Skip ha/drbd_passive test when not in TWO_NODES scenario #15324
Conversation
|
One possible solution would be to create a different schedule for the 3 cluster node scenario, but this implies a PR to this repo, as well as a change to the job settings. - don't we need a PR to add the "TWO_NODES=no" anyway? Anyway, if it passes your vr, lgtm. |
It is already there:
If memory serves, TWO_NODE=no is used by other test modules and added quite a while ago. I am just simply re-using the same setting for this particular test module. In fact, I think L47 on Could be re-written safely as only: But I am leaving the node checks just in case there is already a working test using |
tests/ha/filesystem.pm
Outdated
| $resource = 'drbd_active'; | ||
| } | ||
| elsif ($tag eq 'skip_fs_test') { |
There was a problem hiding this comment.
Isn't this duplicate of line 21 ?
And if it should be at begging because tag drbd_passive would be evaluated before this skip thus not skip.
There was a problem hiding this comment.
Isn't this duplicate of line 21 ?
Not exactly the same. L21 would skip the test when not on drbd Maintenance Updates, but continue it if testing a drbd MU, such as in https://openqa.suse.de/tests/9247409#step/drbd_passive/5
L44 on the other hand is skipping the test on scenarios where the test would not work (such as in 3 node) per L47-L51 in ha/drbd_passive.
And if it should be at begging because tag drbd_passive would be evaluated before this skip thus not skip.
Not a problem. $tag would be one or the other, and never both: https://github.com/os-autoinst/os-autoinst-distri-opensuse/blob/master/lib/hacluster.pm#L458-L480
(This probably could've been implemented with get_var() and set_var() instead, but no idea why Loic went with a local file in the cluster nodes for these tags. Could be a candidate for a re-write in the future)
There was a problem hiding this comment.
But instead of two places where the test will return there could be just one at the beginning of the if statement. Otherwise fine.
There was a problem hiding this comment.
Ah, got it. Let me move both exit conditions to the same line and test.
There was a problem hiding this comment.
New verification runs:
- Alpha Cluster on Milestone Build Validation: node 1, node 2 & support server
Cannot run VR with the drbd MU as the incident repository is gone: http://mango.qa.suse.de/tests/4893#step/iscsi_client/12
Should we merge with the new change or do I rollback 7997ebe?
There was a problem hiding this comment.
Merge this PR, you are writing skip_fs_test tag in first commit
alvarocarvajald
changed the title
Also skip related ha/filesystem test as if ha/drbd_passive test module is skipped, there would be no block device on which to create the FS.
alvarocarvajald
force-pushed
the
drbd-on-3nodes
branch
from
7997ebe
to
81dbecc
Compare
Currently the 3 cluster node scenario tested for Maintenance jobs is always scheduling the
ha/drbd_passivetest module, even when this particular test module is not designed to work in cluster scenarios with more than 2 nodes.As a result, whenever a MU job is triggered with a package that requires a drbd test, the 3 node scenario job will fail as it is skipped in the third node before any of the
barrier_wait()calls. This means the other nodes, remain blocked in abarrier_wait()call until they reach MAX_JOB_TIME and fail the whole scenario.One possible solution would be to create a different schedule for the 3 cluster node scenario, but this implies a PR to this repo, as well as a change to the job settings.
This PR instead modifies
ha/drbd_passiveso it is skipped when running in scenarios with the setting TWO_NODES=no. It will also skip the test moduleha/filesystemthat is usually scheduled right afterha/drbd_passive, as there would be no block device on which to test the filesystem creation.P.S.: I also include here a small optimization in
ha/vgtest module to avoid having multiple calls tohacluster::read_tagwhich sends commands to SUT.