Open ompiteam opened 10 years ago
Imported from trac issue 283. Created by emallove on 2007-08-27T14:48:22, last modified: 2009-03-11T11:03:58
Trac comment by jsquyres on 2007-08-27 15:06:21:
You can implement it similar to how I did #227 -- just look for a file during the run. It's ain't sexy, but it works.
Having a daemon is possible, but would be sorta complex. All MTT instances would need to register with the daemon (and stay connected and have asynchronous handlers for incoming messages from daemons, etc.). You'd also need a way to talk to the daemon to say "do X to all running MTT's" and "do X to MTT Y". Etc. The corner cases become numerous -- I don't think we want a daemon here. :-)
Trac comment by emallove on 2009-03-11 11:03:58:
One way to do this would be to just track the modification time of the INI file. At startup, MTT will record the INI file's modification time, and if this property changes at any point, MTT will reload the newly modified INI file. E.g., if the user is partially through a ten hour test run, and they notice that they need to increase a timeout value, they can simply tweak the timeout parameter in the INI file. Since it would not always be desirable for MTT to reload the INI file partway through a run, a prompt may be appropriate, e.g., ''"The MTT INI configuration changed. Do you want MTT to effect the changes now? (yes/no)"''
Related to #227 (''Stop all testing and submit what you have now''), but instead of stopping MTT, we will be tweaking it's run-time options midway through a run. The use cases I have in mind are:
We would need to report to the user on whether the option they are changing is in fact changeable in the client's current state. E.g., skipping an MPI get section makes no sense if MTT has completed all its MPI get sections. There are MTT options we would want to exclude from this feature, e.g.,
--scratch
--file
Do we need an MTT daemon for this ticket? A la Postgres's
pg_ctrl
?