self.build_locally = False doesn't get a --time specification with Slurm based launching.

[akesandgren]

using this config:

site_configuration = {
    'systems': [
        {   
            'name': 'kebnekaise',
            'descr': 'Kebnekaise cluster',
            'hostnames': ['b-'],
            'modules_system': 'lmod',
            'partitions': [
                {   
                    'name': 'bdw',
                    'descr': 'Broadwell compute nodes',
                    'scheduler': 'slurm',
                    'launcher': 'srun',
                    'access': ['-C broadwell', '-A sysop'],
                    'environs': ['gnu', 'intel'],
                    'resources': [
                        {   
                            'name': 'tasks',
                            'options': ['-n {cores}']
                        },
                    ],
                },
...
}

And a test that does

self.build_locally = False

The build phase created slurm batch file lacks a --time specification, causing it to fail on our system since we require it to be specified.

[teojgo]

@vkarak do you think we should add an extra field for the timelimit of the build or use the same as the one of the running job?

[vkarak]

I would use the same or make it indefinite or expose the build_job to the users. Because the problem of taking the job limit of the run job is that it might be completely irrelevant, but on the other hand I don't want to replicate everything for the build job.

[akesandgren]

indefinite would cause problems for site with a upper limit on time/job.

[vkarak]

Well, I mean "indefinite." I am inclined in removing completely the default limit imposed by the framework on the test. The fact that users must pass --time when running on a specific partition, I think that it should be reflected in the partition configuration. Either in the access option or a new option, so as to cover also the local scheduler. The default time limit in Slurm is that of the partition, as defined by the sysadmins. You would still be able to set the time_limit inside your tests also for the generated build jobs.

[akesandgren]

Well we don't have a default, --time must be explicitly set. So please make sure it's easy yo handle someway.

[vkarak]

Would you be happy to define it in the configuration?

[akesandgren]

well a default there would do, but you need to be able to specify it per test. some build jobs may take a very long time...

[vkarak]

Yes, you will be able to do that, too.