[LU-9980] save_lustre_params() defect

Related — Wed, 11 Oct 2017 20:36:02 +0000

There is a defect in save_lustre_params() in test-framework.sh.
replay-vbr calls save_lustre_params() with wildcard on parameter :

save_lustre_params $(get_facets MDS) "mdt.*.commit_on_sharing" > $cos_param_file

On the setup with facet_HOST != facetfailover_HOST we could have the following stored parameters :

[root@lm0117 ~]# cat /tmp/rvbr-cos-params
mds1 mdt.lustre-MDT0000.commit_on_sharing=0
mds1 mdt.lustre-MDT0001.commit_on_sharing=0  <<< 1st parameter
mds2 mdt.lustre-MDT0000.commit_on_sharing=0
mds2 mdt.lustre-MDT0001.commit_on_sharing=0  <<< 2nd parameter
mds3 mdt.lustre-MDT0002.commit_on_sharing=0
mds4 mdt.lustre-MDT0003.commit_on_sharing=0
[root@lm0117 ~]#

after facet_failover mds1 mds1 is mounted on other node, but mds2 are still on the the same node.

restore_lustre_params () is trying to restore "mdt.lustre-MDT0001.commit_on_sharing=0" on mds1, but it is missing because of mds2 was not failed and it is still on the same place.
I.e. save_lustre_params() duplicates the info, 1st parameter is not related to mds1, it is related to mds2.

Whamcloud Community JIRA

[LU-9980] save_lustre_params() defect