disable index dedupe when rf > 1 and current or upcoming index type is boltdb-shipper

[sandeepsukhani]

What this PR does / why we need it:
We use dedupe caches for chunks and seriesIDs to determine whether we already have written a chunk or a seriesID already to the store. Since boltdb-shipper only uploads boltdb files periodically to make the index available to all the other services, an ingester which first wrote a chunk or seriesID goes down and other ingesters didn't write them due to deduplication, we would be missing some logs in query responses until that ingester is down. This would be a problem even during rollouts.

This PR disables index deduplication and avoid using write dedupe cache(which dedupes seriesIDs) when replication factor > 1 and current or upcoming index type is boltdb-shipper.

Checklist

• Documentation added
• Tests updated

[codecov-commenter]

Codecov Report

Merging #2206 into master will increase coverage by 0.03%.
The diff coverage is 61.53%.

@@            Coverage Diff             @@
##           master    #2206      +/-   ##
==========================================
+ Coverage   61.62%   61.65%   +0.03%     
==========================================
  Files         160      160              
  Lines       13577    13586       +9     
==========================================
+ Hits         8367     8377      +10     
+ Misses       4588     4586       -2     
- Partials      622      623       +1     

Impacted Files	Coverage Δ
pkg/loki/modules.go	11.52% <61.53%> (+2.38%)	⬆️
pkg/querier/queryrange/downstreamer.go	95.87% <0.00%> (-2.07%)	⬇️
pkg/canary/comparator/comparator.go	80.24% <0.00%> (+2.41%)	⬆️

[slim-bean]

I was hoping we wouldn't need this if we could update cortex to fix the underlying issue?

[sandeepsukhani]

I am not sure which issue we are talking about here. This is based on a change in Cortex that I did which helped with this.

[sandeepsukhani]

Another note, this problem is specific to only boltdb shipper because all the other stores have strong consistency while boltdb shipper doesn't.

[slim-bean]

I will look at this closer, maybe it's the naming that confuses me.

We don't want to disable write de-duplication of chunks, we still want the chunks to be written to the cache.

Instead we just want to make sure when a chunk is already in the cache we still write the index entry.

This does get more complicated however if the filesystem is not shared, I will look closer at this.

[sandeepsukhani]

@slim-bean I have changed the flag in Cortex to allow disabling of just index deduplication. If enabled, it would still write just the index even if the chunk is already written.

[sandeepsukhani]

@slim-bean let us merge this PR which has a helper function to check whether current or upcoming index store is of type boltdb-shipper. This would help with doing changes to #2166 that we discussed yesterday.

[slim-bean]

LGTM!