Details
-
Bug
-
Resolution: Cannot Reproduce
-
Minor
-
None
-
Lustre 2.8.0, Lustre 2.9.0
-
None
-
eagle cluster with Lustre tag 2.7.65
-
3
-
9223372036854775807
Description
Running the sanity-krb5 test suite on Lustre systems with a separate MGS and MDS on the same node or running with a combined MGS/MDS, test 151 fails with
'mount with default flavor should have failed'
This does not happen when running with a MGS and MDS on separate nodes. From the test_log, we can see that the complaint is that the MDS is already mounted:
Starting mgs: /dev/vda3 /lustre/scratch/mdt0 pdsh@eagle-51vm6: eagle-51vm1: ssh exited with exit code 1 pdsh@eagle-51vm6: eagle-51vm1: ssh exited with exit code 1 Started scratch-MDT0000 Starting mds1: /dev/vda3 /lustre/scratch/mdt0 eagle-51vm1: mount.lustre: according to /etc/mtab /dev/vda3 is already mounted on /lustre/scratch/mdt0 pdsh@eagle-51vm6: eagle-51vm1: ssh exited with exit code 17 Start of /dev/vda3 on mds1 failed 17 eagle-51vm1: mgc.*.mgs_server_uuid in FULL state after 0 sec sanity-krb5 test_151: @@@@@@ FAIL: mount with default flavor should have failed
It seems like the test code expects to start the MGs and MDS separately. From test 151:
stopall
# start gss daemon on mgs node
combined_mgs_mds || start_gss_daemons $mgs_HOST "$LSVCGSSD -v"
# start mgs
start mgs $(mgsdevname 1) $MDS_MOUNT_OPTS
# mount mgs with default flavor, in current framework it means mgs+mdt1\
.
# the connection of mgc of mdt1 to mgs is expected fail.
DEVNAME=$(mdsdevname 1)
start mds1 $DEVNAME $MDS_MOUNT_OPTS
Should this test be skipped for a combined MGS/MDS set up?
Logs are at:
https://testing.hpdd.intel.com/test_sets/a5ea3b0c-b987-11e5-80e0-5254006e85c2
https://testing.hpdd.intel.com/test_sets/a92042f8-b987-11e5-b318-5254006e85c2
https://testing.hpdd.intel.com/test_sets/af1bee50-b987-11e5-825c-5254006e85c2
https://testing.hpdd.intel.com/test_sets/b06dfcbc-b987-11e5-b318-5254006e85c2