-
Bug
-
Resolution: Unresolved
-
Medium
-
None
-
None
-
None
-
3
-
9223372036854775807
[Claude] "lhsmtool_posix --daemon" calls daemon() (lustre/utils/lhsmtool_posix.c:2152) before llapi_hsm_copytool_register() (:2180), so the parent returns before the copytool has registered with the MDTs. In sanity-hsm, "copytool setup" therefore returns early, and a following copytool_suspend can SIGSTOP the copytool before it registers. HSM requests then stay WAITING until the test times out.
Seen while testing sanity-hsm test_251 on a 3-node cluster with 12 MDTs: the request stayed WAITING for 200s, and the copytool log had only its startup line, with no "waiting for message from kernel". This affects every sanity-hsm subtest that suspends the copytool right after starting it, at any MDT count; more MDTs make registration slower and the window larger.
Fix: have --daemon fork with a notification pipe, and have the parent exit only after the child has registered (exit 1 if the child dies first). sanity-hsm test_31a, test_260b and test_260c start a redundant second copytool that always fails to lock the pid file; with the fix that failure becomes visible, so those calls should be removed.
- is related to
-
LU-14058 Create tests for large number of MDTs
-
- Reopened
-