-
Type: Task
-
Resolution: Fixed
-
Priority: Major - P3
-
Affects Version/s: None
-
Component/s: Not Applicable
-
Storage Engines
-
5
-
2024-05-28 - FOLLOW ON SPRINT
-
v8.0
While debugging various failures in the pre-fetch functionality, we found that adding a delay before pre-fetch threads pop off pages from the pre-fetch queue resulted in faster reproducers. Consider turning this into a timing stress failpoint to assist with detecting future problems.
@@ -82,6 +82,7 @@ __wt_prefetch_thread_run(WT_SESSION_IMPL *session, WT_THREAD *thread) WT_RET(__wt_scr_alloc(session, 0, &tmp)); while (F_ISSET(conn, WT_CONN_PREFETCH_RUN)) { + __wt_sleep(0, 5000); /* * Wait and cycle if there aren't any pages on the queue. It would be nice if this was * interrupt driven, but for now just backoff and re-check.