[LU-9860] conf-sanity test_33a fails with “mount -t lustre failed“ for separate MDS and MGS

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Affects Version/s: Lustre 2.11.0
• Labels: tests
• Environment: Separate MGS and MDS or separate MGT and MDT
• Severity: 3
• Assignee: James Nunez (Inactive)
• Reporter: James Nunez (Inactive)

Description

conf-sanity test 33a fails when run on a Lustre file system configured with a separate MDS and MGS, even with a separate MDT and MGT on the same node. The error message is

conf-sanity test_33a: @@@@@@ FAIL: mount -t lustre failed 

In the test log, we see that mounting the MDS for the newly created file system fails

CMD: onyx-50vm7 mkdir -p /mnt/lustre-fs2mds; mount -t lustre   		                   /dev/lvm-Role_MDS/S1 /mnt/lustre-fs2mds
onyx-50vm7: mount.lustre: mount /dev/mapper/lvm--Role_MDS-S1 at /mnt/lustre-fs2mds failed: Operation already in progress
onyx-50vm7: The target service is already running. (/dev/mapper/lvm--Role_MDS-S1)

This failure looks like it is due to the mgs flags used when the MDT is formatted; “—mgs” flag. When an MGS already exists, specifying that this node will be the MGS for the new file system causes problems.

In addition, the lctl call in the following code needs to be run on the MGS

2418         do_facet $SINGLEMDS "$LCTL conf_param $FSNAME2.sys.timeout=200" ||
2419                 error "$LCTL conf_param $FSNAME2.sys.timeout=200 failed"

We see similar failures for conf-sanity tests 43b, 53b, 54b

Logs that capture test 33a failures are at
https://testing.hpdd.intel.com/test_sets/5176f130-729c-11e7-a0a2-5254006e85c2
https://testing.hpdd.intel.com/test_sets/8df40f4c-729e-11e7-a0a2-5254006e85c2

Note: This tickets description was modified to state the problem more clearly and not only subscribe a solution.

Issue Links

• is related to: LU-8688 All Lustre test suites should run/PASS with separate MDS and MGS
• is related to: LU-10717 several conf-sanity tests failed: FAIL: start mgs failed