Bug 1143996

This is a known weakness with how multi-host testing in Beaker is structured. The tasks in each recipe must line up.

It is mentioned in this doc:

https://beaker-project.org/docs/user-guide/multihost.html
"Firstly, any multihost testing must ensure that the task execution order aligns correctly on all machines..."

But I will be the first to admit that that doc could be improved. It's a really long narrative tutorial-style doc which would benefit from being split into something more task-focused with discrete sections.

You can use /distribution/dummy as a placeholder to fill out gaps in your recipes so that all tasks correctly line up.

I'm converting this to a docs bug, about making this limitation (and the recommended solution) much clearer and easier to find in the docs.

Thanks Dan for the quick response.

It really looks like that this multihost test weakness is by design, but is there any chance that at least the SERVERS and CLIENTS variables are set properly when the tests do not line up?

IMHO this is still a bug and we have a real scenario where this flaw matters. We are using in some tests an ad-hoc synchronisation mechanism (through nfs) other than rhts-sync-* which requires only the SERVERS and CLIENTS variables (so TESTORDER mismatch doas not matter).
So while for legacy synchronisation mechanism we still need to use place-holder tasks, as you pointed out, for this nfs based synchronisation we wouldn't.

Tasks are always paired by their order and the roles in tasks are not visible outside of corresponding "task-set".

But you can set a role in recipe which will be available in all tasks.

The placeholder also won't solve an issue when the respective task on the second system takes significantly more time. I think that systems should be synchronized by harness before each multihost task to ensure that tests starts at the same moment and that one system is not waiting in the TC while the other system is finishing the previous task. Such tests might be killed by local watchdog.

(In reply to Marian Csontos from comment #3)
> Tasks are always paired by their order and the roles in tasks are not
> visible outside of corresponding "task-set".
> 
> But you can set a role in recipe which will be available in all tasks.

I'm not sure if I understand correctly what you mean, but the problem what should be fixed is the following.
Let's consider the following recipe set:

<recipe kernel_options="" kernel_options_post="" ks_meta="method=nfs" role="None" whiteboard="">
	<task name="/distribution/install" role="STANDALONE"/>
	<task name="/multihost test" role="SERVERS"/>
	<task name="/singlehost test" role="None"/>
	<task name="/multihost test" role="SERVERS"/>
</recipe>

<recipe kernel_options="" kernel_options_post="" ks_meta="method=nfs" role="None" whiteboard="">
	<task name="/distribution/install" role="STANDALONE"/>
	<task name="/multihost test" role="CLIENTS"/>
	<task name="/multihost test" role="CLIENTS"/>
</recipe>


In the result the key environment variables will be set like this:

<recipe kernel_options="" kernel_options_post="" ks_meta="method=nfs" role="None" whiteboard="">
	<task name="/distribution/install" role="STANDALONE"/>

	<task name="/multihost test" role="SERVERS"/>
        SERVERS=<servers_hostname>
        CLIENTS=<clients_hostname>

	<task name="/singlehost test" role="None"/>
        CLIENTS=<clients_hostname>

	<task name="/multihost test" role="SERVERS"/>
        SERVERS=<servers_hostname>
</recipe>

<recipe kernel_options="" kernel_options_post="" ks_meta="method=nfs" role="None" whiteboard="">
	<task name="/distribution/install" role="STANDALONE"/>

	<task name="/multihost test" role="CLIENTS"/>
        SERVERS=<servers_hostname>
        CLIENTS=<clients_hostname>

	<task name="/multihost test" role="CLIENTS"/>
        CLIENTS=<clients_hostname>

</recipe>

Beah is no longer supported by Beaker development team.
Instead of that, we are working on Restraint test harness. You can find all the features of Restraint here.

https://restraint.readthedocs.io/en/latest/

If you think your RFE should be still implemented as part of Restraint feel free to create a new BZ ticket.

https://bugzilla.redhat.com/enter_bug.cgi?product=Restraint

In case you have any question feel free to reach out to me
Thank you,
Martin Styk <martin.styk>