Sajith T S <saj...@hcoop.net> writes: > It can't be just me that's causing an Internal Server Error... > > Thanks for looking into this!
The getcwd() bug in afs struck again... I burned out a bit a little over a week ago; I've just been rebooted navajos daily to keep it going. But I think I've finally reintegrated all of the threads at hcoop (having to use dovecot instead of courier on the new mail server made my beautiful plans unravel!) and will be resuming work on getting navajos stable and the new mail server in production. A few notes on how I'm going about restabilizing navajos's afs client, since -sysadmin is pretty much an echo chamber nowadays: The first step will be upgrading to an openafs 1.6.8 prerelease for the client on navajos. The developers said it ought to fix the suexec getcwd() bug. Maybe the rxkad errors and soft lockups in the VFS will go away then... but probably not. Next comes doing traffic dumps, auditing which processes are contacting the afs database servers, and then figuring out *what* to debug to trace the afs client problems. I tried upgrading to linux 3.2 from 2.6.32 on navajos hoping perhaps it was a problem with the old kernel that was causing at least some of the problems. I think it *did* relieve the rxkad errors, but replaced them with even worse lockups... apache was repeatedly hanging in the vfs instead of afs client, and sooner. I am beginning to suspect it may just be a better idea to snapshot navajos, upgrade it in-place to wheezy, and hope it goes OK. Our config packages are up to date with wheezy (mccarthy has apache installed too, and it works) so nothing should go awry. Wheezy was also a fairly minor upgrade as far as web software goes, and navajos has been running things like the backported python since day one... At least then if problems persist we'd be debugging them against a reasonably modern kernel (using the wheezy-backports kernel). We'd also get better TLS support out of it... I've also done some preliminary work for switching from courier to dovecot. The courier mail indexes -> dovecot indexes script claims that no mailboxes at hcoop would have errors upon conversion; I've made the vmail database code amenable to publishing more than one format, and have a pretty solid idea of how we need to configure dovecot. The hard parts remain ;) (mostly making sure dovecot+exim won't start eating your mail during delivery). -- Jessie: i stuck the phone antenna up the dogs nose and he ignored me
pgpxICEipv8dI.pgp
Description: PGP signature
_______________________________________________ HCoop-Help mailing list HCoop-Help@lists.hcoop.net https://lists.hcoop.net/listinfo/hcoop-help