In #81389, I originally moved where Pebble's disk-health checking is layered onto the filesystem. Disk-health checking monitors for excessively slow writes, crashing the process if writes appear to be stalled in the kernel. The changes in #81389 caused disk-health checking to be used in TestExternalDistinct. TestExternalDistinct began to fail with leaked goroutines in the disk-health checking. I dug into it a bit, and traced it to a diskQueue that is never Closed, causing its writeFile to never be closed.

For now I should be able to refactor #81389 to continue omitting disk-health checking in this code path and unblock the vendor bump, so this isn't a blocker for me.

Jira issue: CRDB-15202