Dovecot HA/Resilience

[deano-dovecot at areyes.com]

My own personal setup is a 3-node system using three cheap VPS'.  I also 
helped to set the same thing up for a previous company using proper 
systems, this was handling customer email.

Everything possible is kept in mariadb with galera for master-master 
replication.  Two main mail nodes with 
dovecot/nginx/roundcube/spamassassin/etc and a third as a mariadb quorum 
node.  Dovecot uses replication to keep the encrypted mailstores in 
sync.

This way there is no need for HA storage - you're relying on 
replication.  Oh, and the replication all happens over a tinc vpn mesh 
network, but would work equally well over zerotier or whatever.

I have an ansible playbook to set the whole thing up automagically.  I'm 
working on cleaning it up and documenting it so others can use it as 
well.  So long as you have ssh key'd access to 3 nodes, it will build 
the entire setup.

I'll put it up on github in a few weeks.  NOTE: this is built for MY 
needs.  It might not meet your needs.  But when it's ready(ish) you're 
welcome to try it out.  For example, it's not true HA - you have to hit 
one node or the other.  If you control your own DNS you could set up 
round-robin for a mail.yourdomain.com rather than using 
mail1.yourdomain.com and mail2.yourdomain.com.  For me, I don't bother.

Dean.

On 2020-01-11 3:50 am, Jean-Daniel wrote:
> If you just want active/standby, you can simply use corosync/pacemaker
> as other already suggest and don’t use Director.
> I have a dovecot HA server that uses floating IP and pacemaker to
> managed it, and it works quite well.
> 
> The only real hard part is having a HA storage.
> You can simply use a NFS storage shared by both servers (as long as
> only one has the floating IP, you won’t have issue with the same
> client accessing it from both servers), but the storage will then be a
> single point of failure.
> You may have both server have their own storage and sync it using
> dovecot replicator (I have never tried, so I can’t say for sure), or
> have an other layer taking care of the storage sync (like DRDB).
> 
> While drdb is fine to sync dovecot storage, it may not be enough if
> you really want HA and have other services (postfix, rspamd, …)
> running on that server, as you may need to also have the postfix
> queues (or other data) sync on both servers.
> 
> 
> 
>> Le 10 janv. 2020 à 21:12, Adrian Minta <adrian.minta at gmail.com> a 
>> écrit :
>> 
>> Yes, but it works for small systems if you set IP source address 
>> persistence on LB or even better, if you set priority to be 
>> Active/Standby. I couldn't find a good example with dovecot director 
>> and backend on the same server, so adding another two machines seems 
>> overkill for small setups.
>> 
>> If someone has a working example for this please make it public !
>> 
>> Quote from https://wiki2.dovecot.org/Director
>> 
>> "Director and Backend in same server (broken)
>> NOTE: This feature never actually worked. It would require further 
>> development to fix (director would need to add "proxy" field to extra 
>> fields and notify auth that the auth_request can be freed)."
>> 
>> Also:
>> 
>> https://dovecot.org/pipermail/dovecot/2012-May/135600.htm
>> 
>> https://www.dovecot.org/list/dovecot/2012-June/083983.html
>> 
>> 
>> On 1/10/20 8:09 PM, Aki Tuomi wrote:
>>> Also you should probably use dovecot director to ensure same user 
>>> sessions end up on same server, as it's not supported to access same 
>>> user on different backends in this scenario.
>>> 
>>> Aki
>>> 
>>>> On 10/01/2020 19:49 Adrian Minta <adrian.minta at gmail.com> wrote:
>>>> 
>>>> 
>>>>  Hello,
>>>>  you need to "clone" the first server, change the ip address, mount 
>>>> the same maildir storage and use some mechanism to share the 
>>>> accounts database.
>>>> 
>>>>  Then you need to put a TCP load-balancer in front of the servers an 
>>>> you are good to go. This is the easiest solution if you already have 
>>>> in the network an appliance that can do LB. For instance if you 
>>>> already have a firewall with that function.
>>>> 
>>>> 
>>>> 
>>>>  Another solution is to make a cluster with corosync/pacemaker out 
>>>> of the two servers:
>>>> 
>>>>  
>>>> https://www.digitalocean.com/community/tutorials/how-to-create-a-high-availability-setup-with-corosync-pacemaker-and-floating-ips-on-ubuntu-14-04
>>>>  
>>>> https://linuxacademy.com/blog/linux-academy/configure-a-failover-cluster-with-pacemaker/
>>>> 
>>>> 
>>>> 
>>>> 
>>>> 
>>>>  On 1/10/20 7:16 PM, Kishore Potnuru wrote:
>>>> 
>>>> 
>>>>> Thank you all for the replies....
>>>>> 
>>>>> 
>>>>>  I have the test environment with the same configuration. But I 
>>>>> have been asked to go with same environment for HA/Resilience in 
>>>>> Live.
>>>>>    Yes, I have only one Live server. It is configured in "Maildir" 
>>>>> format. The data stores on a Network / Shared Storage (But 
>>>>> definitely not local disk, its a mount point).
>>>>>    I have been asked to create a HA/Resilience for this 
>>>>> environment. They gave me another server with same ram/cpu/os and I 
>>>>> need to configure the dovecot on it.
>>>>>    Please provide your suggestions/steps as I am new to this kind 
>>>>> of environment.
>>>>>    Is it possible, when any email comes to any one or both of the 
>>>>> two servers, how it will be read by the user from Outlook? How to 
>>>>> create the environment?
>>>>> 
>>>>> 
>>>>>  Thanks,
>>>>>  Kishore Potnuru
>>>>>    On Fri, Jan 10, 2020 at 7:55 AM Sami Ketola 
>>>>> <sami.ketola at dovecot.fi> wrote:
>>>>> 
>>>>> 
>>>>>>    > On 10 Jan 2020, at 9.20, Emmanuel Dreyfus <manu at netbsd.org> 
>>>>>> wrote:
>>>>>>  >
>>>>>>  > On Fri, Jan 10, 2020 at 09:07:24AM +0200, Aki Tuomi wrote:
>>>>>>  >> Replication is not supported with mbox. Most features are not.
>>>>>>  >
>>>>>>  > It would be nice if the document about replication could tell
>>>>>>  > what setup works.
>>>>>>      First step in setting up HA system would be to migrate away 
>>>>>> from mbox.
>>>>>>    Sami