Add sampling info to Tracestate headers

[mikker]

See elastic/apm#307 (WIP)

[apmmachine]

💚 Build Succeeded

Expand to view the summary

Build stats

• Build Cause: [Pull request #858 updated]

• Start Time: 2020-09-01T11:43:53.159+0000

• Duration: 24 min 7 sec

Test stats 🧪

Test	Results
Failed	0
Passed	66917
Skipped	47
Total	66964

Steps errors

Expand to view the steps failures

• Name: Shell Script

  • Description: ./spec/scripts/spec.sh ruby:2.7 rails-6.0

  • Duration: 2 min 30 sec

  • Start Time: 2020-09-01T11:51:34.379+0000

  • log

• Name: Shell Script

  • Description: ./spec/scripts/spec.sh ruby:2.5 sinatra-1.4

  • Duration: 1 min 24 sec

  • Start Time: 2020-09-01T11:51:29.136+0000

  • log

• Name: Shell Script

  • Description: [2020-09-01T11:50:42.886Z] + ./spec/scripts/spec.sh docker.elastic.co/observability-ci/jruby:9.2-13-

  • Duration: 4 min 16 sec

  • Start Time: 2020-09-01T11:50:42.597+0000

  • log

[axw]

Looks good overall. What happens if the es vendor value is invalid?
Is there a test that covers the case where a downstream transaction receives traceparent, and either tracestate is missing or doesn't have a valid es entry? In that case the transaction should have no sample_rate field.

[mikker]

What happens if the es vendor value is invalid?

Not sure but depends on what we mean by invalid? Not matching the k:v pattern or including an illegal utf char or something?

[axw]

What happens if the es vendor value is invalid?

Not sure but depends on what we mean by invalid? Not matching the k:v pattern or including an illegal utf char or something?

Yes (all of the above :))

Here's a few test cases which I think we should consider, to be defensive:

• "es=foo"
• "es=s:foo"
• "es=s:1.5" (s should be in the range 0..1)

[axw]

LGTM. I feel like we might all benefit from a Gherkin spec for this, WDYT?

[simitt]

Didn't follow semantics to closely, but LGTM from a ruby perspective.

[mikker]

@axw

I feel like we might all benefit from a Gherkin spec for this, WDYT?

I agree. As this would add significant complexity (incoming and outgoing reqs) to the currently very simple gherkin specs (spec, really), I'm going to add it in a separate PR.