On January 21, 2003 05:38 pm, Claudio Jeker wrote:
> Thanks for the patches, I will add variatioins of both. See
> attached patch.
>
I've patched qmail-reply.c with your improved patches and will
continue testing against that. I had to add a mini-change into
recent_update() - "for(; size > MAX_SIZE; )" changed to "for(; n >
MAX_SIZE; )".
> The main problem with the NOTYET stuff is that multiple concurrent
> accesses to the .qmail-reply.db file will dump some users (by
> overwriting a just written file with a new file). I'm not sure if
> we should use the maildir++ approach with timestamps or if it is
> better to use file locking (which sucks over NFS). Perhaps we can
> also ignore it and some senders get more than one reply in the
> REPLY_TIMEOUT time.
Would it be dangerous to always use the filename .qmail-reply.db.tmp
as the temp file, then check for the existence of that file before
updating .qmail-reply.db? ie. if .qmail-reply.db.tmp exists, then
sleep for a random number of seconds and try again up to the x minute
timeout period. That way if it exists it means someone else is
trying to update the db and we wait until they finish before doing
the current update - if the timeout expires, then just ignore the
update and move on. Not sure what nasty side-effects that could have
(could the potential race be eliminated? would it really matter?).
Considering how critical this is (not), ignoring it might not be so
bad - if someone wants to enable this feature, then having occasional
duplicate replies is better than always having duplicate replies.
-neil
