Parsing Output Data

The data package parses output from benchmark and verification testing into data objects.

Data Flow

The overall data flow between the various major processes in this repository is shown in the following diagram:

The processes on the left are completely disconnected from the ones on the right. Configuration objects necessary for benchmarks and verification testing are not available to the commands tabular and server, they only have text data output from the testing.

Benchmark tests generate two main types of output:

The first type is the benchmark data itself, generated by go test -bench in lines like the following:

BenchmarkSlogJSON/BenchmarkLogging-8   50026   21898 ns/op   20678166.73 MB/s   0 B/op   0 allocs/op

There is no way to control the output, so it is necessary to parse the string at the beginning of the line to get the handler tag (SlogJSON) and the test tag (Bench:Logging).

The second type is the warning data collected during the benchmark testing process:

# Warnings for chanchal/zaphandler:
#   Implied
#      1 [SourceKey] Source data not logged when AddSource flag set
#          SimpleSource: no 'source' key
#            {"level":"info","time":"2024-02-25T08:21:51-08:00","msg":"This is a message"}
# 
# Warnings for phsym/zeroslog:
#   Implied
#      1 [SourceKey] Source data not logged when AddSource flag set
#          SimpleSource: no 'source' key
#            {"level":"info","caller":"/home/marc/work/go/src/github.com/madkins23/go-slog/bench/tests/benchmarks.go:71","time":"2024-02-25T08:22:26-08:00","message":"This is a message"}
# 
# Warnings for samber/slog-zap:
#   Implied
#      1 [SourceKey] Source data not logged when AddSource flag set
#          SimpleSource: no 'source' key
#            {"level":"info","time":"2024-02-25T08:23:02-08:00","caller":"tests/benchmarks.go:71","msg":"This is a message"}

Verification tests generate only one type of output: verification warnings. These look like the warnings generated by the benchmark testing (see above) but there are generally a lot more of them.

Parser Data

The parsing of output data from benchmark and verification tests is done by two entities from the data package:

• the benchmark parser internal/data/Benchmarks.ParseBenchmarkData and
• the warnings parser internal/data/Warnings.ParseWarningData.

Handler Names and Tags

Handlers have tags and names:

• tags are restricted in form
  • they are parsed from benchmark test function names and
  • are to be used in web URLs by server and
• names are free form and generally of the form Phsym/zeroslog or Slog/JSONHandler.

The only time when tags and names are both accessible is during benchmark testing. The test function name can be parsed and the handler name is provided by the infra.Creator object composed into the SlogBenchmarkSuite object. However, this data is required by the Warnings parser which is a separate process.

The solution to this problem implemented herein is to declare the relationships between handler tags and names to be a separate data element. This third type of data is emitted during the benchmark testing process as a set of lines:

# Handler[SlogJSON]="slog/JSONHandler"

which link the handler tag (SlogJSON) with the handler name ("slog/JSONHandler"). One of these lines is emitted for each defined handler.

Parser Setup

As shown in the following diagram:

the benchmark parser gathers from the benchmark output:

• benchmark data,
• handler name/tag mapping, and
• benchmark warnings as unprocessed text.

After this the warning parser takes the handler name/tag mapping from the benchmark parser and parse the benchmark warnings and verification warnings separately, combining the result together into one warning data block.

The data.Setup function is designed to execute both parsers with the appropriate order and arguments. Create one parser of each type and pass them to the data.Setup function. The new parser objects will contain the parsed data unless an error is returned.

Scoring

Scoring data is derived from benchmark and verification results. The scoring algorithms should not be considered authoritative, they were constructed just to be able to draw the Handler Scores cmd/server pages (Default page shown, other pages with experimental scoring algorithms available). Documentation for the scoring algorithms is available on bottom of the score pages.