Unassigned shards when replica count updated to 0

We are using ES 2.4.0. We are noticing un-assigned shards in the below scenario:

  1. Have 3 nodes designated as data nodes. (4 indexes with only 1 shard for each index)
  2. Replica count set as 2 --> Cluster state GREEN
  3. Bring down 1 of the node
  4. Replica count set as 1 --> Cluster state GREEN
  5. Bring down another node (Left with only one "data" node)
  6. Replica count set as 0 --> Cluster state RED

During Step 6, noticed that we get a cluster state as RED for those indices for which primary shard was the node that went down. (If we had all primary shards allocated to the node that was left in the cluster, we don't see an issue)

We have "cluster.routing.allocation.enable" set to "ALL". Anything else, we are missing due to which the shards are not being allocated to the only node thats left ?

The issue was due to updating the replica count to 0 before the node goes down. If we shutdown the node and then update the replica count to 0, it works fine.

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.