Forget it. I was the only one foolished :) strace showed it was effectively an attempt to open more than 1018 change log... Then in absence of garbage collect every socket attempt just fails... (Too many open files) If we find the guilty change log opener, this will fix the issue at least on my linux box. Otherwise inserting a retryWithGC: here and there is a possible temporary workaround... But those files will have to be closed... 2009/5/1 Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com>:
Hmm... I suspect my linux /proc has been foolished for some reason and it was in fact some sockets... I tried again, during loading, the socket opened and close and were recognized as such in /proc, but after a certain point, listing /proc/$pid/fd was reporting change log open and close... though the loading did continue normally. I will try the whole procedure again, but all this is really slow (I had some other traffic going on my limited bandwith). Nicolas
2009/5/1 John M McIntosh <johnmci@smalltalkconsulting.com>:
Well yes, and the magic check would be
FileStream allSubInstances size inspect
same place as the socket.
Nothing of course finalization is involved in getting rid of zombie file handles too.
And Dave is right you should only have one file handle open (read/ write) for *.changes
On 30-Apr-09, at 4:30 PM, David T. Lewis wrote:
Well that's a bug for sure. Somebody is opening new file streams on the changes file, probably by accident. And not closing the one they opened last time.
The bug in changes file handling could easily appear to be a socket problem, because the per-process limit on open descriptors would be shared between files and sockets (and pipes and ...). So find out why the changes file is being accidentally reopened, and the "socket problem" will probably go away.
Dave
-- = = = ======================================================================== John M. McIntosh <johnmci@smalltalkconsulting.com> Twitter: squeaker68882 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com = = = ========================================================================
_______________________________________________ Pharo-project mailing list Pharo-project@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project