SnapshotTooOld can occur when an operation’s read snapshot is no longer available before the operation completes. In these cases, the command should be treated as transient and retried with a newly established snapshot rather than surfaced as a permanent failure.

There are currently at least two examples:

• CheckMetadataConsistencyInBackground (CMC): SERVER-133516 proposes aborting the current check and retrying the full check later with a fresh snapshot timestamp.

• Catalog cache refresh: The catalog-cache implementation is another internal path that should handle SnapshotTooOld through the common retry mechanism.

The investigation should:

• Identify other internal commands that should receive the same behavior.

• Identify an appropriate common layer for classifying SnapshotTooOld as retryable for internal commands and define safe retry semantics.