We pace file deletions to avoid spikes in IO. However, the data directory is never synced, so IO is not necessarily scheduled yet. This could theoretically allow a buildup of unsynced file deletions. When the data directory is finally synced, IO may spike.

Noticed during the investigation of cockroachlabs/support#2673. It's unlikely to be the root cause of that particular.

Jira issue: PEBBLE-66