Merge e0182b85e0 into e909be6380

Fixes incorrect application of ilm expiry rules on versioned objects
when replication is enabled.

Regression from https://github.com/minio/minio/pull/20441 which sends
DeleteObject calls to all pools. This is a problem for replication + ilm
scenario since replicated version can end up in a pool by itself instead of
pool where remaining object versions reside.

For example, if the delete marker is set on pool1 and object versions exist on
pool2, the second rule below will cause the delete marker to be expired by ilm
policy since it is the single version present in pool1
```
{
  "Rules": [
   {
    "ID": "cs6il1ri2hp48g71mdjg",
    "NoncurrentVersionExpiration": {
     "NoncurrentDays": 14
    },
    "Status": "Enabled"
   },
   {
    "Expiration": {
     "ExpiredObjectDeleteMarker": true
    },
    "ID": "cs6inj3i2hp4po19cil0",
    "Status": "Enabled"
   }
  ]
}
```

cmd/erasure-server-pool.go

@@ -1185,6 +1185,13 @@ func (z *erasureServerPools) DeleteObject(ctx context.Context, bucket string, ob
 		return z.deleteObjectFromAllPools(ctx, bucket, object, opts, noReadQuorumPools)
 	}
 
+	// All replication requests needs to go to pool with the object.
+	if opts.ReplicationRequest {
+		objInfo, err = z.serverPools[pinfo.Index].DeleteObject(ctx, bucket, object, opts)
+		objInfo.Name = decodeDirObject(object)
+		return objInfo, err
+	}
+
 	for _, pool := range z.serverPools {
 		objInfo, err := pool.DeleteObject(ctx, bucket, object, opts)
 		if err != nil && !isErrObjectNotFound(err) && !isErrVersionNotFound(err) {