[SOLR-11656] TLOG replication doesn't work properly after rebalancing leaders.

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 7.1
• Fix Version/s: 8.7
• Component/s: replication (java)
• Assignee: Erick Erickson
• Reporter: Yuki Yano
• Created: 17/Nov/17 10:48
• Updated: 04/Nov/20 15:21
• Resolved: 24/Jul/20 20:54

Description

With TLOG replica type, the replication may stop after invoking rebalance leaders API.
This can be reproduced by following steps:

1. Create SolrCloud with TLOG replica type.
2. Set perferredleader flag to some of no-leader nodes.
3. Invoke rebalance leaders API.
4. The replication stops in nodes which were "leader" before rebalancing.

Because the leader node doesn't have the replication thread, we need to create it when the status is changed from "leader" to "replica". On the other hand, rebalance leaders API doesn't consider this matter, and the replication may stop if the status is changed from "leader" to "replica" by rebalance leaders.

Note that, we can avoid this problem if we reload or restart Solr after rebalancing leaders.

Issue Links

• is related to: SOLR-11794 PULL replicas stop replicating after a RELOAD collection action
• relates to: SOLR-10233 Add support for different replica types in Solr