalance would fail on source if it tries to create indexes on a shard that is being copied to destination.
handle the same