QA » openQA Project » openQA Tests

Communication with Egbert for the snapshot test on HPC, it is not so important. We can skip grub_test and snapper_rollback, etc which is blocked by btrfs FS(currently it is xfs FS on HPC).

Email content:

Hi Egbert,

Thanks for so quickly reply.

On 2/20/20 7:32 PM, Egbert Eich wrote:

Hi Lemon,

Lemon writes:

Hi Egbert,

On 2/20/20 5:28 PM, Egbert Eich wrote:

Lemon writes:

Hi Egbert,

 I found from the description of 'HPC management server' role, xfs

is the default root file system.
http://openqa.nue.suse.com/tests/3896264#step/system_role/4

 I have a question about this, does the xfs is the must file system

for HPC testing? Will there any effect for HPC test if we change to
btrfs? I ask this question since we have snapper rollback test after
migration is done which depends on btrfs. Or maybe snapper rollback test
is not needed any more for HPC migration test if migration is success
(talked in our previous mail), right?
Could you help to confirm for the last question? It's very important for
us.

I would say the migration tests as we perform them for HPC management
servers are not relevant for HPC at all: nobody would start a 'yast migration'
go for coffee, come back and expect everything to work.
Management servers are usually carefully planned, it is recommended to
make backups of important data (like the accounting database for the
workload manager). Here snapshotting may help, however, databases are
exempt from snapshotting.
To me, the entire 'yast or zypper online migration on a life and running
system' is just a checkbox item by PM.
In my opinion it has no real life relevance for HPC systems: as soon as
a life workload manager is running such a scenario is unrealistic. Even
if a rollback is technically possible the system could well be in a state
after a migration from which a recovery is more difficult than just
running 'snapper rollback'.
Our tests cover 'idle' systems only.

Bottom line, snapshotting may be useful during a migration of an HPC
system, but we have not yet made a recommendation for such a scenario.
Sure, understood it, thanks a lot for the info.

For the management server, we don't need xfs necessarily, it was
done due to preferences by the HPC community. There are arguments
which speak for btrfs as well.
I'm not sure if we want to change this that late in the game -
ie at the eve of the RC phase.

I think you misunderstanding my meaning, I just mean if needed we QA
modify it on test side while not change the default to btrfs.

IHMO, without changing the default on the product, there are two options:

1. QA adds a workflow to change the FS from xfs to btrfs
2. QA changes the workflow to not perform any of the rollback tests.

So, are you asking which one of the two options we should do?

We QA migration team will take the option 2. Thanks.

Best regards,

lemon