Default to internal_error rather than unknown_error

[cyberdelia]

When using sentry-opentelemetry to trace code that isn't related to an HTTP call, the SpanProcessor will set the status of a Transaction to be "unknown_error", which is benign, but within Sentry UI this status is not interpreted as an error at all.

The result is that all the impacted transactions appears to have 0% failure rate, etc. Changing it to "internal_error" prevents this, and also is in line with other sentry SDKs behaviour.

Sidebar: The SpanEvent metadata around exceptions seems to be dropped in the conversion process to Transaction , which is too bad.

[solnic]

This makes sense, thanks. Could you also add a simple spec that covers this change to span_processor_spec.rb?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.17%. Comparing base (a9b3687) to head (26e5605).
Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2473   +/-   ##
=======================================
  Coverage   98.17%   98.17%           
=======================================
  Files         128      128           
  Lines        4825     4829    +4     
=======================================
+ Hits         4737     4741    +4     
  Misses         88       88           

Components	Coverage Δ
sentry-ruby	98.57% <ø> (+<0.01%)	⬆️
sentry-rails	97.08% <ø> (ø)
sentry-sidekiq	96.96% <ø> (ø)
sentry-resque	92.85% <ø> (ø)
sentry-delayed_job	95.65% <ø> (ø)
sentry-opentelemetry	99.31% <100.00%> (ø)

Files with missing lines	Coverage Δ
...lemetry/lib/sentry/opentelemetry/span_processor.rb	100.00% <100.00%> (ø)

... and 1 file with indirect coverage changes

[cyberdelia]

@solnic Done. The new spec cover all type of errors handled by update_span_status, since it was not tested before.

[solnic]

Looks good, thank you! Please update the top-level CHANGELOG.md too, and I'll be able to merge it in 🙂