Hi there, I recently started using Mailcow for both my email my clients’ email, and was wondering whether anybody had any advice with regards to making this setup redundant? I’m assuming that Mailcow doesn’t have this functionality built-in, so would my only option be to make a contribution to Mailcow to add this functioanlity, or do some magic with Postfix / Dovecot on a separate node? Has anybody managed to do this? Thanks, A

We will publicate docs for that soon and sell support for it. They will be releasd under the premise that we will not provide support for these setups in the community. 🙂

Perfect! Thanks for the update. Very much looking forward to seeing what you come up with!

We just need to rework it a bit and translate the text. :o Are you german?

I’m also interested in this topic, how the solution would look like 🙂 @diekuh I’m speaking German.

diekuh Sorry for the delayed response! Unfortunately I’m not German, but this is sounding great. If I can help with anything, let me know!

diekuh Any news on this? Would really be nice to hear how far you came in the last months 🙂

diekuh , I’m also interested in the progress of this. Can you share an update?

I will probably release a doc for it in the next few days/weeks. We decided to create new support packages for clustered setups, too. That will be the only way to get support for these kind of setups. The community is open for discussion of course, but we will not provide any kind of support here. I hope you understand.

Hi diekuh , Do you have an ETA for this (i.e. an approximate due date)? We are wanting to move to a clustered mail setup ASAP (as we are coming to the end of the contract with our current MX provider), and would love to know when we can expect to be able to implement this. I know you’ve said a “couple days/weeks”, but it’d be great to know whether we are talking ‘a couple of days’ or ‘several weeks’. 😃 Also, could you give a little more info on these support packages, please? I presume these would just be for Mailcow-provided assistance with any problems on an instance, rather than putting clustering behind a paywall (as is done with some other platforms, like Axigen)? Thanks in advance for all your hard work on putting this (and the entire solution) together! It’s truly amazing…

Thanks. 🙂 We offer them currently hosted in Willich and FFM (active/active). Docs are not available at the moment. I don’t have any more information as of today. A “paywall” sounds strange. You don’t go in the supermarket and call the product price a paywall, do you? 🙂

Mailcow Clustering

BusterB

I would gladly make a contribution to a Stay-Awesome-License (SAL) for the clustering documentation or ability – please let me know what’s reasonable. If Google Translate will kind of work with the German documentation, I’m fine with that. I even offer to help with some of the English documentation and how-to, plus I have some VMs and systems I can test with, to validate.

Bleala

Are there some news about the cluster docs?
I would really like to setup Mailcow in a cluster, but i don’t know exactly how.

BusterB

Unfortunately, I’ve not received any update and am still going with my two-host hot/warm kind of config. Basically, I run mailcow on two hosts (same config) and use KEEPALIVED to check on their health. It’s not elegant, took a lot of custom coding/scripting (e.g. custom certs renewal) and it’s almost perfect, but I’m having an issue with DKIM right now but I think it’s unrelated.)

pamart

Hi,
Any news about clustering docs ?

Thank you to Mailcow’s devs for the great job ;-)

BusterB

No. I’ve not heard from anyone, unfortunately. Mailcow’s a great product that would be AMAZING if it could be easily clustered for H/A. I know it’s possible and heard it involves running the DB in an H/A config, but I didn’t want to experiment too much with this, since I’m still learning docker and didn’t want to do too much “customization” to mailcow.

My config is not truly H/A in that there is no real load balancing. Keepalived properly keeps mail1 as the primary, but no mail flows to mail2 unless mail1 is down. The other issue is what really should happen is the DB needs to sync, not just the emails.

My config is basically setup for <disaster recovery> in that if mail1 goes down unexpectedly, I’ll still receive incoming mail on mail2. The problem I have is restoring mail1 once it’s back and getting the new emails from mail2 over to mail1. I can answer the challenges separately if anyone’s interested, but it’s manual and a bit ‘messy’. For small (home) configs and short outages, my setup works fine in that I won’t lose mail with one host down, which is great, but for large # of mailboxes or prolonged outages, mail recovery this way is a lot more problematic.

To make this work, I had to: 1) make sure the crypt keys were the same for both servers. 2) Use slightly different docker-compose.override files <and> directories 3) set up each server as stand-alone, so they use their own DB instance. 4) Leverage the GUI’s sync (imapsync).

Mailcow is great and customization is possible, just not ‘out-of-the-box’ nor for novices. I went with my config this way since it’s basically two ‘cloned’ stand-alone instances with keepalived running. I can live with the mailsync issue as it’s not that big of a deal if/when there is a crash.

MrChrisRodriguez

BusterB if you’re able and willing to share, I’d also benefit greatly from this. Thanks!

BusterB

OK, I think I finally have it working properly now. It’s not a complete or full high-availability cluster for everything as it doesn’t replicate all settings (e.g. I don’t replicate rspamd and a few other things), but it does what I want, which is it syncs emails between two separate hosts, and if the primary goes down, the other takes over seamlessly, transparently, and with no lost emails. My implementation is not something I would call “enterprise ready” without more work, but it’s “good enough” for me and might be a good starting point for someone else that wants to learn more.

The configuration involves a combination of keepalived at the OS level, and SQL master-master and dovecot replication at the app layer. It’s not a difficult or complex config, but there are some prerequistes, you have to do some things in a certain way/order and you also need to change the mailcow.conf and docker-compose.override.yml file as well the .cnf files for both dovecot and sql. I’m guessing it <should> survive an update.sh but I haven’t tested that out yet.

I don’t mind sharing my build book with anyone (devs?), but it’s long, step-by-step and I don’t have a place to host it. I’m also sure I missed a bunch of things and there are probably some config settings that I’m not doing right, nor did I do any optimizations you’d expect for large systems. (Note that I only have four (4) email accounts and low volume….) If you’re interested, like this post, reply to it, and don’t forget to thank and/or support the amazing mailcow devs for providing what we have now.

boohoomoo

BusterB
Sounds great! I, too, am interested in setting this up. Would you mind sharing your notes?

Have you considered posting them on GitHub (or similar)? This way, others would be able to contribute to your work. I am not an expert of this myself, unfortunately, so can’t do more than suggest you try it out.

And, of course, many thanks to the devs for giving us such a great tool! die Kuh rockt!

prival

BusterB I’m very interessed in this. Much appriciated for sharing.

Kyzoeadmin

WE would be very interrested in this.

DerLinkman

Hi Niklas here from which André spoke a while (almost a year) ago 👋

We have a instruction guide prepared which is (at least now) only accessible internally.

The plan is to rewrite that instruction and publish it (probably the docs site?).

However we know about a major issue regarding the mailcow UI which might be the reason why it hasn´t been released yet which is about the Redis DB and cluster behaviour. Currently the clustering setup is (as far as i`m informed) only build for 2 mailcow Servers not 3 or more. But exactly here comes the current Redis Issue into place which i was talking about. Because in the current design of HA mailcow your UI dies if the master Server dies or is not reachable. All Mails etc. will still be received and send but the Admin UI (not SOGo) isn´t accessible then.

The reason for this behaviour is the 3 Node (min) Redis Clustering requirement which the current mailcow HA don´t fulfills.

Besides that UI issue mailcow is usable in a HA Cluster.

So yeah. That is one if not the main issue why it´s been quiet about that topic.

When the documentation is releasing is not sure, probably if we fix that issue with Redis mentioned above. It´s not done with: “Simply deploy a third redis instance anywhere” as i know.

But yeah to sum it up: The Topic is not cancelled or has been forgotten internally!

trob

DerLinkman
Hi Niklas,
any chance you could share this guide? I’m in the process of setting up a simple load balancer in front of my mail servers and would like to cross check.

Thanks in advance!

ruonline

Hi, I’d also love to hear any developments on this.

At the moment I’ve got a setup that I think is pretty close to being redundant: I have 2 VPS’s running mailcow, with CloudFlare managing the DNS for my domain. The primary server (mail.mydomain.com) uses the cold-standby backup script (/opt/mailcow-dockerized/create_cold_standby.sh) to create/update a mirror image onto the 2nd VPS every hour (1 line cron job). It uses rsync, so doesn’t seem to transfer much data each time it does this.

So my cold-standby server is always ready to roll, and is only ever an hour out of date, so if my primary server ever has a major crash, I can switch the A record for mail.mydomain.com over to the IP address of the cold-standby server and boom, my clients can continue using their emails straight away, as though nothing is wrong (there’ll be a few emails missing that arrived in that last hour).

I also use the other backup script (helper-scripts/backup_and_restore.sh) to send a daily backup of my primary server to my Google Drive, and I keep a couple of days’ worth of these, just in case my cold-standby server has an issue.

I still need to figure out how I’m going to retrieve and restore those missing emails from that 1-hour gap, once I recover the data from my primary server again.

I’m also toying with setting up a backup MX server, but it would take a bit of tweaking to keep it updated with the list of domains/mailboxes/spam rules from the primary server, so I’ll worry about that later.

Does anyone have any opinions/improvements/suggestions on this setup? Should I run my cold-standby cron job every 5 minutes, or at some point of frequency will this cause issues (5 minutes/1 minute)?

justiceserv

I’m currently wondering if shared docker volume with ocfs2 (shared Block Volume in Oracle Cloud) would do the trick with few disabled & custom scripts.
Maybe with Load Balancers too?

jonboy86

I would be very interested in setting up a cluster as our mail provider has decided email hosting is not for them anymore . having dual locations each with there own internet and haing this fall over would be awesome

simonc

Hi there!

Any progress on the thread? I would really like to setup this.

DerLinkman

Hello 👋

No i’m afraid not. We cancelled this cluster setup so there won’t be any guide on how to. Even if will be such cluster stuff more part of a paid model to give the project some value to grow except Donations.

If we release a clustering method we let you know and how to get it.

Thanks!

simonc

DerLinkman Okay. So what is the recommended method to run mailcow in failover setup?

I know I can setup a second MX with a server that will relay messages - so nothing from inbound will get lost when mailcow is down. But how Can I do the same for SMTP? So we can also send emails?

thanks!

esackbauer

simonc so nothing from inbound will get lost when mailcow is down.

Nothing will get lost even without secondary MX, because sending mail servers will retry at least for 48 hours.
WIth a secondary, you will have the problem that the sender will think the mail was received (and read), but if the sending mail server will retry only the one and only mail server, the sender will get a bounce message after 4 hours informing him about delayed delivery.

aggiepilot

I reached this post searching for high availability as I have my home server and am colocating a server. Given the current availability I’ll probably just migrate mail to the colocated server and keep a cold backup at home but would be really nice to have a cluster between home and the datacenter. I would certainly pay for a “premium” level for this feature.

« Previous Page Next Page »