StorageGRID CloudMirror Replication Delays and Checkpoint Inconsistency
Applies to
- NetApp StorageGRID 11.9
- Platform Services – CloudMirror Replication
- Multi-site StorageGRID deployments with cross-site replication
Issue
Customers observed the following symptoms after expanding their StorageGRID environment and enabling cross-site CloudMirror replication:
-
Significant delays in S3 object ingest operations.
-
High number of pending and in-progress platform service replication requests.
-
Surges in S3 request duration and large active request queues (up to 1,500+).
-
Target tables in destination buckets became inaccessible due to checkpoint mismatches:
The_last_checkpoint (xxx000) is ahead of the latest checkpoint.parquet (xxxx30) in the target delta_log. This is causing the target table to be inaccessible with the error: LastCheckpoint(xxx000...) mentions a non-existent checkpoint file. -
StorageGRID logs showed repeated errors such as:
WARNING: [endpoint.go:zzz] RequestError: send request failedcaused by: Put "https://<endpoint>": unexpected EOF -
Disabling CloudMirror replication stabilized client performance, but pending queues drained very slowly (e.g., ~50 objects per 20 minutes).
