Matt Weber created SOLR-8743: -------------------------------- Summary: Data loss on shard recovery and leader election Key: SOLR-8743 URL: https://issues.apache.org/jira/browse/SOLR-8743 Project: Solr Issue Type: Bug Components: SolrCloud Affects Versions: 5.5 Reporter: Matt Weber
SolrCloud with 3 external ZK nodes in quorum, 2 data nodes (NodeA and NodeB). Single collection that has a single shard and a single replica (replication factor 2). The primary shard is initially on NodeA and the replica on NodeB. 1. Index Doc1 via NodeA ./solr/bin/post -c people -d '<add><doc><field name="id">1</field></doc></add>' 2. Shutdown NodeA, NodeB becomes the primary. Doc1 is searchable. 3. Delete Doc1 and index Doc2 via NodeB. ./solr/bin/post -c people -d "<delete><id>1</id></delete>" ./solr/bin/post -c people -d '<add><doc><field name="id">2</field></doc></add>' 4. Shutdown NodeB, no nodes online. ZK still up. 5. Start NodeA, it remains in "down" status since NodeB is down and it was last seen as the primary. 6. Start NodeB, it comes online in "down" status. NodeA is elected leader and sync starts. Doc2 is gone, Doc1 exists in both shards. -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org