High retention on Prometheus

[metalmatze]

In the example the retention for Prometheus itself is set to 24h.
When setting this to, for example, 1 week (168h) does Thanos query the object storage less for that time range?

[bwplotka]

Hello!
Good question. (:

TL;DR: No. For Thanos query it does not matter if the underlying Store API is store, sidecar or ruler.

This is related discussion: #82 I remember I started doing some implementation of it, but essentially this is bit fuzzy problem and not straightforward, so we aborted. Basically:

• you don't know how many stores you have in the system. We aimed to support one per bucket for now, but you can have multiple buckets.
• you don't really know what "sources" (what metrics) store node has access too.
• races can happen (although that could be mitigated by some time margin).
• we are not really sure if it's "better" to ask store or Prometheus (sidecar/source). Better means cheaper? faster? simpler? more resilient?

[bwplotka]

@metalmatze does it answer your question? Can we close this?

[metalmatze]

Yes. Thanks