On 01/30/2015 03:09 PM, Andrew Hodgson wrote:

My question revolves around the list archives. If I start Mailman up before they are copied across, it is likely there will be waiting messages and these will get archived to a new mbox file. I want to reduce downtime in moving the lists, but the mbox files are around 25GB, and there are a couple of lists which take the majority of that. I was going to index the messages once the mbox files are copied over, but what is the smartest way to get these moved without downing Mailman for too long?

It seems like you are planning to just move the archives/private/LIST.mbox/LIST.mbox files and rebuild the archives with bin/arch --wipe after the .mbox files are moved.

This in fact is the recommended procedure. One consideration is whether you want to preserve the message numbers in the new archive. If you don't do this, and someone has a saved URL to an archived message, that URL will return a different message from the new archive. So let's assume you do want to do this.

The second consideration is whether your .mbox files are "good". They probably are, but if they go back several years and you haven't previously checked, they may contain unescaped "From " lines in message bodies. bin/arch is very lax about validating "From " separators. Any line that begins with those 5 characters is considered a message separator and can result in trailing parts of messages being in the archive without headers and separately from the leading parts.

This is normally not an issue with .mbox files written only by 2.1.x versions of Mailman. Mailman's bin/cleanarch script can help detect and fix such problems.

OK. Now we'll assume you have a bunch of good LIST.mbox files on the old server which you just stopped. I think there are two ways to proceed. You'll have to decide which works better for you. While typing, I decided Way 2. is better.

Way 1. Bring up the new server and start Mailman. Let the lists operate and messages archive. These messages will ultimately be renumbered, but this is a relatively short window so it may not be an issue.

Bring the LIST.mbox files as say LIST.mbox/List.mbox.old

After you have the files moved, find the PID of ArchRunner via something like

ps -fwwC python

and kill it with SIGTERM OR SIGINT. Do not use SIGKILL or the master will just restart it.

Then do something to the effect of

cat LIST.mbox/List.mbox >> LIST.mbox/List.mbox.old mv LIST.mbox/List.mbox LIST.mbox/List.mbox.new mv LIST.mbox/List.mbox.old LIST.mbox/List.mbox

And verify the ownership and permissions on LIST.mbox/List.mbox which now should have all the messages with the new ones at the end.

Then copy the ArchRunner command from the previous ps command output and issue that command with sudo -u mailman's_uid and a trailing & to restart ArchRunner. If You don't start in in the background, it will remain attached to your terminal and breaking out will SIGINT it and it will die again. Command is like

sudo -u mailman /usr/bin/python /var/MM/21/bin/qrunner
--runner=ArchRunner:0:1 -s&

Now your mbox is good and new messages are being archived and you can run bin/arch --wipe LIST to build the pipermail archive. This should be OK as bin/arch will lock the archive to prevent conflicts with incoming messages.

Way 2. I think this is better.

Bring up the new server, start Mailman and immediately stop ArchRunner as above before any new messages arrive, perhaps before starting the MTA.

Now you can just bring the LIST.mbox/List.mbox files over and check permissions and rebuild the archives with bin/arch with --wipe or not as the pipermail archive should be empty anyway. Meanwhile lists are running and messages for the archive are queued.

Then when the pipermail archives have been rebuilt, start ArchRunner as above.

Now there are potentially some cleanup things to do. It is possible, depending on how many posts arrived while ArchRunner was not running, the qfiles/archive directory grew large. This can be a performance issue. See the last paragraph in the FAQ at <http://wiki.list.org/x/4030638>. You can shrink the queue directory by waiting until it's empty, stopping Mailman, moving the directory aside and recreating it with the same ownership and permissions and restarting Mailman. and if by chance, one or two entries snuck in to the old, moved aside directory, move them individually back to the new directory.

Finally, see the script at <http://www.msapiro.net/scripts/update_archive_mtime> (mirrored at <http://fog.ccsf.edu/~msapiro/scripts/update_archive_mtime>) if you care about the modification times of the archived message files.

Caveat: I just typed this out. I'm *sure* it will work, but I've been wrong before.

-- Mark Sapiro <mark@msapiro.net> The highway is for gamblers, San Francisco Bay Area, California better use your sense - B. Dylan