mirror of
https://github.com/SigNoz/signoz.git
synced 2026-05-27 04:10:28 +01:00
6cf22e98dd
* add maintenanceMuteStage to move planned maintenance to alertmanager Rules previously skipped rule.Eval() entirely during maintenance windows. This change moves suppression to MaintenanceMuter, injected as a Stage in the alertmanager notification pipeline. Now rules always evaluate and everys suppression is handled by alertmanager. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * refactor: wrap routing pipeline once instead of per-route injection Replace the per-route-entry loop with a single MultiStage wrap so maintenance suppression runs once per dispatch group before routing. * refactor: move maintenance mute stage into custom pipelineBuilder Copy notify.PipelineBuilder locally so we can inject mms between the silence stage and the receiver stage (GossipSettle → Inhibit → TimeActive → TimeMute → Silence → mms → Receiver), matching the correct suppression order the team requires. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * chore: add license header to pipeline_builder.go Copied code originates from Apache-2.0 licensed Prometheus Alertmanager; add dual copyright + SPDX identifier following the repo's convention. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * chore: replace SPDX tag with full Apache 2.0 license boilerplate The full license text is unambiguously compliant with Apache 2.0 Section 4(a), which requires giving recipients "a copy of this License". Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * refactor: pass MaintenanceMuter directly to pipelineBuilder Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * refactor: remove dead orgID param from task constructors Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * rename buildReceiverStage -> createReceiverStage * refactor: replace maintenanceMuteStage with notify.NewMuteStage MaintenanceMuter already satisfies types.Muter, and pipelineBuilder has its own pb.metrics, so the hand-rolled maintenanceMuteStage wrapper is redundant. Use notify.NewMuteStage(pb.muter, pb.metrics) directly. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * refactor: hoist MuteStage construction out of the receiver loop MuteStage holds no per-receiver state, so one instance shared across all receivers is sufficient — matching how is/ss are handled upstream. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * refactor: always initialize maintenanceStore; remove nil guards Tests now use a real sqlrulestore-backed MaintenanceMuter instead of passing nil. With nil no longer a valid input, remove the nil guards in server.go and pipeline_builder.go. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * refactor: move MaintenanceMuter to Server and pass it to pipelineBuilder.New - Remove muter from pipelineBuilder struct and newPipelineBuilder(); pass it as a parameter to New() instead, consistent with inhibitor/silencer - Store muter on Server so GetAlerts can call Mutes() alongside the inhibitor and silencer, ensuring maintenance-suppressed alerts show the correct muted status in API responses Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com> * remove redundant MemMarker wrapper * feat: surface maintenance-suppressed alerts via mutedBy in GetAlerts Alerts suppressed by an active maintenance window were being correctly muted in the notification pipeline but appeared as state=active in the v2 GetAlerts response, since MaintenanceMuter.Mutes had no marker side-effect (unlike inhibitor/silencer). Add MaintenanceMuter.MutedBy returning the matching window IDs, and plumb a mutedByFunc callback through NewGettableAlertsFromAlertProvider into AlertToOpenAPIAlert. The upstream v2 API forces state=suppressed when mutedBy is non-empty, so the frontend's existing state-based rendering picks it up without further changes. Use the dedicated mutedBy field rather than SilencedBy to avoid violating the "complete set of silence IDs" contract that anything querying silences by ID would rely on. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com> * code cleanup * refactor: move maintenance (planned downtime) to alertmanager packages Types move from pkg/types/ruletypes/ to pkg/types/alertmanagertypes/: - maintenance.go, recurrence.go, schedule.go (+ tests) Store impl moves from pkg/ruler/rulestore/sqlrulestore/ to pkg/alertmanager/alertmanagerstore/sqlalertmanagerstore/. Maintenance windows mute alerts, so they belong with alertmanager rather than the rule types. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * test: add unit tests for MaintenanceMuter Covers Mutes/MutedBy semantics (empty label, rule match, empty-RuleIDs matches-all, future windows, multi-window) and the result cache (single-fetch within TTL, stale-cache fallback on store error, re-fetch after expiry, concurrency safety). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Update schema changes * Re-add marker * fix NewMaintenanceStore in tests * Go lint fixes * test: use mockery-generated mock for MaintenanceStore in muter tests Replace hand-written fakeMaintenanceStore with a mockery-generated MockMaintenanceStore, consistent with the alertmanagertest pattern. Also adds MaintenanceStore to .mockery.yml so the mock stays in sync. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * chore: regenerate mocks via make gen-mocks Picks up new MockHandler for the Handler interface in pkg/alertmanager and regenerates MockMaintenanceStore with canonical mockery formatting. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * cleanup test * test: add e2e muting tests for maintenance window behaviour * Add label expression support to planned downtime Alert instances can now be scoped by label expression (e.g. env == "prod"), scoping suppression below the rule level. A window with no rule IDs and a label expression silences any alert whose labels match, regardless of which rule fired it; when rule IDs are also present, the expression is evaluated only within the matched rules. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Remove redundant || undefined from labelExpression assignment * Move label expression evaluation into ShouldSkip ShouldSkip now owns all three suppression checks in sequence: rule ID match → schedule active → label expression. IsActive passes nil labels so the expression check is skipped (no instance labels available for UI status). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * remove redundant LabelSet->map conversion * implement Down migration to drop label_expression column * fix lset type and update openapi spec * fix(tests): resolve envprovider env isolation, factory name length, and ShouldSkip signature Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * remove unused function `evaluateLabelExpression` * chore: rename label expression to scope * test(maintenance): add tests for scope label expression filtering Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * test(e2e): verify scope-based maintenance muting in alertmanager flow Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Use `AND` instead of `&&` Co-authored-by: Srikanth Chekuri <srikanth.chekuri92@gmail.com> * fix(maintenance): consolidate label-set-to-env conversion to avoid expr panic Move ConvertLabelSetToEnv to alertmanagertypes so both the maintenance scope evaluator and the route-policy evaluator share one implementation. Dotted label keys (e.g. kubernetes.node) are expanded into nested maps, preventing the expr-lang panic that occurs when one key is a prefix of another. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * chore(planned-downtime): use clickable learn more link in scope tooltip * chore(sqlmigration): renumber add_scope_to_planned_maintenance to 086 078 collided with add_sa_managed_role_txn; bumped to the next free number and reordered registration after add_source_to_dashboard (085). * refactor(maintenance): extract scope expression eval and surface errors - Move ConvertLabelSetToEnv and EvalScopeExpression into expression.go with companion tests in expression_test.go. - EvalScopeExpression now returns (bool, error) instead of swallowing compile/run failures and non-bool outputs; ShouldSkip logs the error via slog.Default() and falls back to not suppressing (safety-first). - Update test fixtures to the SQL-style operator form (`=`, AND, OR) matching the placeholder and reviewer suggestions. * chore: use `=` instead of `==` in expressions * fix(maintenance): satisfy forbidigo/sloglint in scope eval - Replace fmt.Errorf with pkg/errors Wrapf/Newf using a new ErrCodeInvalidScopeExpression code. - Use slog ErrorContext (with context.Background()) instead of Error to satisfy sloglint. * perf(maintenance): fold prefix-conflict detection into ConvertLabelSetToEnv ConvertLabelSetToEnv now returns (env, conflict). The rulebased provider drops its O(n^2) pre-scan and logs based on the flag, restoring the previous O(n*d) cost while keeping the shared helper. * chore: add docs URL for invalid scope * refactor: don't log in types package * remove down migration --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> Co-authored-by: Srikanth Chekuri <srikanth.chekuri92@gmail.com>