Skip to content

Commit 2af28c9

Browse files
committed
test: size the daemon-application wait budget for TSan's slowdown tail
The suite's observation waits shared a 2000 ms budget that is a hang detector, not a latency assertion — every waiter returns the moment its condition holds, so green runs never pay it. Under ThreadSanitizer's 5-20x slowdown on a loaded 3-core macOS CI runner the cancel-delivery wait lost the tail of that distribution once in seven otherwise-green TSan rounds (the cancel itself had succeeded; only the fake worker's counter observation timed out). The budget is now 10000 ms. Verified: nine consecutive daemon_application runs under native macOS TSan with the widened budget, all green, zero ThreadSanitizer warnings. Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
1 parent 5a5ed51 commit 2af28c9

1 file changed

Lines changed: 6 additions & 1 deletion

File tree

‎tests/test_daemon_application.c‎

Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -27,7 +27,12 @@
2727
#include <stdlib.h>
2828
#include <string.h>
2929

30-
enum { APP_TEST_TIMEOUT_MS = 2000, APP_TEST_PATH_CAP = 1024 };
30+
/* Observation-wait budget. This is a HANG DETECTOR, not a latency
31+
* assertion: every waiter returns the moment its condition holds, so green
32+
* runs never pay it. It must absorb ThreadSanitizer's 5-20x slowdown on a
33+
* loaded 3-core CI runner — at 2000 ms the cancel-delivery wait lost the
34+
* tail of that distribution once in seven otherwise-green TSan rounds. */
35+
enum { APP_TEST_TIMEOUT_MS = 10000, APP_TEST_PATH_CAP = 1024 };
3136

3237
typedef struct {
3338
char runtime_parent[APP_TEST_PATH_CAP];

0 commit comments

Comments
 (0)