gitlab_quality-test_tooling 3.18.3 → 3.19.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: f524117673a4935ef736ec70c0fbbdaef37e47ce9f6745560b39d27498b3d6dc
-  data.tar.gz: b576628aa613113b8969659d2e3f1e3267eb0ea8af12871fc062c9dd5c738b1a
+  metadata.gz: 5fb5356dd9a684e76ccf4de875f66d77ee818bef86d1cd9792eb8db9d688cfe9
+  data.tar.gz: 75e5afd55644a124fbf09f4fe0edd8f298fb2614f54dd0884a8a18002d2fe6d8
 SHA512:
-  metadata.gz: c01d93893558f7f7cdbfcc96787d8367583f83f66527799cf161def8f9d1d2444f7bd31f65b96926f7610b9992f95b676a34471be03cc3964ea3b8935fd358f0
-  data.tar.gz: d830ae9098aa8c636390350d6c12c706545b4a35bcd4022d1c194515b9a7c9f8d08dfa3079e2abb461e7647f521200acd951da496073f744ebd6cec459070620
+  metadata.gz: 35ff364a8fbbd437bef5e94a6bedccf66b8288965fd6c136ce6c21baa3d9052ac44281e54dd3a11eb92dae156f6e9e123756fa526ddb24fd39c00450b1d7187e
+  data.tar.gz: 1de79819dcb8aadf7ef3994f54387008ff9d1e15ea4a0b0590caebe9cf2af99a2821dcc7008b7bff5b2df9da69d4103622624983ce7434272ec2e5663c1e6695

data/Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    gitlab_quality-test_tooling (3.18.3)
+    gitlab_quality-test_tooling (3.19.1)
       activesupport (>= 7.0)
       amatch (~> 0.4.1)
       fog-google (~> 1.24, >= 1.24.1)

data/exe/test-coverage

@@ -14,6 +14,7 @@ require_relative '../lib/gitlab_quality/test_tooling/code_coverage/click_house/p
 require_relative '../lib/gitlab_quality/test_tooling/code_coverage/click_house/test_file_mappings_table'
 require_relative '../lib/gitlab_quality/test_tooling/code_coverage/click_house/jest_quarantined_tests_table'
 require_relative '../lib/gitlab_quality/test_tooling/code_coverage/click_house/test_health_risk_aggregator'
+require_relative '../lib/gitlab_quality/test_tooling/code_coverage/click_house/test_health_score_snapshotter'
 require_relative '../lib/gitlab_quality/test_tooling/code_coverage/coverage_data'
 require_relative '../lib/gitlab_quality/test_tooling/code_coverage/lcov_file'
 require_relative '../lib/gitlab_quality/test_tooling/code_coverage/artifacts'
@@ -177,6 +178,13 @@ clickhouse_data = {
 case mode
 when :aggregation
   GitlabQuality::TestTooling::CodeCoverage::ClickHouse::TestHealthRiskAggregator.new(**clickhouse_data).run
+  # Best-effort: the score snapshot is a non-critical add-on, so it must not
+  # fail the job after the aggregation has already succeeded.
+  begin
+    GitlabQuality::TestTooling::CodeCoverage::ClickHouse::TestHealthScoreSnapshotter.new(**clickhouse_data).run
+  rescue StandardError => e
+    warn "Warning: score snapshot failed (aggregation succeeded), continuing: #{e.message}"
+  end
 when :per_test
   GitlabQuality::TestTooling::CodeCoverage::PerTestCoverageExporter.new(
     coverage_glob: params[:per_test_coverage],

data/lib/gitlab_quality/test_tooling/code_coverage/click_house/test_health_risk_aggregation.sql

@@ -71,27 +71,44 @@ WITH
     GROUP BY test_file
     HAVING is_flaky
   ),
-  -- Jest quarantine comes from a flat file (quarantined_vue3_specs.txt) rather
-  -- than the issue-based RSpec ledger. It is populated into this helper table
-  -- by JestQuarantinedTestsTable before each nightly aggregation run.
-  jest_quarantine_status AS (
-    SELECT test_file
-    FROM code_coverage.jest_quarantined_tests_today FINAL
+  test_ownership AS (
+    -- Authoritative test-file ownership from test_metrics, which resolves
+    -- feature_category to group/stage/section from a current source. The
+    -- per-test coverage export's own CategoryOwners mapping is stale (it fetches
+    -- a dated stages.yml), so many rows arrive with a blank group; this is the
+    -- fallback. argMax pins the latest ownership per file over the risk window.
+    SELECT
+      file_path,
+      argMax(`group`, timestamp) AS owner_group,
+      argMax(stage, timestamp) AS owner_stage,
+      argMax(section, timestamp) AS owner_section
+    FROM test_metrics.test_results
+    WHERE timestamp >= now() - INTERVAL {risk_window}
+      AND `group` NOT IN ('', 'unknown')
+    GROUP BY file_path
   ),
   per_test_status AS (
     SELECT
-      tc.source_file,
-      tc.`group`,
-      tc.stage,
-      tc.section,
-      tc.total_lines,
-      tc.covered_lines,
-      (coalesce(qs.is_quarantined, FALSE) OR (jq.test_file IS NOT NULL)) AS is_quarantined,
+      tc.source_file AS source_file,
+      -- Fall back to test_metrics ownership when the coverage row has no group
+      -- (stale CategoryOwners mapping). Resolve all three off the same condition
+      -- so the team stays internally consistent. Alias the passthrough columns to
+      -- bare names too: the analyzer otherwise keeps the `tc.` qualifier once the
+      -- ownership join is in play, and per_file references them unqualified.
+      if(tc.`group` != '', tc.`group`, own.owner_group) AS `group`,
+      if(tc.`group` != '', tc.stage, own.owner_stage) AS stage,
+      if(tc.`group` != '', tc.section, own.owner_section) AS section,
+      tc.total_lines AS total_lines,
+      tc.covered_lines AS covered_lines,
+      -- Jest quarantine is a flat list (quarantined_vue3_specs.txt), loaded into
+      -- jest_quarantined_tests_today by JestQuarantinedTestsTable before each run.
+      (coalesce(qs.is_quarantined, FALSE)
+        OR tc.test_file IN (SELECT test_file FROM code_coverage.jest_quarantined_tests_today FINAL)) AS is_quarantined,
       coalesce(fs.is_flaky, FALSE) AS is_flaky
     FROM code_coverage.test_coverage_per_file tc FINAL
-    LEFT JOIN quarantine_status      qs ON qs.test_file = tc.test_file
-    LEFT JOIN jest_quarantine_status jq ON jq.test_file = tc.test_file
-    LEFT JOIN flaky_status           fs ON fs.test_file = tc.test_file
+    LEFT JOIN quarantine_status qs ON qs.test_file = tc.test_file
+    LEFT JOIN flaky_status      fs ON fs.test_file = tc.test_file
+    LEFT JOIN test_ownership own ON own.file_path = tc.test_file
     WHERE tc.timestamp >= now() - INTERVAL {coverage_window}
   ),
   per_file AS (

data/lib/gitlab_quality/test_tooling/code_coverage/click_house/test_health_score_snapshot.sql

@@ -0,0 +1,31 @@
+-- Freezes today's per-pillar Test Health Index scores into the history table.
+--
+-- Reads one pillar's score view (the same view the dashboard reads, so the
+-- snapshot and the dashboard cannot diverge) and tags each row with the
+-- snapshot date and the pillar. Run once per pillar by
+-- TestHealthScoreSnapshotter, right after the daily risk aggregation refreshes
+-- test_health_risk_per_group, which the view reads.
+--
+-- Idempotency: the target is a SharedReplacingMergeTree keyed by
+-- (snapshot_date, level, group, stage, section, pillar) with a version column,
+-- so re-running for the same snapshot_date upserts and the last run of the day
+-- wins on merge. No "final run of the day" detection is needed.
+--
+-- Parameter substitution uses {name} braces, replaced via gsub in
+-- TestHealthScoreSnapshotter#build_sql after each value passes a regex check:
+--   {snapshot_date} : the date stamp for this run, e.g. '2026-05-07'
+--   {pillar}        : quarantine | slow (names the view and tags the rows)
+INSERT INTO code_coverage.test_health_score_history
+  (snapshot_date, level, `group`, stage, section, pillar, score, rag_status, formula_version, factors)
+SELECT
+  toDate('{snapshot_date}') AS snapshot_date,
+  level,
+  `group`,
+  stage,
+  section,
+  '{pillar}' AS pillar,
+  score,
+  rag_status,
+  formula_version,
+  factors
+FROM code_coverage.test_health_scores_{pillar};

data/lib/gitlab_quality/test_tooling/code_coverage/click_house/test_health_score_snapshotter.rb

@@ -0,0 +1,95 @@
+# frozen_string_literal: true
+
+require_relative 'client'
+require_relative 'table'
+
+module GitlabQuality
+  module TestTooling
+    module CodeCoverage
+      module ClickHouse
+        # Freezes the day's per-pillar Test Health Index scores into
+        # `code_coverage.test_health_score_history`. Reads the per-pillar score
+        # views (the same views the dashboard reads, so the snapshot cannot
+        # diverge from the live score) and tags each row with the snapshot date
+        # and the pillar.
+        #
+        # Runs right after TestHealthRiskAggregator in the daily job: the views
+        # read `test_health_risk_per_group`, which the aggregator has just
+        # refreshed, so the freeze captures the current day. The producer never
+        # holds the score formula; it copies whatever the views compute.
+        class TestHealthScoreSnapshotter
+          include Client
+
+          SQL_FILE = File.expand_path('test_health_score_snapshot.sql', __dir__)
+
+          # One INSERT per pillar. The token names the view
+          # (test_health_scores_<pillar>) and tags the row's `pillar` column.
+          # Flaky is excluded until it is scored live.
+          PILLARS = %w[quarantine slow].freeze
+
+          DATE_PATTERN = /\A\d{4}-\d{2}-\d{2}\z/
+
+          def initialize(url:, database:, username: nil, password: nil, logger: nil)
+            @url = url
+            @database = database
+            @username = username
+            @password = password
+            @logger = logger || ::Logger.new($stdout, level: ::Logger::INFO)
+          end
+
+          # @param snapshot_date [Date, String] date stamp for this run; defaults to today.
+          # @return [void]
+          def run(snapshot_date: Date.today)
+            PILLARS.each do |pillar|
+              logger.info("#{LOG_PREFIX} Snapshotting #{pillar} scores snapshot_date=#{snapshot_date}")
+              client.query(build_sql(snapshot_date: snapshot_date, pillar: pillar), format: "TabSeparated")
+            end
+            inserted = fetch_row_count(snapshot_date)
+            logger.info("#{LOG_PREFIX} Score snapshot wrote #{inserted} rows for snapshot_date=#{snapshot_date}")
+          rescue StandardError => e
+            logger.error("#{LOG_PREFIX} Score snapshot failed for #{snapshot_date}: #{e.message}")
+            raise
+          end
+
+          private
+
+          attr_reader :url, :database, :username, :password, :logger
+
+          def build_sql(snapshot_date:, pillar:)
+            File.read(SQL_FILE)
+              .gsub('{snapshot_date}', validate_date(snapshot_date))
+              .gsub('{pillar}', validate_pillar(pillar))
+          end
+
+          # DateTime/Time `to_s` includes the time portion and is rejected.
+          def validate_date(value)
+            str = value.to_s
+            raise ArgumentError, "Invalid snapshot_date: #{value.inspect}" unless DATE_PATTERN.match?(str)
+
+            str
+          end
+
+          # The pillar names the view and is interpolated into the SQL, so
+          # restrict it to the known set rather than a looser pattern.
+          def validate_pillar(value)
+            raise ArgumentError, "Invalid pillar: #{value.inspect}" unless PILLARS.include?(value)
+
+            value
+          end
+
+          # Returns 'unknown' on any error so a transient count-query failure
+          # can't mask the success of the actual INSERTs.
+          def fetch_row_count(snapshot_date)
+            count_sql = "SELECT count() FROM code_coverage.test_health_score_history FINAL " \
+              "WHERE snapshot_date = toDate('#{validate_date(snapshot_date)}')"
+            result = client.query(count_sql, format: "JSONCompact")
+            result&.dig('data', 0, 0) || 'unknown'
+          rescue StandardError => e
+            logger.debug("#{LOG_PREFIX} Could not fetch post-snapshot row count: #{e.message}")
+            'unknown'
+          end
+        end
+      end
+    end
+  end
+end

data/lib/gitlab_quality/test_tooling/code_coverage/per_test_coverage_exporter.rb

@@ -52,7 +52,10 @@ module GitlabQuality
 
           insert(coverage_files)
           jest_quarantine_error = ingest_jest_quarantine_best_effort
-          aggregate unless skip_aggregation
+          unless skip_aggregation
+            aggregate
+            snapshot_scores
+          end
 
           # Re-raise only after aggregation has run, so a jest quarantine ingest
           # failure can't silently stop the table-wide rollup while still
@@ -112,9 +115,24 @@ module GitlabQuality
           ClickHouse::TestHealthRiskAggregator.new(**clickhouse_credentials).run
         end
 
+        # Freezes the day's per-pillar scores into test_health_score_history,
+        # right after the aggregation the score views read. Best-effort: the
+        # snapshot is a non-critical add-on, so a failure (for example before
+        # the score views exist) must not fail the daily job once the
+        # aggregation has succeeded. A missed freeze is recoverable because the
+        # dashboard reads the live view for today.
+        def snapshot_scores
+          ClickHouse::TestHealthScoreSnapshotter.new(**clickhouse_credentials).run
+        rescue StandardError => e
+          logger.error(
+            "#{ClickHouse::LOG_PREFIX} Score snapshot failed; the aggregation still " \
+              "succeeded, continuing: #{e.message}"
+          )
+        end
+
         # `tests_to_categories` comes from the test report JSON; without it, rows
-        # carry blank categories. `CategoryOwners` fetches stages.yml over HTTP,
-        # so only build it when there are reports to enrich.
+        # carry blank categories. The category-to-team map is a ClickHouse read,
+        # so only build both when there are reports to enrich.
         def resolve_categories
           return [{}, {}] unless test_reports
 
@@ -123,7 +141,19 @@ module GitlabQuality
           report_files = Dir.glob(patterns)
           return [{}, {}] if report_files.empty?
 
-          [tests_to_categories_from(report_files), CategoryOwners.new.feature_categories_to_teams]
+          [tests_to_categories_from(report_files), feature_categories_to_teams]
+        end
+
+        # Category to group/stage/section, read from `code_coverage.category_owners`
+        # (populated by sync-category-owners). This is the source the full
+        # coverage export already uses, and it retains categories the live
+        # stages.yml fetch has since dropped (global_search and the Duo/AI groups),
+        # so more rows resolve to a real group instead of blank. `owner_records`
+        # gives string-keyed values; PerTestCoverageData reads symbol keys.
+        def feature_categories_to_teams
+          ClickHouse::CategoryOwnersTable.new(**clickhouse_credentials).owner_records.transform_values do |owner|
+            { group: owner["group"], stage: owner["stage"], section: owner["section"] }
+          end
         end
 
         def tests_to_categories_from(report_files)

data/lib/gitlab_quality/test_tooling/version.rb

@@ -2,6 +2,6 @@
 
 module GitlabQuality
   module TestTooling
-    VERSION = "3.18.3"
+    VERSION = "3.19.1"
   end
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: gitlab_quality-test_tooling
 version: !ruby/object:Gem::Version
-  version: 3.18.3
+  version: 3.19.1
 platform: ruby
 authors:
 - GitLab Quality
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2026-06-24 00:00:00.000000000 Z
+date: 2026-07-02 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: climate_control
@@ -499,6 +499,8 @@ files:
 - lib/gitlab_quality/test_tooling/code_coverage/click_house/test_file_mappings_table.rb
 - lib/gitlab_quality/test_tooling/code_coverage/click_house/test_health_risk_aggregation.sql
 - lib/gitlab_quality/test_tooling/code_coverage/click_house/test_health_risk_aggregator.rb
+- lib/gitlab_quality/test_tooling/code_coverage/click_house/test_health_score_snapshot.sql
+- lib/gitlab_quality/test_tooling/code_coverage/click_house/test_health_score_snapshotter.rb
 - lib/gitlab_quality/test_tooling/code_coverage/coverage_data.rb
 - lib/gitlab_quality/test_tooling/code_coverage/lcov_file.rb
 - lib/gitlab_quality/test_tooling/code_coverage/per_test_coverage_data.rb