-
Bug
-
Resolution: Unresolved
-
Medium
-
None
-
None
-
None
-
<distro/arch>, <zfs|ldiskfs>
-
3
-
9223372036854775807
sanity-hsm test_606 stops mds1, mounts the MDT device directly as ldiskfs/zfs, and registers a single EXIT trap: unmount_fstype mds1; start mds1 .... If that umount fails (EBUSY), set -e aborts the trap before start mds1 runs, so MDT0000 stays down. The next test, test_607b, then hangs until the test session times out (24 h here).
Fix: split the cleanup into separate stack_trap commands. First retry the umount for up to 60 s, then run unmount_fstype mds1 || true, then start mds1. A failed umount can then no longer skip the restart, and unmount_zfs still restores the dataset mountpoint property.