Viktor: > If the files are guaranteed to only have a single recipient at the > point > in your processing stream at which they are "held", you don't have to > do > anything nearly so complex. Just retain the file's original name and > inode, by renaming it into a suitable directory tree in the same file- > system. > > Releasing can be accomplished by just moving it back into the maildrop > queue > with its original name.
Perfect... I didn't know it was safe to physically move the files out of a queue directory. We were copying them out and then deleting them from the hold queue with postsuper. Just moving the queue files would significantly reduce disk i/o too. > A better design (working on this for our next gen quarantine) is I > think > to deliver the quarantined messages to a custom SMTP or LMTP agent that > saves the message envelope and data in an ASCII form. Then messages > released by users from the quarantine are re-injected also via SMTP. I believe that is what Wietse is suggesting in his subsequent email. It's going to take a major overhall to get our quarantine system to be able to accept submissions by SMTP/LMTP, but based on all that I'm hearing from Wietse, I need to eventually make that happen. In the mean time, it seems like using doing "postsuper -r" to re-activate old queue files would be a good alternative. Hopefully that resolves the expiration cycle issue that is caused when you inject a queue file directly into the maildrop queue? Curtis
