<[EMAIL PROTECTED]> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Mailer: Mutt 1.0pre3us In-Reply-To: <[EMAIL PROTECTED]>
Clifton Royston on Tue 9/04 14:01 -1000: > > [uw-imap with LOCKPGM set to mlock from imap-utils allows qpopper > > and uw-imap to co-exist] > > This part actually works, yes. > > > [but not in server mode] > > This is the part everyone's concerned about. > > > I still think the lock should surround the session and let MDA > > handle temporary errors, but maybe that is bad policy, I don't know. > > One problem with that solution is that lock is by convention > considered dead and removable by the MDA after 6 minutes, and a POP > session can go longer. I don't know what you mean by `convention.' I recall a discussion where somebody said their copy of sendmail did this after a configurable default of 6 minutes. We are using procmail for delivery, which waits for a configurable default of 17 minutes, then clears the lock like sendmail. And looks like qmail with its own DA timeouts in 30 seconds but the message gets re-queued instead of forcing the lock off (which, IMO is the right thing to do, something has not cleaned up after itself and the bug should be found, not ignored by destroying the lock) - If your MDA timeout is configured too short and POP sessions overrun it commonly, block longer on delivery (or better yet forever), dont use server mode, or disable full pop session lock. I realize that a user could be getting a gig through email over a modem and holding up all his deliveries, but this user has other problems and we should not worry about him, besides we have quotas to enforce upper bounds and prevent timeouts from getting too crazy, and idle timeouts on the server to prevent this from being a problem for hungup clients. - If your MTA does not properly handle MDA errors, get a new one. - If something is ignoring both dotlocks and file locks and just overwriting files, you have a misbehaving program somewhere, fix it, or get a new one. - In all cases timeouts are configurable. So why wouldn't it follow that a full session lock be optional to people that know what they are doing. I am interested in how you solved this problem with the blackbox devices you were looking into for in an earlier thread. I also wonder how ipop3d handles it, I suspect it just isn't optimized for this case like qpopper in server mode and thus doesn't run into this problem.
