Corrupted table because manifest list file of an active snapshot is removed when commit is empty

[priyankar-stripe]

Apache Iceberg version

1.8.1

Query engine

Flink

Please describe the bug 🐞

In https://github.com/apache/iceberg/pull/10523/files, we changed the cleanup logic to stop fetching the latest snapshot from the metastore and instead maintain an in-memory snapshot instance for cleanup operations.
Specifically what we saw happen was:

1. Initial Commit Attempt: Flink attempts to commit snapshot <snapshot_id> to metastore. The commit succeeds on the metastore side, but Flink receives a transient network error and incorrectly marks the commit as failed.
2. Retry with Stale Metadata: RetryingMetaStoreClient retries the commit, but since the table has already been modified, metastore returns a The table has been modified error. This triggers a CommitFailedException (see
   https://github.com/apache/iceberg/blob/1.8.x/hive-metastore/src/main/java/org/apache/iceberg/hive/HiveTableOperations.java#L277-L278).
3. SnapshotProducer Retry: SnapshotProducer catches this exception and retries the operation. It reuses the same snapshot ID but generates a new manifest list file: snap-<snapshot_id>-2-<uuid>.avro (note the incremented attempt number), different from the already-committed manifest list snap-<snapshot_id>-1-<uuid>.avro.
4. No-Op Detection: Since there are no actual changes between these two attempts (same snapshot content), Iceberg detects this as a no-op and skips the commit https://github.com/apache/iceberg/blob/1.8.x/core/src/main/java/org/apache/iceberg/SnapshotProducer.java#L448-L453.
5. Incorrect Cleanup: The cleanup logic then runs, but it incorrectly assumes snap-<snapshot_id>-2-<uuid>.avro is the committed manifest list (since it's the most recent attempt). It therefore deletes snap-<snapshot_id>-1-<uuid>.avro as an "uncommitted" file, thereby corrupting the active snapshot

Willingness to contribute

• I can contribute a fix for this bug independently
• I would be willing to contribute a fix for this bug with guidance from the Iceberg community
• I cannot contribute a fix for this bug at this time