mirror of
https://github.com/prometheus/prometheus.git
synced 2026-10-28 10:44:59 +08:00
rbizos
ffe0b67b81
* adding OM2 scrape parsing Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Renaming native histogram buckets Also adding test case for omitted buckets / span Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * better test unknown metrics Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Enforcing examplar timestamp and remove requirement on unit suffix Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * adding test exemplar above previous limit Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * supporting having both native and classic histograms if both provided Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * adding benchmarks for the diff between om1 and om2 Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * reusing the label builder Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * using an iter for splitting composite field Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * fixing native histogram parsing Before we would not check if buckets are floats or not, this ensures that we don't miss some edgecases Also adding test coverage on the different cases Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * reduce allocations in composite value parsing - converts parseBuckets/parseQuantile to iterators - reusing pending series Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * fixing native histograms format renamed correctly the gauge histogram om encode buckets as positive absolutes count in stead of delta Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Update model/textparse/testdata/alltypes.bench.om2.txt Co-authored-by: George Krajcsovits <krajorama@users.noreply.github.com> Signed-off-by: rbizos <58781501+rbizos@users.noreply.github.com> * Update model/textparse/testdata/alltypes.bench.om2.txt Co-authored-by: George Krajcsovits <krajorama@users.noreply.github.com> Signed-off-by: rbizos <58781501+rbizos@users.noreply.github.com> * Update model/textparse/testdata/alltypes.bench.om2.txt Co-authored-by: George Krajcsovits <krajorama@users.noreply.github.com> Signed-off-by: rbizos <58781501+rbizos@users.noreply.github.com> * Update model/textparse/testdata/alltypes.bench.om2.txt Co-authored-by: George Krajcsovits <krajorama@users.noreply.github.com> Signed-off-by: rbizos <58781501+rbizos@users.noreply.github.com> * Update model/textparse/testdata/alltypes.bench.om2.txt Co-authored-by: George Krajcsovits <krajorama@users.noreply.github.com> Signed-off-by: rbizos <58781501+rbizos@users.noreply.github.com> * Moving handwritten code from lex to parse There was some leftover manual code on top of generated code Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Fixing type and unit labels - added tests covering each types - fixing missing type and unit for composite metrics - fixed unit not being cleaned between metrics families Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Fixing bucket ordering for test data added an explicit unit test to cover the unorder case Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Reset Type and unit with any metadata Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * reusing parsingLVals for examplars Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * reused ParserOptions in stead of dedicated type Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * fail on missing fields for NH Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * fixing wrong link Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * reject duplicated st Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Fixing conflicting Series return for composite metrics Reused the logic from protobuf that suffers the same issue Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Update model/textparse/openmetrics2parse_test.go Co-authored-by: George Krajcsovits <krajorama@users.noreply.github.com> Signed-off-by: rbizos <58781501+rbizos@users.noreply.github.com> * Adding UTF-8 validation in labels and metadata also fixed a small mistake in the tests where the input was updated but not the test expected value Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Making metrics name parsing stricter - metric name must not be empty - metric name without explicite __name__ must be first Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * simplify native histogram parsing using schema Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * Force ordering in buckets for histograms also fixed some texts using om1 data Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * reseting properly when seeing a new metric Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * adding missing gsum/gcount metrics for classic gauge histograms Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * making composite parsing reject unknown and duplicate field Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * reject invalid schema Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * fix invalid metric name/label parsing Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * adding strict validation of zero_threshold Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * making stricter composite value validation - enforcing +inf bucket for legacy histograms - enforce the presence of requeired field - force composite for types that require them Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * fixing timestamp ts order + summary quantile order Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * adding stricter validation of statesets Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * rejecting NaN counters Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * model/textparse: fix failing OM2 NaN counter test The "counter sample value must not be NaN" case declared MetricFamily "foo" via `# TYPE foo counter` but sampled `foo_total`. OpenMetrics 2.0 removes implicit suffix stripping and requires a Metric name to exactly match its MetricFamily name, so `foo_total` is treated as a different family. The parser therefore resets the type to Unknown, the counter NaN guard is skipped, NaN is accepted, and the test reaches EOF instead of the expected error. Declare the family as `foo_total` so the name matches and the guard is exercised. This makes `go test ./model/textparse/` pass again. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com> * model/textparse: make OM2 benchmark fixtures spec-compliant Several counters in the OM2 benchmark fixtures declared their MetricFamily via `# TYPE`/`# HELP`/`# UNIT` using a name without the `_total` suffix (e.g. `cache_hits`, `http_requests`) while sampling the `_total` name. OpenMetrics 2.0 removes implicit suffix stripping and requires a Metric name to exactly match its MetricFamily name, so those samples were treated as a different, metadata-less family and parsed as untyped (Unknown) instead of counters. Align the metadata names with the sampled `_total` names so the fixtures are valid OM2 and the counters parse as counters. Histogram, summary, gauge, info and stateset families already matched and are unchanged; histograms.bench.om2.txt needed no changes. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com> * model/textparse: make OM2 unit-test snippets spec-compliant Several inline test snippets used OpenMetrics 1.0 naming conventions that are invalid under OpenMetrics 2.0, which removes implicit suffix stripping and requires a Metric name to exactly match its MetricFamily name: - Counters declared as `# TYPE foo counter` / `# TYPE requests counter` while sampling `foo_total` / `requests_total`. The mismatch made the parser treat the samples as a different, untyped family. Align the descriptor name with the `_total` sample. - The Info family `ii` did not end in `_info`; the spec mandates Info MetricFamily names MUST end in `_info`. Rename to `ii_info`. Expected entries are updated accordingly. The affected error cases still exercise the same errors (timestamp handling, missing exemplar timestamp), which are independent of the metric type. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com> * model/textparse: test OM2 type resolution by metric/family name match Add a table-driven test contrasting the two OM2 naming outcomes under the exact-match rule (implicit suffix stripping is removed): - A sample whose name only suffix-extends the declared family (`foo_total` under `# TYPE foo counter`) does not belong to that family and is parsed as untyped (no __type__ label). - A sample whose name matches the descriptor exactly (`foo_total` under `# TYPE foo_total counter`) keeps the declared type, exposing __type__="counter". The existing family-reset test only covered a wholly unrelated undeclared name, not this OM1-style suffix case. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com> * adding a feature flag to activate OM2 support Signed-off-by: Raphael Bizos <r.bizos@criteo.com> * chore: pin OpenMetrics testdata fixtures to LF line endings Windows checkout (core.autocrlf=true) rewrites the new OM1/OM2 benchmark testdata fixtures to CRLF. The OpenMetrics/OM2 parsers treat "\r" as part of the token, which turns "counter" into "counter\r" and fails type validation, breaking TestOM1OM2BenchPairsEquivalent on the "Go tests on Windows" CI job. Pin these fixtures to LF via .gitattributes so checkout no longer mangles them. Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com> --------- Signed-off-by: Raphael Bizos <r.bizos@criteo.com> Signed-off-by: rbizos <58781501+rbizos@users.noreply.github.com> Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com> Signed-off-by: r.bizos <r.bizos@criteo.com>
7 lines
321 B
Plaintext
7 lines
321 B
Plaintext
web/api/v1/testdata/openapi_golden.yaml linguist-generated
|
|
|
|
# Ensure OpenMetrics test fixtures keep LF line endings on checkout.
|
|
# The parsers reject "\r" as part of tokens (OpenMetrics/OM2 text format is
|
|
# LF-only), so CRLF-mangled fixtures break parsing on Windows checkouts.
|
|
model/textparse/testdata/*.txt text eol=lf
|