Ganesh Vernekar
Ganesh Vernekar
And empty span seems like the least bad way to do it.
Speaking specifically with out-of-order support in mind, since it is already out in the wild, the simpler fix seems attractive for a start while we can look at the complicated...
Ignoring out-of-order, we still need to cover a minimum of 1h old queries, since they can still get new timeseries that are not out-of-bounds. OOO time window might be 30m...
@juliusv @Nexucis: we can use your help on this whenever you get time. (apologies for the double notification, this is same as https://github.com/prometheus/prometheus/issues/11171, but a more focused issue)
Hey @prymitive, just popping in here to say that I haven't forgotten about this. I hope to get back to this soon. Multiple conferences (and hence conference driven development) taking...
> I was wondering, wouldn't it be easier and less complex to log to regular WAL after commit (some new committed_sample type) and implement cleanup to account for out-of-order samples,...
TODO before v2.39 - add docs and fix https://github.com/prometheus/prometheus/issues/11255 to make it simpler to use out-of-order.
/prombench af6167df58c206daddccdeba3e9d9860f4ff611b
/prombench pre-ooo I created a branch out of the above commit, to only test this PR. This benchmark is to check any overheads without enabling OOO.
Looks like I can't test against a branch or commit. So I will do it as part of the usual release benchmark then (v2.39 coming soon).