search

Mailu / helm-charts

Development repo for helm charts
126 stars 130 forks source link

Add replicaCount for admin, dovecot, webmail and oletools #303

Open Deltachaos opened 1 year ago

Deltachaos commented 1 year ago

In general those services can easily deployed with multiple replicas, when the storage is shared, without any issues. So why not be able to configure them?

fastlorenzo commented 1 year ago

We'd need to test this in length in term of session management. However, as stated in #304 , there is potential issues with running 2 dovecot instances on the same backend storage.

github-actions[bot] commented 11 months ago

This PR is stale because it has been open 45 days with no activity. Remove stale label or comment or this will be closed in 10 days.

nextgens commented 11 months ago

That is definitely broken. The only things you can safely scale up are: unbound, oletools, clamav, tika and webmail; All the other containers can't^wshoudn't be scaled up without changes in Mailu

github-actions[bot] commented 10 months ago

This PR is stale because it has been open 45 days with no activity. Remove stale label or comment or this will be closed in 10 days.

Deltachaos commented 8 months ago

@nextgens why should admin not be scaled up? i run this setup since a few months and have not experienced any problems.

Regarding postfix the interesting point would be what problems do you expect there? If they dont share storage, but instead all have there own storage, then one mail should be randomly delivered to one of the postfixes and send from there.

I also have not experienced any issue with this so far.

github-actions[bot] commented 7 months ago

This PR is stale because it has been open 45 days with no activity. Remove stale label or comment or this will be closed in 10 days.

nextgens commented 1 month ago

@nextgens why should admin not be scaled up? i run this setup since a few months and have not experienced any problems.

Because until 2024.06 it had some shared state. I don't think that's true anymore so it is probably okay to do so now.

The imap container (running dovecot) is still problematic. To scale that up you should find a way to emulate what 'dovecot director' does: ensure that all clients sharing the same username are steered towards the same dovecot backend regardless of the protocol they use. Failing to do so means that indexes will never be kept up to date across backends ... and that will put the underlying FS under a lot of strain and will get you terrible performance. A basic implementation of this is trivial to do (probably 3 lines of code); the devil is in the detail. To do it right you probably need to have visibility on the healthchecks of the various backends... which means reinventing the wheel in a docker-compose context. Is there a standardized way of doing it in k8s land?

nextgens commented 1 month ago

Correcting myself above: webmails cannot be safely scaled as is. PHP sessions and whatknot are not in redis; If requests are load-balanced across instances it will break.

github-actions[bot] commented 1 day ago

This PR is stale because it has been open 45 days with no activity. Remove stale label or comment or this will be closed in 10 days.