The mailing list for listmasters using Sympa

[Matt Taggart <address@concealed>]

On 6/14/19 10:10 AM, kill-9 wrote:
> Does anyone have any anecdotal evidence of whether current Sympa will
> work properly as multiple servers sharing the spools, list configs, etc.
> over shared storage, such as NFS? The DB will be a multi-master mariadb
> setup as well.

We haven't done this, but we have done:

* DB on two other hosts in mysql replication setup, read/write io to the
master, some additional read-only going to the slave
* have postfix on the sympa machine relay mail to a bunch of other
outbound MXs using SSDs before going out to the world
* split the i/o load of the sympa machine up by putting them on separate
raid1 SSD arrays: postfix spools, sympa spools, expl dir, archives. If
you don't do something like this, each list mail that comes in results
in a bunch of i/o contention as it attempts to do all these thing at
once (this was _super_ bad before SSDs too)

Currently we have 18k lists, 4.7m subscriptions total and the main sympa
system is a 4 core Haswell era system and it's barely using the CPUs.
Delivering list mail is mostly an i/o thing I think.

Things we've wanted to try
* have the additional outbound MXs run bulk.pl themselves and grab mail
to send from the db directly
* have redundancy for the inbound mail and web interface so we could
take things down for maintenance without downtime

-- 
Matt Taggart
address@concealed