[META] Segment Replication Issue list

[Poojita-Raj]

Please comment on this issue with any feedback on our recent experimental release.

This is a meta issue for all work relating to Segment Replication.

• [Segment Replication] Design Proposal #2229
• [Segment Replication] Experimental Release Tracking #3969
• [Meta] Promote Segment Replication out of experimental. #5147

Performance Improvements:

• [Segment Replication] Optimise network bandwidth on primaries #4245
• [META] [Segment Replication] Integration with remote store #4448
• [Segment Replication & Remote Translog] Back-pressure and Recovery for lagging replica copies #4478

Compatibility:

• [Segment Replication] Support for mixed cluster versions (Rolling Upgrade) #3881
• [Segment Replication] Review Cross Cluster Replication compatibility with Segment Replication Enabled #3823

Bugs:

• [BUG] [Segment Replication] testCleanupAndPreserveLatestCommitPoint test failure #4538
• [BUG] [Segment Replication] Flaky engine AlreadyClosedException exception on closed index action #4530
• [Segment Replication] Synchronize checkpoint updates with failover. #4136

Benchmarking:

• [Segment Replication] Benchmarking Segment Replication. #2583

Future Extensions:

• [Segment Replication] Support enabling/disabling segment replication for an existing index. #3685
• [Segment Replication] Trigger segment copy after initial recovery. #4036
• [Segment Replication] Update ReplicationTarget and ReplicationListener failure exception type. #4291
• [Segment Replication] Replicate *.liv file may cause performance issue #3929
• Use pull based model for Segment Replication #4577

Completed:

• [Segment Replication] Merge back to main tracking issue. #2355
• [Segment Replication] - Toggle replication strategy based on segrep index setting. #2197
• [Segment Replication] - Clean up transport layer communication when publishing checkpoints. #2199
• [Segment Replication] Refactor SegmentInfosRef to top level class. #2202
• [Segment Replication] Fix delete operations to only write to xlog on replicas. #2330
• [Segment Replication] - Fix corrupt index when verifying after file copy. #2331
• [Segment Replication]. Support graceful replication failure. #2332
• [Segment Replication] Fix replicas to correctly update their processed local checkpoint. #2358
• [Segment Replication] Clean up Recovery State transitions. #2588
• [Segment Replication] Implement Replaying on new checkpoints received by replica shard #3558
• Bugs:
  • [BUG][Segment Replication] Gradle check failures #2414
  • [Segment Replication] - Fix Integ test with reproducible seeds #2285
  • [Segment Replication] Bug - Missing store stats from /_cat/shards API. #2431
  • [Segment Replication] Bug - Replica shards are incorrectly trying to send checkpoints. #2815

[kartg]

Capturing pending refactor work from the last refactor PR - #2468

• Remove ShardRoutingReplicationListener and use a single class based on extracting PeerRecoveryTargetService.RecoveryListener instead - Refactoring of Segment Replication classes #2468 (comment)
• Refactor the *Collection classes and merge common functionality - Refactoring of Segment Replication classes #2468 (comment)

[kartg]

Action items from another PR - #2551

• Improve replication lifecycle management by avoiding distributed ownership across classes - [Segment Replication] Ensure replica's store always contains the previous commit point. #2551 (comment)
• Update segment replication test verification steps to validate file structure (and therefore point-in-time consistency) over and above hit count - [Segment Replication] Ensure replica's store always contains the previous commit point. #2551 (comment)

[navneet1v]

I want to understand few cases and how they are handled overall, my understanding is based on this GH issue: #2229

1. If during syncing of segments to replicas if the node containing primary dies how the cluster will recover?
2. What will happen to the shard re-routing process when a node left the cluster during the fsync is happening?