Date: Sat, 01 Mar 2014 14:07:53 +0100 From: magnum <john.magnum@...hmail.com> To: john-dev@...ts.openwall.com Subject: Re: Status request signallling to children or MPI nodes. On 2014-03-01 04:37, Solar Designer wrote: > On Wed, Feb 26, 2014 at 08:01:18PM +0100, magnum wrote: >> Is there a reason you picked SIGUSR2 for fork mode's signalling to >> children instead of SIGUSR1, that the MPI code has used for ages for the >> same thing? I have a patch that changes your code to use SIGUSR1 and >> drops the specific MPI code but I thought you might have had a good >> reason for not doing that? > > I recall briefly thinking about this when writing that code, but I no > longer recall why I chose that specific signal. Perhaps you wasn't sure whether MPI *and* fork could be in use at times. But that is currently not supported and will bail out anyway. > Maybe you can try moving the MPI code's signal to SIGUSR2, and free up > SIGUSR1 for your new use? That way, no change to core would be needed. That would technically be no problem at all but MPI has used USR1 forever so it will break things (including becuase of die-hard habits). Fork's signal to children is not used externally nor manually so it would be better to change that. I accidentally committed the experimental move to USR1 for fork prematurely a day ago or so but opted to let it stay for now. It works just fine. If it turns out to be bad for some reason, we can change it later. On a related note I have also committed the "reload on USR2" patch including --reload-every=N (which signals USR2 to children). The good news is it seem to work like a champ (although ineffective if used often) but the bad news is it (ie. execv) can never work under MPI :-( So now I'm struggling with a *real* reload, not using execv. Not trivial, more on that (including more questions :) later. thanks, magnum
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.