The timeofdeath field in the WT_DATA_HANDLE structure is read and written by both the sweep server and the cursor->close API. It is not atomic or protected by a lock. I believe this is a race condition, but it's (a) very rare (since the sweep server runs relatively infrequently) and (b) of little consequence.

If it gets covnerted to an atomic, I believe it can use a relaxed memory ordering.

TSAN will possible pick it up once that gets rolled out more widely.