[mimir-distributed] Add support for the query-frontend results cache

[Logiraptor]

Fixes #1403

Tested locally that the results cache is used for subsequent requests for the same data.

Note fetched_chunk_bytes, etc are 0, and response_time is sub 10ms.

level=info ts=2022-05-24T17:33:25.8620796Z caller=handler.go:223 org_id=anonymous msg="query stats" component=query-frontend method=POST path=/prometheus/api/v1/query_range response_time=123.8518ms query_wall_time_seconds=0.116888 fetched_series_count=1000 fetched_chunk_bytes=3394000 fetched_chunks_count=3000 sharded_queries=0 param_end=1653411855 param_query=sum(cortex_load_generator_sine_wave) param_start=1653409470 param_step=15                                                                                          
level=info ts=2022-05-24T17:33:31.9627418Z caller=handler.go:223 org_id=anonymous msg="query stats" component=query-frontend method=POST path=/prometheus/api/v1/query_range response_time=2.3415ms query_wall_time_seconds=0 fetched_series_count=0 fetched_chunk_bytes=0 fetched_chunks_count=0 sharded_queries=0 param_start=1653409470 param_step=15 param_end=1653411855 param_query=sum(cortex_load_generator_sine_wave)                                                                                                               

Also tested upgrade from previous chart and enabling / disabling the results cache.

Signed-off-by: Patrick Oyarzun patrick.oyarzun@grafana.com

[krajorama]

hi, I think we need to update the small, large, capped-small, capped-large yamls as well and decide what defaults to use. Technically this is backwards compatible, however if we enable it by default then we should mention something about the increased resource use in the CHANGELOG.

[Logiraptor]

@krajorama Thanks for the feedback. I updated the example values files. I chose 4 replicas for the large.yaml by checking on some of our larger cells in Grafana Cloud. I think we will need to do additional tuning of all four memcached replicasets later when we start to validate against the 500M AS scale settings. I opted not to do that now since we should really take a wholistic view of the system.

[krajorama]

LGTM

[Logiraptor]

Closing this PR to migrate it to https://github.com/grafana/mimir