Store span warnings as tags in Cassandra

[utsavoza]

Which problem is this PR solving?

• Resolves Span.Warnings not stored in Cassandra #705
• Resolves Cassandra's dbmodel's Span.SpanHash depends on data that isn't in dbmodel #704

Short description of the changes

• Instead of creating a separate column in the table, we encode warnings as tags, with a magic string prefix so that they can be parsed and converted back to warnings during reads.

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.01 🎉

Comparison is base (43c29a0) 97.09% compared to head (2c422d9) 97.10%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4313      +/-   ##
==========================================
+ Coverage   97.09%   97.10%   +0.01%     
==========================================
  Files         300      300              
  Lines       17668    17700      +32     
==========================================
+ Hits        17155    17188      +33     
+ Misses        413      412       -1     
  Partials      100      100              

Flag	Coverage Δ
unittests	97.10% <100.00%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...n/storage/cassandra/spanstore/dbmodel/converter.go	100.00% <100.00%> (ø)

... and 1 file with indirect coverage changes

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[yurishkuro]

please add tests for error handling paths that are not covered (see codecov comments)

[yurishkuro]

please change it back. Original version has no copying and no memory allocations, yours has both. And pre-allocating array of known size is usually more efficient because it avoids re-allocation as the vector grows.

[yurishkuro]

because your input and output lengths may not batch, you have to allocate with make(type, 0, len) and use append, not indexed assignment.

[utsavoza]

done

[yurishkuro]

@utsavoza how/why do you think this fixes #704?

[utsavoza]

@yurishkuro I don't think this solves/fixes #704 unless we add Warnings as a separate field in the dbmodel's Span. I've updated the PR description to exclude the issue.

[yurishkuro]

More I think about it, I think it does solve the issue. #704 doesn't actually explain the problem, it instead points to implementation details. I think the problem is that the spanHash used in the stored version is supposed to be different if anything about the span is different, including the warnings. This is how hash is calculated:

jaeger/model/span.go

Lines 63 to 68 in a631623

	func (s *Span) Hash(w io.Writer) (err error) {
	// gob is not the most efficient way, but it ensures we don't miss any fields.
	// See BenchmarkSpanHash in span_test.go
	enc := gob.NewEncoder(w)
	return enc.Encode(s)
	}

Since warnings previously were not part of the DB model, two identical domain spans that are only different in warnings would produce different hash code, even though when they are stored in the DB their stored image would be identical. I am not sure why even that would be a problem, sounds more like a theoretical issue that one would not come across in actual usage. But now that warnings are converted into special tags, the stored versions will be different, thus justifying the different hash codes, as was the intention. So feel free to keep the "resolves #704" in the description.