Add code to fix the red index on close

Signed-off-by: Ashish Singh <[email protected]>

server/src/main/java/org/opensearch/index/translog/TranslogWriter.java

@@ -405,13 +405,25 @@ public int totalOperations() {
 
     @Override
     synchronized Checkpoint getCheckpoint() {
+        // On a remote store enabled cluster, the global checkpoint is always the max seq no of the shard.
+        // case 1- Whenever a sync translog happens, we create an empty translog writer (TR) with value as -1
+        // (SequenceNumbers.NO_OPS_PERFORMED). If we try to sync the translog on account of ensureSynced / flush / roll generation, then
+        // checkpoint contains value as -1.
+        // Case 2 - In other cases where at-least operation has been added to TR, then the maxSeqNo is set
+        // correctly in the checkpoint. Since the in-memory global checkpoint is read from translog checkpoint, we set the same correctly as
+        // max of seqNo and globalCheckpoint supplied by the supplier.
+        // In case 1, the max value would be supplied global checkpoint. In case 2, it would be the maxSeqNo.
+        assert maxSeqNo == SequenceNumbers.NO_OPS_PERFORMED || maxSeqNo >= globalCheckpointSupplier.getAsLong();
+        long globalCheckpoint = remoteTranslogEnabled
+            ? Math.max(maxSeqNo, globalCheckpointSupplier.getAsLong())
+            : globalCheckpointSupplier.getAsLong();
         return new Checkpoint(
             totalOffset,
             operationCounter,
             generation,
             minSeqNo,
             maxSeqNo,
-            globalCheckpointSupplier.getAsLong(),
+            globalCheckpoint,
             minTranslogGenerationSupplier.getAsLong(),
             SequenceNumbers.UNASSIGNED_SEQ_NO
         );