[Mailman-Users] Mailman on multiple web and mail servers

Mark Sapiro mark at msapiro.net
Fri Jan 22 03:28:39 CET 2010


Guy wrote:
>
>I've been going through the archives and I've picked up bits and pieces of
>the information I need, but in most cases NFS is discussed.
>I've got 5 Ubuntu servers, 2 postfix mail gateways (MXs for all our domains)
>and 3 apache web servers. All of those have access to an ISCSI SAN with
>OCFS2 partitions.
>
>Ideally I'd like to have the mail section running on both gateways
>(potentially more than just the 2 in the near future) and the web interface
>available from any of the web servers. This is more for availability than
>performance.
>Should separate installations of the web section on each web server be fine
>or is there shared information needed there?


There is shared information in all of the archives, data, lists, locks,
logs and qfiles directories. These need to be on shared storage so
they are accessible to all web and mail servers.

By the 'mail section', I'm guessing you mean the qrunners. The qrunners
need access to all the shared data, and if there are multiple qrunners
processing the same queue, they need to be sliced so they don't
interfere with each other.

I suppose you could have a separate set of qrunners and separate mail
delivery on each mail server, each with its own set of local qfiles,
but they would still need access to the shared archives, data, lists,
locks and logs directories (well, they could have their own logs too).


>And since I'm using a clustered fs, would I still need to slice up the
>qfiles directory structure for the mail section?


Slicing is not a directory structure thing. Each queue entry is a file
in the corresponding queue directory. E.g., qfiles/in/file. Slicing
just divides the file name space into sections each of which is
processed by its own runner.


>Or would having the entire mailman installation on shared storage and
>individual apache/postfix configs per machine be the way to go?


It's one way to go. Keep in mind that much of the code is common to all
processes, so having local code will result in much duplication. OTOH,
it will reduce contention on the shared store. This is not a tradeoff
that I can give a pat answer for since I don't know what your load is
or what the capability of your shared fs is, and I have virtually no
experience with this anyway.


>Any other suggestions or advice about this would be greatly appreciated as
>well.


I hope this helps. If it doesn't answer your questions or raises other,
ask again.

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



More information about the Mailman-Users mailing list