Follow up on known PostingsForMatchersCache issues

[aknuds1]

As mentioned by @charleskorn in #546, there are a couple of outstanding issues in PostingsForMatchersCache.PostingsForMatchers that should be followed up on:

• Should other kinds of errors also not be cached?
• If all waiting calls to PostingsForMatchersCache.PostingsForMatchers() have cancelled their context, should we cancel the context passed to the underlying PostingsForMatchers implementation?

[charleskorn]

One other thing that needs to be done once those questions are answered: we should upstream the fix in #546 (and any other related changes) to Prometheus.

[charleskorn]

@aknuds1 I see you've assigned yourself to this issue, but let me know if you need any help.

[aknuds1]

we should upstream the fix in #546 (and any other related changes) to Prometheus

@charleskorn which changes from #546, considering PostingsForMatchersCache isn't upstream? Sorry if I'm missing the obvious.

[aknuds1]

I see you've assigned yourself to this issue, but let me know if you need any help.

@charleskorn I unassigned myself, since I figured it's better to leave it free in the (query) backlog, so whoever has the capacity to pick it up can.

[charleskorn]

we should upstream the fix in #546 (and any other related changes) to Prometheus

@charleskorn which changes from #546, considering PostingsForMatchersCache isn't upstream? Sorry if I'm missing the obvious.

Oh, I misunderstood, I thought this was an upstream feature. In that case, no upstreaming required 🙂

[aknuds1]

Haven spoken offline with @charleskorn, it might be preferable if I pick this issue up if I have the capacity, given that I possess the most context wrt. making Prometheus support query cancellation. I'll have to see if bandwidth opens up to work on this.

[pracucci]

Fixed by #644