Łukasz Mierzwa

> @prymitive thanks for that; somehow the buffer it's trying to use is too small. Anything you can tell me about the query being executed when it crashed? Sadly no,...

> The query executed is printed on the next startup when Prometheus reads the active query logging Le mer. 6 juil. 2022, 18:46, Łukasz Mierzwa ***@***.***> a écrit : >...

I've added patch from #10995 and I see a few panics in logs. They all seem to have range queries. Here's an example: ``` component="query engine" msg="runtime panic in parser"...

Running a range query for `traffic_manager_local_interface_reactor_action_successful` over past 30d returns a list of time series and they're not all present over that time, that's about it really.

Adding some debug logs with: ```diff @@ -717,6 +719,9 @@ func marshalLabelsToSizedBuffer(lbls []Label, data []byte) int { func marshalLabelToSizedBuffer(m *Label, data []byte) int { i := len(data) i -= len(m.Value)...

```diff @@ -708,6 +709,15 @@ func (b *Builder) Labels() Labels { func marshalLabelsToSizedBuffer(lbls []Label, data []byte) int { i := len(data) for index := len(lbls) - 1; index >= 0;...

I did deploy it to the same test server and, so far, I don't see any panics, so it does look like the issue is resolved. I've also captured heap...

Somehow `external_labels` are missing from `/api/v1/status/config` response, as if marshaling to JSON didn't work properly.

This is go alloc size (`avg_over_time(go_memstats_alloc_bytes[30m])`) on a test instance. We have: - v2.36.0 + patches (#10782 & #10709 patches) running initially - at 08:50 main+this patch starts - at...

Thanks! I can confirm that it fixed external_labels issue