From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ben Schmidt Date: Thu, 11 Nov 2010 03:58:59 +0000 Subject: Re: [mlmmj] Potential mail loss in postfix? Message-Id: <4CDB6A03.1000901@yahoo.com.au> List-Id: References: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: mlmmj@mlmmj.org I might have found this bug. If init_sockfd() fails (e.g. because Postfix has shut down so there is no smtpd listening) it calls exit(). Mail would then fail to be archived or requeued. It will be in a queue file only until mlmmj-maintd cleans it up (which it will do as soon as it finds it, as it won't have accompanying .mailfrom etc. files). Do you have logs from when this happened? Do you see "Could not get socket" or "Could not connect to %s, exiting..." (%s probably is 127.0.0.1) in them? Ben. On 29/09/10 9:21 AM, Robin H. Johnson wrote: > Hi > > Noticed something, and I don't have a testcase for it yet unfortunately > or a suitable setup to re-test on. Instead I've got my analysis of the > problem how it's occurred twice now. > > - Using verp and postfix together first of all (string 'postfix' in the > verp file, '100' in maxverprecips). > - Pick a list with a lot of subscribers. > - This leads to a case where the mlmmj-send invocation takes several > minutes to complete for a normal list mail. > - (optional) set postfix to hold incoming mail, and you can release it > just at the right moment to see it be mlmmj-recieve. > - The postfix log will show delivery to mlmmj-recieve. > - mlmmj.operation.log will contain a line from mlmmj-process stating > that the message was allowed (by your access rules). > - Now, while mlmmj-send is running, you're going to execute a normal > shutdown of postfix: 'postfix stop' [1] > - The mail will be lost completely now. There is no record of it in > archive, or any of the queues :-(. > > [1] The description for 'postfix stop': Stop the Postfix mail system in > an orderly fashion. If possible, running processes are allowed to > terminate at their earliest convenience. >