Expand description
Parse, pair, aggregate, compare, and report compact benchmark markers.
The normal pipeline is parse_benchmark_events,
pair_benchmark_spans, aggregate_benchmark_spans, and optionally
compare_benchmark_aggregates plus write_benchmark_report_dir. Marker
producers can use format_marker on the host or
crate::performance::Performance in canister code.
Totals and run counts remain exact integers. Means use IC Metrics’ exact scaled integer projection, rounded down. Callers choose the scale and handle projection errors; reports use scale one in the original counter units. Comparisons retain current and previous means instead of percentage estimates.
Structs§
- Benchmark
Aggregate Error - Overflow rejects the complete aggregation, without returning partial totals.
- Benchmark
Aggregate Report - Benchmark
Aggregate Row - Benchmark
Averages - Floor-rounded scaled means of exact aggregate totals and run counts.
- Benchmark
Comparison Report - Benchmark
Comparison Row - Current and previous scaled means, preserving missing rows and measured zero.
- Benchmark
Counters - Benchmark
Parse Report - Benchmark
Parser Config - Benchmark
RunDirectory - Benchmark
RunMetadata - Benchmark
RunReport - Benchmark
Span - Benchmark
Span Report - Invalid
Benchmark Span - Malformed
Benchmark Marker - RawBenchmark
Event - Unpaired
Benchmark Marker
Enums§
- Benchmark
Counter - Counter whose aggregate cannot be represented.
- Benchmark
Event Kind - Benchmark
Event Source - Suite
Derivation - Unpaired
Benchmark Marker Kind
Constants§
- DEFAULT_
PREFIX - Default prefix for benchmark marker lines.
Functions§
- aggregate_
benchmark_ spans - Aggregate spans by named suite and across all suites.
- benchmark_
run_ directory_ name - compare_
benchmark_ aggregates - Pair current and previous means at the same caller-selected scale. Returns a typed error if either side cannot supply the selected projection.
- find_
latest_ previous_ run - Find the latest earlier run with readable metadata and a matching command.
- format_
marker - next_
benchmark_ run_ directory - Compute the next available benchmark run path for one timestamp and commit.
- pair_
benchmark_ spans - parse_
benchmark_ events - parse_
benchmark_ events_ from_ captured_ output - Parse separately captured stdout and stderr.
- parse_
benchmark_ events_ from_ source - read_
benchmark_ run_ metadata - write_
benchmark_ report_ dir - Write one complete benchmark report into the requested directory.