per IRC talk: we need more details how to stop and merge the queues etc

I've looked at this. If we add lists.wikimedia.org to the hold domains and increase the retry time (to at least 4 hours, the window length), then all emails should be written to /var/spool/exim4/input.

From there, we can rsync sodium onto fermium's directory (exim4 will hold on fermium too so this can be a puppet change and the directory will be used by exim already) and allow exim to process emails once we're confident nothing else is going to sodium's exim.

This creates two levels of stoppage. Mailman will be stopped so it will not process any emails on both servers which should be done after exim begins to hold emails.

I propose:

• hold lists.wikimedia.org for exim (via puppet/git), deploy it
• wait 10 minutes, check mailman is not doing anything mail wise - then stop it.
• do the migration
• start mailman on fermium only.
• don't hold anymore emails on fermium only. Sodium should have puppet disabled and should remain to queue emails otherwise any unlucky people will never have their emails acknowledged by mailman and will just bounce after 24-48 hours.

And assigning back to Daniel.