@oneuptime/common 11.5.3 → 11.5.5

package/build/dist/Server/Services/IncidentService.js

@@ -10,6 +10,9 @@ var __metadata = (this && this.__metadata) || function (k, v) {
 import DatabaseConfig from "../DatabaseConfig";
 import QueryHelper from "../Types/Database/QueryHelper";
 import DatabaseService from "./DatabaseService";
+import AIRunService from "./AIRunService";
+import AIRunType from "../../Types/AI/AIRunType";
+import AIRunStatus from "../../Types/AI/AIRunStatus";
 import IncidentOwnerTeamService from "./IncidentOwnerTeamService";
 import IncidentOwnerUserService from "./IncidentOwnerUserService";
 import IncidentStateService from "./IncidentStateService";
@@ -73,7 +76,7 @@ import OnCallDutyPolicy from "../../Models/DatabaseModels/OnCallDutyPolicy";
 import ProjectService from "./ProjectService";
 import IncidentTemplateService from "./IncidentTemplateService";
 import AIService from "./AIService";
-import SentinelIncidentInvestigationRunner from "../Utils/AI/Sentinel/IncidentInvestigationRunner";
+import AIIncidentInvestigationRunner from "../Utils/AI/SRE/IncidentInvestigationRunner";
 import IncidentAIContextBuilder from "../Utils/AI/IncidentAIContextBuilder";
 export class Service extends DatabaseService {
     constructor() {
@@ -940,21 +943,21 @@ export class Service extends DatabaseService {
             .then(async () => {
             var _a, _b;
             /*
-             * Sentinel (AI SRE): automatically investigate the new incident and post
+             * AI (AI SRE): automatically investigate the new incident and post
              * a cited root cause analysis to the incident timeline + Slack/Teams.
              * Runs last so the workspace channels already exist, and is gated per
              * project (opt-in) + requires a configured LLM provider. Read-only.
              */
             try {
                 if (createdItem.projectId && createdItem.id) {
-                    await SentinelIncidentInvestigationRunner.investigateNewIncident({
+                    await AIIncidentInvestigationRunner.investigateNewIncident({
                         incidentId: createdItem.id,
                         projectId: createdItem.projectId,
                     });
                 }
             }
             catch (error) {
-                logger.error(`Sentinel incident investigation failed in IncidentService.onCreateSuccess: ${error}`, {
+                logger.error(`AI incident investigation failed in IncidentService.onCreateSuccess: ${error}`, {
                     projectId: (_a = createdItem.projectId) === null || _a === void 0 ? void 0 : _a.toString(),
                     incidentId: (_b = createdItem.id) === null || _b === void 0 ? void 0 : _b.toString(),
                 });
@@ -2070,12 +2073,21 @@ ${incidentSeverity.name}
         }
         return Service.DEFAULT_METRIC_RETENTION_DAYS;
     }
-    async refreshIncidentMetrics(data) {
-        var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l, _m, _o, _p, _q, _r, _s, _t, _u;
+    /*
+     * Load the incident + its owners and build the attribute set shared by
+     * every incident metric (all values strings for ClickHouse
+     * Map(String, String) storage; arrays joined comma-separated). Factored
+     * out of refreshIncidentMetrics so one-off metric writers — e.g. the
+     * AI time-to-rca metric recorded when an investigation posts its
+     * analysis — record the exact same attribute shape.
+     */
+    async getIncidentMetricContext(data) {
+        var _a, _b, _c, _d, _e, _f;
         const incident = await this.findOneById({
             id: data.incidentId,
             select: {
                 projectId: true,
+                createdAt: true,
                 declaredAt: true,
                 postmortemPostedAt: true,
                 monitors: {
@@ -2164,6 +2176,115 @@ ${incidentSeverity.name}
             .filter((name) => {
             return Boolean(name);
         });
+        /*
+         * common attributes shared by all incident metrics
+         * All values must be strings for ClickHouse Map(String, String) storage.
+         * Arrays are joined as comma-separated strings.
+         */
+        const baseMetricAttributes = {
+            incidentId: data.incidentId.toString(),
+            projectId: incident.projectId.toString(),
+            monitorIds: (((_a = incident.monitors) === null || _a === void 0 ? void 0 : _a.map((monitor) => {
+                var _a;
+                return (_a = monitor._id) === null || _a === void 0 ? void 0 : _a.toString();
+            }).filter(Boolean)) || []).join(", "),
+            monitorNames: (((_b = incident.monitors) === null || _b === void 0 ? void 0 : _b.map((monitor) => {
+                var _a;
+                return (_a = monitor.name) === null || _a === void 0 ? void 0 : _a.toString();
+            }).filter(Boolean)) || []).join(", "),
+            incidentSeverityId: (_d = (_c = incident.incidentSeverity) === null || _c === void 0 ? void 0 : _c._id) === null || _d === void 0 ? void 0 : _d.toString(),
+            incidentSeverityName: (_f = (_e = incident.incidentSeverity) === null || _e === void 0 ? void 0 : _e.name) === null || _f === void 0 ? void 0 : _f.toString(),
+            ownerUserIds: ownerUserIds.join(", "),
+            ownerUserNames: ownerUserNames.join(", "),
+            ownerTeamIds: ownerTeamIds.join(", "),
+            ownerTeamNames: ownerTeamNames.join(", "),
+        };
+        return { incident, baseMetricAttributes };
+    }
+    /*
+     * AI measurement layer: record how long the incident waited for its
+     * AI root-cause analysis — seconds from incident creation to now, written
+     * once from IncidentInvestigationRunner.postAnalysis. Deliberately NOT
+     * part of refreshIncidentMetrics: the refresh's replace-list excludes
+     * this metric name, so refreshes never tombstone it.
+     */
+    async recordTimeToRootCausePostedMetric(data) {
+        const { incident, baseMetricAttributes, } = await this.getIncidentMetricContext({ incidentId: data.incidentId });
+        const now = OneUptimeDate.getCurrentDate();
+        const incidentCreatedAt = incident.createdAt || incident.declaredAt || now;
+        const metricRetentionDays = await this.getMetricRetentionDays();
+        const retentionDate = OneUptimeDate.addRemoveDays(OneUptimeDate.getCurrentDate(), metricRetentionDays);
+        const timeToRcaMetric = new MutableMetric();
+        timeToRcaMetric.projectId = incident.projectId;
+        timeToRcaMetric.primaryEntityId = incident.id;
+        timeToRcaMetric.primaryEntityType = ServiceType.Incident;
+        timeToRcaMetric.name = IncidentMetricType.TimeToRootCausePosted;
+        timeToRcaMetric.metricPointId = IncidentMetricType.TimeToRootCausePosted;
+        timeToRcaMetric.value = OneUptimeDate.getDifferenceInSeconds(now, incidentCreatedAt);
+        timeToRcaMetric.attributes = Object.assign(Object.assign({}, baseMetricAttributes), { 
+            // Every time-to-rca point is AI-posted by construction.
+            aiInvestigated: "true" });
+        timeToRcaMetric.attributeKeys = TelemetryUtil.getAttributeKeys(timeToRcaMetric.attributes);
+        timeToRcaMetric.time = now;
+        timeToRcaMetric.timeUnixNano = OneUptimeDate.toUnixNano(now);
+        timeToRcaMetric.metricPointType = MetricPointType.Sum;
+        timeToRcaMetric.retentionDate = retentionDate;
+        await MutableMetricService.createMutableMetrics({
+            metrics: [timeToRcaMetric],
+        });
+        // Register the metric type so it shows up in the type catalog.
+        const metricType = new MetricType();
+        metricType.name = IncidentMetricType.TimeToRootCausePosted;
+        metricType.description =
+            "Time from incident creation to the AI investigation's posted root-cause analysis";
+        metricType.unit = "seconds";
+        TelemetryUtil.indexMetricNameServiceNameMap({
+            metricNameServiceNameMap: {
+                [IncidentMetricType.TimeToRootCausePosted]: metricType,
+            },
+            projectId: incident.projectId,
+        }).catch((err) => {
+            var _a, _b;
+            logger.error(err, {
+                projectId: (_a = incident.projectId) === null || _a === void 0 ? void 0 : _a.toString(),
+                incidentId: (_b = incident.id) === null || _b === void 0 ? void 0 : _b.toString(),
+            });
+        });
+    }
+    async refreshIncidentMetrics(data) {
+        var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l, _m, _o, _p, _q;
+        const { incident, baseMetricAttributes, } = await this.getIncidentMetricContext({ incidentId: data.incidentId });
+        // getIncidentMetricContext guarantees this; re-checked for TS narrowing.
+        if (!incident.projectId) {
+            throw new BadDataException("Incident Project ID not found");
+        }
+        /*
+         * aiInvestigated dimension for MTTA/MTTR (AI measurement layer):
+         * did a completed AI investigation run for this incident? One
+         * indexed countBy per metric refresh (triggeredByIncidentId is indexed)
+         * — acceptable at refresh frequency. Failure must never break metric
+         * recording, so this is best-effort false.
+         */
+        let aiInvestigated = false;
+        try {
+            const completedInvestigationCount = await AIRunService.countBy({
+                query: {
+                    runType: AIRunType.Investigation,
+                    status: AIRunStatus.Completed,
+                    triggeredByIncidentId: data.incidentId,
+                },
+                props: {
+                    isRoot: true,
+                },
+            });
+            aiInvestigated = completedInvestigationCount.toNumber() > 0;
+        }
+        catch (err) {
+            logger.error(err, {
+                projectId: (_a = incident.projectId) === null || _a === void 0 ? void 0 : _a.toString(),
+                incidentId: (_b = incident.id) === null || _b === void 0 ? void 0 : _b.toString(),
+            });
+        }
         // get incident state timeline
         const incidentStateTimelines = await IncidentStateTimelineService.findBy({
             query: {
@@ -2208,8 +2329,8 @@ ${incidentSeverity.name}
         }
         catch (err) {
             logger.error(err, {
-                projectId: (_a = incident.projectId) === null || _a === void 0 ? void 0 : _a.toString(),
-                incidentId: (_b = incident.id) === null || _b === void 0 ? void 0 : _b.toString(),
+                projectId: (_c = incident.projectId) === null || _c === void 0 ? void 0 : _c.toString(),
+                incidentId: (_d = incident.id) === null || _d === void 0 ? void 0 : _d.toString(),
             });
         }
         try {
@@ -2222,29 +2343,6 @@ ${incidentSeverity.name}
                 incident.createdAt ||
                 OneUptimeDate.getCurrentDate();
             const metricTypesMap = {};
-            /*
-             * common attributes shared by all incident metrics
-             * All values must be strings for ClickHouse Map(String, String) storage.
-             * Arrays are joined as comma-separated strings.
-             */
-            const baseMetricAttributes = {
-                incidentId: data.incidentId.toString(),
-                projectId: incident.projectId.toString(),
-                monitorIds: (((_c = incident.monitors) === null || _c === void 0 ? void 0 : _c.map((monitor) => {
-                    var _a;
-                    return (_a = monitor._id) === null || _a === void 0 ? void 0 : _a.toString();
-                }).filter(Boolean)) || []).join(", "),
-                monitorNames: (((_d = incident.monitors) === null || _d === void 0 ? void 0 : _d.map((monitor) => {
-                    var _a;
-                    return (_a = monitor.name) === null || _a === void 0 ? void 0 : _a.toString();
-                }).filter(Boolean)) || []).join(", "),
-                incidentSeverityId: (_f = (_e = incident.incidentSeverity) === null || _e === void 0 ? void 0 : _e._id) === null || _f === void 0 ? void 0 : _f.toString(),
-                incidentSeverityName: (_h = (_g = incident.incidentSeverity) === null || _g === void 0 ? void 0 : _g.name) === null || _h === void 0 ? void 0 : _h.toString(),
-                ownerUserIds: ownerUserIds.join(", "),
-                ownerUserNames: ownerUserNames.join(", "),
-                ownerTeamIds: ownerTeamIds.join(", "),
-                ownerTeamNames: ownerTeamNames.join(", "),
-            };
             const incidentCountMetric = new MutableMetric();
             incidentCountMetric.projectId = incident.projectId;
             incidentCountMetric.primaryEntityId = incident.id;
@@ -2292,7 +2390,8 @@ ${incidentSeverity.name}
                         IncidentMetricType.TimeToAcknowledge;
                     timeToAcknowledgeMetric.value = OneUptimeDate.getDifferenceInSeconds((ackIncidentStateTimeline === null || ackIncidentStateTimeline === void 0 ? void 0 : ackIncidentStateTimeline.startsAt) ||
                         OneUptimeDate.getCurrentDate(), incidentStartsAt);
-                    timeToAcknowledgeMetric.attributes = Object.assign({}, baseMetricAttributes);
+                    // aiInvestigated: the MTTA with/without-AI dimension.
+                    timeToAcknowledgeMetric.attributes = Object.assign(Object.assign({}, baseMetricAttributes), { aiInvestigated: aiInvestigated.toString() });
                     timeToAcknowledgeMetric.attributeKeys =
                         TelemetryUtil.getAttributeKeys(timeToAcknowledgeMetric.attributes);
                     timeToAcknowledgeMetric.time =
@@ -2330,7 +2429,8 @@ ${incidentSeverity.name}
                 timeToResolveMetric.metricPointId = IncidentMetricType.TimeToResolve;
                 timeToResolveMetric.value = OneUptimeDate.getDifferenceInSeconds((resolvedIncidentStateTimeline === null || resolvedIncidentStateTimeline === void 0 ? void 0 : resolvedIncidentStateTimeline.startsAt) ||
                     OneUptimeDate.getCurrentDate(), incidentStartsAt);
-                timeToResolveMetric.attributes = Object.assign({}, baseMetricAttributes);
+                // aiInvestigated: the MTTR with/without-AI dimension.
+                timeToResolveMetric.attributes = Object.assign(Object.assign({}, baseMetricAttributes), { aiInvestigated: aiInvestigated.toString() });
                 timeToResolveMetric.attributeKeys = TelemetryUtil.getAttributeKeys(timeToResolveMetric.attributes);
                 timeToResolveMetric.time =
                     (resolvedIncidentStateTimeline === null || resolvedIncidentStateTimeline === void 0 ? void 0 : resolvedIncidentStateTimeline.startsAt) ||
@@ -2397,7 +2497,7 @@ ${incidentSeverity.name}
                 if (!timeline.startsAt || !timeline.endsAt) {
                     continue;
                 }
-                const stateName = ((_k = (_j = timeline.incidentState) === null || _j === void 0 ? void 0 : _j.name) === null || _k === void 0 ? void 0 : _k.toString()) || "Unknown";
+                const stateName = ((_f = (_e = timeline.incidentState) === null || _e === void 0 ? void 0 : _e.name) === null || _f === void 0 ? void 0 : _f.toString()) || "Unknown";
                 const timeInStateMetric = new MutableMetric();
                 timeInStateMetric.projectId = incident.projectId;
                 timeInStateMetric.primaryEntityId = incident.id;
@@ -2405,7 +2505,7 @@ ${incidentSeverity.name}
                 timeInStateMetric.name = IncidentMetricType.TimeInState;
                 timeInStateMetric.metricPointId = `time-in-state:${timeline.id.toString()}`;
                 timeInStateMetric.value = OneUptimeDate.getDifferenceInSeconds(timeline.endsAt, timeline.startsAt);
-                timeInStateMetric.attributes = Object.assign(Object.assign({}, baseMetricAttributes), { incidentStateName: stateName, incidentStateId: (_l = timeline.incidentStateId) === null || _l === void 0 ? void 0 : _l.toString(), isCreatedState: ((_o = (_m = timeline.incidentState) === null || _m === void 0 ? void 0 : _m.isCreatedState) === null || _o === void 0 ? void 0 : _o.toString()) || "false", isAcknowledgedState: ((_q = (_p = timeline.incidentState) === null || _p === void 0 ? void 0 : _p.isAcknowledgedState) === null || _q === void 0 ? void 0 : _q.toString()) || "false", isResolvedState: ((_s = (_r = timeline.incidentState) === null || _r === void 0 ? void 0 : _r.isResolvedState) === null || _s === void 0 ? void 0 : _s.toString()) || "false" });
+                timeInStateMetric.attributes = Object.assign(Object.assign({}, baseMetricAttributes), { incidentStateName: stateName, incidentStateId: (_g = timeline.incidentStateId) === null || _g === void 0 ? void 0 : _g.toString(), isCreatedState: ((_j = (_h = timeline.incidentState) === null || _h === void 0 ? void 0 : _h.isCreatedState) === null || _j === void 0 ? void 0 : _j.toString()) || "false", isAcknowledgedState: ((_l = (_k = timeline.incidentState) === null || _k === void 0 ? void 0 : _k.isAcknowledgedState) === null || _l === void 0 ? void 0 : _l.toString()) || "false", isResolvedState: ((_o = (_m = timeline.incidentState) === null || _m === void 0 ? void 0 : _m.isResolvedState) === null || _o === void 0 ? void 0 : _o.toString()) || "false" });
                 timeInStateMetric.attributeKeys = TelemetryUtil.getAttributeKeys(timeInStateMetric.attributes);
                 timeInStateMetric.time = timeline.startsAt;
                 timeInStateMetric.timeUnixNano = OneUptimeDate.toUnixNano(timeInStateMetric.time);
@@ -2457,8 +2557,8 @@ ${incidentSeverity.name}
                 }
                 catch (err) {
                     logger.error(err, {
-                        projectId: (_t = incident.projectId) === null || _t === void 0 ? void 0 : _t.toString(),
-                        incidentId: (_u = incident.id) === null || _u === void 0 ? void 0 : _u.toString(),
+                        projectId: (_p = incident.projectId) === null || _p === void 0 ? void 0 : _p.toString(),
+                        incidentId: (_q = incident.id) === null || _q === void 0 ? void 0 : _q.toString(),
                     });
                 }
             }
@@ -2760,6 +2860,18 @@ __decorate([
     __metadata("design:paramtypes", [Object]),
     __metadata("design:returntype", Promise)
 ], Service.prototype, "changeIncidentState", null);
+__decorate([
+    CaptureSpan(),
+    __metadata("design:type", Function),
+    __metadata("design:paramtypes", [Object]),
+    __metadata("design:returntype", Promise)
+], Service.prototype, "getIncidentMetricContext", null);
+__decorate([
+    CaptureSpan(),
+    __metadata("design:type", Function),
+    __metadata("design:paramtypes", [Object]),
+    __metadata("design:returntype", Promise)
+], Service.prototype, "recordTimeToRootCausePostedMetric", null);
 __decorate([
     CaptureSpan(),
     __metadata("design:type", Function),