Fix reporting of tracing spans from PromQL engine

[pracucci]

What this PR does

In the PR #1695 we upgraded the vendored Prometheus, which included a change to the tracing, migrating from opentracing to opentelemetry. This change broke the reporting of tracing spans from PromQL engine, which are particularly useful when investigating slow queries.

I think the clean way to fix it would be migrating Mimir (and so weaveworks/common) to opentelemetry library, but it's risky change to ensure we don't break backward compatibility and I would recommend to do it but carefully.

As a quicker fix, I'm proposing to introduce a bridge to implement the opentelemetry tracer interface but using opentracing (and jaeger client specifically) under the hood. The opentelemetry library provides the bridge for the opposite direction only, so I had to manually write it.

I manually tested it and you can see a difference between before / after.

Before:

After:

Which issue(s) this PR fixes or relates to

Fixes #1954

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

[dimitarvdimitrov]

A few questions about some features that weren't ported. I am not sure how well these bridges should translate between OTEL and OT.

[pracucci]

A few questions about some features that weren't ported. I am not sure how well these bridges should translate between OTEL and OT.

Thanks for the valuable review! The bridge we need should be "good enough", given the limited real usage we're doing right now. Anyway, feedback was good!

[dimitarvdimitrov]

LGTM, thank you for addressing the feedback

[github-actions]

Helm <> Jsonnet Diff

⚠️ A difference was detected between the Helm chart and the Jsonnet library.

1. Use make check-helm-jsonnet-diff to reproduce the output locally.
2. This test is experimental while we gather feedback about its usefulness.
3. It does not block your PR from being merged, but we would appreciate you trying to keep feature parity between the Helm chart and Jsonnet library if possible.

If you get stuck on this step and the Mimir maintainers aren't able to help, feel free to merge without making this step pass and tag @Logiraptor so the Mimir maintainers can gather feedback later.

Please see the contribution docs here for more info.

Expand to see the output

Output of https://github.com/grafana/mimir/actions/runs/2847132247

Warning: policy/v1beta1 PodDisruptionBudget is deprecated in v1.21+, unavailable in v1.25+; use policy/v1 PodDisruptionBudget
diff -r -u -N scratch/./helm/07-config/ingester-MimirConfig.yml scratch/./jsonnet/08-config/ingester-MimirConfig.yml
--- scratch/./helm/07-config/ingester-MimirConfig.yml	2022-08-12 13:48:18.849649810 +0000
+++ scratch/./jsonnet/08-config/ingester-MimirConfig.yml	2022-08-12 13:48:25.953662070 +0000
@@ -553,7 +553,7 @@
     max_fetched_chunks_per_query: 2000000 (default)
     max_fetched_series_per_query: 0 (default)
     max_global_exemplars_per_user: 0 (default)
-    max_global_series_per_metric: 20000 (default)
+    max_global_series_per_metric: 0
     max_global_series_per_user: 150000 (default)
     max_label_name_length: 1024 (default)
     max_label_names_per_series: 30 (default)
diff -r -u -N scratch/./helm/07-config/overrides-exporter-MimirConfig.yml scratch/./jsonnet/08-config/overrides-exporter-MimirConfig.yml
--- scratch/./helm/07-config/overrides-exporter-MimirConfig.yml	2022-08-12 13:48:18.853649821 +0000
+++ scratch/./jsonnet/08-config/overrides-exporter-MimirConfig.yml	2022-08-12 13:48:25.961662089 +0000
@@ -498,7 +498,7 @@
     max_fetched_chunks_per_query: 2000000 (default)
     max_fetched_series_per_query: 0 (default)
     max_global_exemplars_per_user: 0 (default)
-    max_global_series_per_metric: 20000 (default)
+    max_global_series_per_metric: 0
     max_global_series_per_user: 150000 (default)
     max_label_name_length: 1024 (default)
     max_label_names_per_series: 30 (default)