[Mailman-Users] mailman 2.1.1 not taking actions (subscribe)
Todd A. Green
tag at cs.utah.edu
Wed Sep 17 00:17:38 CEST 2003
> Mailman sends out a welcome message after doing a read of the database
> and finding that the user is not a current member. That means it could
> have failed in it's attempt to write to the database - though that would
> definitely send some errors to see in the logs.
It would seem that verifying the write to the database and then the
confirmation would be more appropriate no?
> I wonder if there is a problem with commiting the data via NFS? Are you
> running Sync or Async? Sync doesn't send back a thumbs up to the NFS
> process until the write is committed on the local drive. Async smiles,
> and winks and says: Of course it's been committed to the local drive;
> pay no attention to the bits behind the curtain.
They are synchronous writes. We have ~266 lists and do gobs of NFS (peaks
of 20+K ops/sec, averaging around 6-8K ops), so I don't think it is an NFS
issue. Everything we do is NFS based. (user's Windows profiles and unix
home dirs, Maildir, etc. are all on our Netapps).
This is what is driving me nuts. Everything should work but isn't. And it
is just a few lists from time to time that go into this phantom mode. The
fact that running "add_member" fixes the list tells me it has to be
something internal to Mailman. The question is what does add_member do that
is different than the email or web interface for subscribing. (We've also
have problems with unsub'ing when the list goes in phantom mode too. I'd
guess it is any action dealing with updating the database.)
I need to check what kind of file locking mailman does, as that would be my
next guess. It is failing to get a lock, but not checking the error status
perhaps. Our Netapps run in mixed mode, which means they support both NT
style ACLs and unix user modes. However we never access the mailman files
from Window's machines, so that shouldn't come into play (in theory.)
More information about the Mailman-Users