Kaviraj Kanagaraj

Some findings. So the docker-driver (with fix) seems to work(at least some cases) With `grafana/docker-driver:2.1.0` (without the fix) - Container takes long to stop. ``` -bash5.0$ docker plugin install grafana/loki-docker-driver:2.1.0...

take

> @kavirajk -- have you had a chance to work on this issue? If not, do you mind if someone else does? Sorry @alamb didn't get a chance to work...

@pstibrany @scalalang2 Sorry for late response, just back from holidays. I do have some working example for ring package will push it today or tmrw.

Going to use this same issue to track some more examples/documents I want to add for `ring` package. - [x] [Simple example on how to use `ring` package](https://github.com/grafana/dskit/pull/272) - [...

> Does having a lot of pods shut down at the same time not cause other problems, like difficulty updating routing for the Service? No, not really (AFAIS). Only problem...

> How come the replacement pods can insert themselves in the ring and become ready at the same rate? I think that's because we [have `initRing` called in the main...

I guess different tradeoff here. Tried following cases to get some metrics about CAS rate. 1. In normal rollout, the CAS operations rate reaches almost 700 reqs per sec. (with...

I guess we can play with these k8 configs to reach good default values for distributors, but looks like we always have to pay some rollout time as a cost.

Here is a another one. Keep k8 configs same (maxSurge and maxUnavailable to 25%) But just increase the retry from 10 -> 100. Rollout tooks same ~2mins, No Unhealthy. But...