feat(otel): Add support for storing run spans and log data in Clickhouse

.vscode/settings.json

@@ -6,5 +6,6 @@
     "**/node_modules/**": true,
     "packages/cli-v3/e2e": true
   },
-  "vitest.disableWorkspaceWarning": true
+  "vitest.disableWorkspaceWarning": true,
+  "typescript.experimental.useTsgo": false
 }

apps/webapp/app/env.server.ts

@@ -1109,6 +1109,21 @@ const EnvironmentSchema = z
     CLICKHOUSE_LOG_LEVEL: z.enum(["log", "error", "warn", "info", "debug"]).default("info"),
     CLICKHOUSE_COMPRESSION_REQUEST: z.string().default("1"),
 
+    EVENTS_CLICKHOUSE_URL: z
+      .string()
+      .optional()
+      .transform((v) => v ?? process.env.CLICKHOUSE_URL),
+    EVENTS_CLICKHOUSE_KEEP_ALIVE_ENABLED: z.string().default("1"),
+    EVENTS_CLICKHOUSE_KEEP_ALIVE_IDLE_SOCKET_TTL_MS: z.coerce.number().int().optional(),
+    EVENTS_CLICKHOUSE_MAX_OPEN_CONNECTIONS: z.coerce.number().int().default(10),
+    EVENTS_CLICKHOUSE_LOG_LEVEL: z.enum(["log", "error", "warn", "info", "debug"]).default("info"),
+    EVENTS_CLICKHOUSE_COMPRESSION_REQUEST: z.string().default("1"),
+    EVENTS_CLICKHOUSE_BATCH_SIZE: z.coerce.number().int().default(1000),
+    EVENTS_CLICKHOUSE_FLUSH_INTERVAL_MS: z.coerce.number().int().default(1000),
+    EVENT_REPOSITORY_CLICKHOUSE_ROLLOUT_PERCENT: z.coerce.number().optional(),
+    EVENT_REPOSITORY_DEFAULT_STORE: z.enum(["postgres", "clickhouse"]).default("postgres"),
+    EVENTS_CLICKHOUSE_MAX_TRACE_SUMMARY_VIEW_COUNT: z.coerce.number().int().default(250_000),
+
     // Bootstrap
     TRIGGER_BOOTSTRAP_ENABLED: z.string().default("0"),
     TRIGGER_BOOTSTRAP_WORKER_GROUP_NAME: z.string().optional(),

apps/webapp/app/presenters/v3/ApiRetrieveRunPresenter.server.ts

@@ -75,6 +75,7 @@ export class ApiRetrieveRunPresenter {
       },
       select: {
         ...commonRunSelect,
+        traceId: true,
         payload: true,
         payloadType: true,
         output: true,

apps/webapp/app/presenters/v3/RunPresenter.server.ts

@@ -3,7 +3,8 @@ import { createTreeFromFlatItems, flattenTree } from "~/components/primitives/Tr
 import { prisma, type PrismaClient } from "~/db.server";
 import { createTimelineSpanEventsFromSpanEvents } from "~/utils/timelineSpanEvents";
 import { getUsername } from "~/utils/username";
-import { eventRepository } from "~/v3/eventRepository.server";
+import { resolveEventRepositoryForStore } from "~/v3/eventRepository/index.server";
+import { SpanSummary } from "~/v3/eventRepository/eventRepository.types";
 import { getTaskEventStoreTableForRun } from "~/v3/taskEventStore.server";
 import { isFinalRunStatus } from "~/v3/taskStatus";
 
@@ -45,9 +46,11 @@ export class RunPresenter {
         id: true,
         createdAt: true,
         taskEventStore: true,
+        taskIdentifier: true,
         number: true,
         traceId: true,
         spanId: true,
+        parentSpanId: true,
         friendlyId: true,
         status: true,
         startedAt: true,
@@ -140,18 +143,51 @@ export class RunPresenter {
       };
     }
 
+    const eventRepository = resolveEventRepositoryForStore(run.taskEventStore);
+
     // get the events
-    const traceSummary = await eventRepository.getTraceSummary(
+    let traceSummary = await eventRepository.getTraceSummary(
       getTaskEventStoreTableForRun(run),
+      run.runtimeEnvironment.id,
       run.traceId,
       run.rootTaskRun?.createdAt ?? run.createdAt,
       run.completedAt ?? undefined,
       { includeDebugLogs: showDebug }
     );
+
     if (!traceSummary) {
-      return {
-        run: runData,
-        trace: undefined,
+      const spanSummary: SpanSummary = {
+        id: run.spanId,
+        parentId: run.parentSpanId ?? undefined,
+        runId: run.friendlyId,
+        data: {
+          message: run.taskIdentifier,
+          style: { icon: "task", variant: "primary" },
+          events: [],
+          startTime: run.createdAt,
+          duration: 0,
+          isError:
+            run.status === "COMPLETED_WITH_ERRORS" ||
+            run.status === "CRASHED" ||
+            run.status === "EXPIRED" ||
+            run.status === "SYSTEM_FAILURE" ||
+            run.status === "TIMED_OUT",
+          isPartial:
+            run.status === "DELAYED" ||
+            run.status === "PENDING" ||
+            run.status === "PAUSED" ||
+            run.status === "RETRYING_AFTER_FAILURE" ||
+            run.status === "DEQUEUED" ||
+            run.status === "EXECUTING",
+          isCancelled: run.status === "CANCELED",
+          isDebug: false,
+          level: "TRACE",
+        },
+      };
+
+      traceSummary = {
+        rootSpan: spanSummary,
+        spans: [spanSummary],
       };
     }
 
@@ -220,6 +256,7 @@ export class RunPresenter {
         queuedDuration: run.startedAt
           ? millisecondsToNanoseconds(run.startedAt.getTime() - run.createdAt.getTime())
           : undefined,
+        overridesBySpanId: traceSummary.overridesBySpanId,
       },
     };
   }

apps/webapp/app/presenters/v3/RunStreamPresenter.server.ts

@@ -3,7 +3,7 @@ import { eventStream } from "remix-utils/sse/server";
 import { PrismaClient, prisma } from "~/db.server";
 import { logger } from "~/services/logger.server";
 import { throttle } from "~/utils/throttle";
-import { eventRepository } from "~/v3/eventRepository.server";
+import { tracePubSub } from "~/v3/services/tracePubSub.server";
 
 const pingInterval = 1000;
 
@@ -41,7 +41,7 @@ export class RunStreamPresenter {
 
     let pinger: NodeJS.Timeout | undefined = undefined;
 
-    const { unsubscribe, eventEmitter } = await eventRepository.subscribeToTrace(run.traceId);
+    const { unsubscribe, eventEmitter } = await tracePubSub.subscribeToTrace(run.traceId);
 
     return eventStream(request.signal, (send, close) => {
       const safeSend = (args: { event?: string; data: string }) => {

apps/webapp/app/presenters/v3/SpanPresenter.server.ts

@@ -10,21 +10,23 @@ import {
 import { AttemptId, getMaxDuration, parseTraceparent } from "@trigger.dev/core/v3/isomorphic";
 import { RUNNING_STATUSES } from "~/components/runs/v3/TaskRunStatus";
 import { logger } from "~/services/logger.server";
-import { eventRepository, rehydrateAttribute } from "~/v3/eventRepository.server";
+import { rehydrateAttribute } from "~/v3/eventRepository/eventRepository.server";
 import { machinePresetFromRun } from "~/v3/machinePresets.server";
 import { getTaskEventStoreTableForRun, type TaskEventStoreTable } from "~/v3/taskEventStore.server";
 import { isFailedRunStatus, isFinalRunStatus } from "~/v3/taskStatus";
 import { BasePresenter } from "./basePresenter.server";
 import { WaitpointPresenter } from "./WaitpointPresenter.server";
 import { engine } from "~/v3/runEngine.server";
+import { resolveEventRepositoryForStore } from "~/v3/eventRepository/index.server";
+import { IEventRepository, SpanDetail } from "~/v3/eventRepository/eventRepository.types";
 
 type Result = Awaited<ReturnType<SpanPresenter["call"]>>;
 export type Span = NonNullable<NonNullable<Result>["span"]>;
 export type SpanRun = NonNullable<NonNullable<Result>["run"]>;
 type FindRunResult = NonNullable<
   Awaited<ReturnType<InstanceType<typeof SpanPresenter>["findRun"]>>
 >;
-type GetSpanResult = NonNullable<Awaited<ReturnType<(typeof eventRepository)["getSpan"]>>>;
+type GetSpanResult = SpanDetail;
 
 export class SpanPresenter extends BasePresenter {
   public async call({
@@ -74,14 +76,20 @@ export class SpanPresenter extends BasePresenter {
       return;
     }
 
+    const { traceId } = parentRun;
+
+    const eventRepository = resolveEventRepositoryForStore(parentRun.taskEventStore);
+
     const eventStore = getTaskEventStoreTableForRun(parentRun);
 
     const run = await this.getRun({
       eventStore,
-      environmentId: parentRun.runtimeEnvironmentId,
+      traceId,
+      eventRepository,
       spanId,
       createdAt: parentRun.createdAt,
       completedAt: parentRun.completedAt,
+      environmentId: parentRun.runtimeEnvironmentId,
     });
     if (run) {
       return {
@@ -93,10 +101,12 @@ export class SpanPresenter extends BasePresenter {
     const span = await this.#getSpan({
       eventStore,
       spanId,
+      traceId,
       environmentId: parentRun.runtimeEnvironmentId,
       projectId: parentRun.projectId,
       createdAt: parentRun.createdAt,
       completedAt: parentRun.completedAt,
+      eventRepository,
     });
 
     if (!span) {
@@ -112,29 +122,30 @@ export class SpanPresenter extends BasePresenter {
   async getRun({
     eventStore,
     environmentId,
+    traceId,
+    eventRepository,
     spanId,
     createdAt,
     completedAt,
   }: {
     eventStore: TaskEventStoreTable;
     environmentId: string;
+    traceId: string;
+    eventRepository: IEventRepository;
     spanId: string;
     createdAt: Date;
     completedAt: Date | null;
   }) {
-    const span = await eventRepository.getSpan({
-      storeTable: eventStore,
+    const originalRunId = await eventRepository.getSpanOriginalRunId(
+      eventStore,
+      environmentId,
       spanId,
       environmentId,
-      startCreatedAt: createdAt,
-      endCreatedAt: completedAt ?? undefined,
-    });
+      createdAt,
+      completedAt ?? undefined
+    );
 
-    if (!span) {
-      return;
-    }
-
-    const run = await this.findRun({ span, spanId });
+    const run = await this.findRun({ originalRunId, spanId, environmentId });
 
     if (!run) {
       return;
@@ -259,7 +270,7 @@ export class SpanPresenter extends BasePresenter {
       workerQueue: run.workerQueue,
       traceId: run.traceId,
       spanId: run.spanId,
-      isCached: !!span.originalRun,
+      isCached: !!originalRunId,
       machinePreset: machine?.name,
       externalTraceId,
     };
@@ -294,7 +305,15 @@ export class SpanPresenter extends BasePresenter {
     };
   }
 
-  async findRun({ span, spanId }: { span: GetSpanResult; spanId: string }) {
+  async findRun({
+    originalRunId,
+    spanId,
+    environmentId,
+  }: {
+    originalRunId?: string;
+    spanId: string;
+    environmentId: string;
+  }) {
     const run = await this._replica.taskRun.findFirst({
       select: {
         id: true,
@@ -404,12 +423,14 @@ export class SpanPresenter extends BasePresenter {
           },
         },
       },
-      where: span.originalRun
+      where: originalRunId
         ? {
-            friendlyId: span.originalRun,
+            friendlyId: originalRunId,
+            runtimeEnvironmentId: environmentId,
           }
         : {
             spanId,
+            runtimeEnvironmentId: environmentId,
           },
     });
 
@@ -418,27 +439,32 @@ export class SpanPresenter extends BasePresenter {
 
   async #getSpan({
     eventStore,
+    eventRepository,
+    traceId,
     spanId,
     environmentId,
     projectId,
     createdAt,
     completedAt,
   }: {
+    eventRepository: IEventRepository;
+    traceId: string;
     spanId: string;
     environmentId: string;
     projectId: string;
     eventStore: TaskEventStoreTable;
     createdAt: Date;
     completedAt: Date | null;
   }) {
-    const span = await eventRepository.getSpan({
-      storeTable: eventStore,
-      spanId,
+    const span = await eventRepository.getSpan(
+      eventStore,
       environmentId,
-      startCreatedAt: createdAt,
-      endCreatedAt: completedAt ?? undefined,
-      options: { includeDebugLogs: true },
-    });
+      spanId,
+      traceId,
+      createdAt,
+      completedAt ?? undefined,
+      { includeDebugLogs: true }
+    );
 
     if (!span) {
       return;
@@ -451,23 +477,29 @@ export class SpanPresenter extends BasePresenter {
         spanId: true,
         createdAt: true,
         number: true,
-        lockedToVersion: {
-          select: {
-            version: true,
-          },
-        },
+        taskVersion: true,
       },
       where: {
         parentSpanId: spanId,
       },
     });
 
     const data = {
-      ...span,
+      spanId: span.spanId,
+      parentId: span.parentId,
+      message: span.message,
+      isError: span.isError,
+      isPartial: span.isPartial,
+      isCancelled: span.isCancelled,
+      level: span.level,
+      startTime: span.startTime,
+      duration: span.duration,
       events: span.events,
+      style: span.style,
       properties: span.properties ? JSON.stringify(span.properties, null, 2) : undefined,
+      entity: span.entity,
+      metadata: span.metadata,
       triggeredRuns,
-      showActionBar: span.show?.actions === true,
     };
 
     switch (span.entity.type) {

apps/webapp/app/routes/_app.orgs.$organizationSlug.projects.$projectParam.env.$envParam.runs.$runParam/route.tsx

@@ -97,6 +97,7 @@ import { useCurrentPlan } from "../_app.orgs.$organizationSlug/route";
 import { SpanView } from "../resources.orgs.$organizationSlug.projects.$projectParam.env.$envParam.runs.$runParam.spans.$spanParam/route";
 import { useSearchParams } from "~/hooks/useSearchParam";
 import { CopyableText } from "~/components/primitives/CopyableText";
+import type { SpanOverride } from "~/v3/eventRepository/eventRepository.types";
 
 const resizableSettings = {
   parent: {
@@ -301,7 +302,8 @@ function TraceView({ run, trace, maximumLiveReloadingSetting, resizable }: Loade
     return <></>;
   }
 
-  const { events, duration, rootSpanStatus, rootStartedAt, queuedDuration } = trace;
+  const { events, duration, rootSpanStatus, rootStartedAt, queuedDuration, overridesBySpanId } =
+    trace;
   const shouldLiveReload = events.length <= maximumLiveReloadingSetting;
 
   const changeToSpan = useDebounce((selectedSpan: string) => {
@@ -323,6 +325,8 @@ function TraceView({ run, trace, maximumLiveReloadingSetting, resizable }: Loade
     // WARNING Don't put the revalidator in the useEffect deps array or bad things will happen
   }, [streamedEvents]); // eslint-disable-line react-hooks/exhaustive-deps
 
+  const spanOverrides = selectedSpanId ? overridesBySpanId?.[selectedSpanId] : undefined;
+
   return (
     <div className={cn("grid h-full max-h-full grid-cols-1 overflow-hidden")}>
       <ResizablePanelGroup
@@ -371,6 +375,7 @@ function TraceView({ run, trace, maximumLiveReloadingSetting, resizable }: Loade
             <SpanView
               runParam={run.friendlyId}
               spanId={selectedSpanId}
+              spanOverrides={spanOverrides as SpanOverride | undefined}
               closePanel={() => replaceSearchParam("span")}
             />
           </ResizablePanel>