Segment Replication - Recover all translog ops when flipping to write…

…able engine.

Signed-off-by: Marc Handalian <[email protected]>

server/src/main/java/org/opensearch/index/shard/IndexShard.java

@@ -4314,7 +4314,9 @@ public void close() throws IOException {
 
         // When the new engine is created, translogs are synced from remote store onto local. Since remote store is the source
         // of truth for translog, we play all translogs that exists locally. Otherwise, the recoverUpto happens upto global checkpoint.
-        long recoverUpto = this.isRemoteTranslogEnabled() ? Long.MAX_VALUE : globalCheckpoint;
+        // We also replay all local translog ops with Segment replication, because on engine swap our local translog may
+        // hold more ops than the global checkpoint.
+        long recoverUpto = this.isRemoteTranslogEnabled() || indexSettings().isSegRepEnabled() ? Long.MAX_VALUE : globalCheckpoint;
         newEngineReference.get()
             .translogManager()
             .recoverFromTranslog(translogRunner, newEngineReference.get().getProcessedLocalCheckpoint(), recoverUpto);