[Mimir] Add a "simple-scalable" version of helm chart

[ghost]

Similar to Loki deployment charts add a "simple-scalable" option for Mimir (--target=all) where deploying in full microservices mode on K8s is too complex/unnecessary.
Alternatively split only the read and write path without having each component separate.

[krajorama]

Hi, both are under discussion at Grafana, and read/write path should be coming first, but I don't have a timeline at the moment.

[flokli]

Also very interested in this :-)

[interpeix]

That could be nice to have, to be able to run Mimir in a small platforms in a simple way, also with monolithic deployments.

[joshuapare]

This would be great simply for the write path consolidation, as the inter-AZ networking costs between ingesters and distributors is very expensive on the write path (talking thousands of dollars) - we've had to buffer this by zone-locking the ingesters and distributors.

[AlexisDucastel]

Can't wait to have such chart :) I would be very awesome to provide a step between single monolithic and full micro services.
👍

[gclawes]

Would this effectively be a monolithic deployment if it was configured with just 1 replica?

[joshuapare]

@gclawes not necessarily, as you would still need a read pod and a write pod - the goal here is to eliminate the high inter-AZ networking costs on each path, while still allowing for distributed load with separately scalable write and read needs across availability zones for fault-tolerance.

We use Mimir at Mosaic Learning, and we had some very high inter-AZ networking costs - we were essentially forced to zone lock each path with affinity rules (many thousands of dollars a month).

[juanchovelezpro]

any news about the mimir simple scalable helm chart ?

[rubenvw-ngdata]

Since the helm chart has been moved to https://github.com/grafana/mimir, probably good to log a ticket for this over there.
I created grafana/mimir#4832