Prettier errors and related improvements

.changeset/plenty-spoons-build.md

@@ -0,0 +1,6 @@
+---
+"trigger.dev": patch
+"@trigger.dev/core": patch
+---
+
+Prettier and more specific errors with links to docs

apps/coordinator/src/index.ts

@@ -401,13 +401,16 @@ class TaskCoordinator {
               success: true,
             };
           } catch (error) {
-            logger.error("Error while waiting for checkpointable state", { error });
+            logger.error("Error while waiting for checkpointable state", {
+              error,
+              runId: socket.data.runId,
+            });
 
             if (error instanceof CheckpointReadinessTimeoutError) {
-              await crashRun({
-                name: error.name,
-                message: `Failed to become checkpointable in ${CHECKPOINTABLE_TIMEOUT_SECONDS}s for ${reason}`,
-              });
+              logger.error(
+                `Failed to become checkpointable in ${CHECKPOINTABLE_TIMEOUT_SECONDS}s for ${reason}`,
+                { runId: socket.data.runId }
+              );
 
               return {
                 success: false,
@@ -490,7 +493,7 @@ class TaskCoordinator {
             updateAttemptFriendlyId(executionAck.payload.execution.attempt.id);
             updateAttemptNumber(executionAck.payload.execution.attempt.number);
           } catch (error) {
-            logger.error("Error", { error });
+            logger.error("READY_FOR_EXECUTION error", { error, runId: socket.data.runId });
 
             await crashRun({
               name: "ReadyForExecutionError",
@@ -524,7 +527,10 @@ class TaskCoordinator {
             }
 
             if (!lazyAttempt.success) {
-              logger.error("failed to get lazy attempt payload", { runId: socket.data.runId });
+              logger.error("failed to get lazy attempt payload", {
+                runId: socket.data.runId,
+                reason: lazyAttempt.reason,
+              });
 
               await crashRun({
                 name: "ReadyForLazyAttemptError",
@@ -546,7 +552,7 @@ class TaskCoordinator {
               return;
             }
 
-            logger.error("Error", { error });
+            logger.error("READY_FOR_LAZY_ATTEMPT error", { error, runId: socket.data.runId });
 
             await crashRun({
               name: "ReadyForLazyAttemptError",
@@ -1004,7 +1010,10 @@ class TaskCoordinator {
           });
 
           if (!createAttempt?.success) {
-            logger.debug("no ack while creating attempt", message);
+            logger.debug("no ack while creating attempt", {
+              runId: message.runId,
+              reason: createAttempt?.reason,
+            });
             callback({ success: false, reason: createAttempt?.reason });
             return;
           }

apps/kubernetes-provider/src/taskMonitor.ts

@@ -3,13 +3,14 @@ import { SimpleLogger } from "@trigger.dev/core/v3/apps";
 import { EXIT_CODE_ALREADY_HANDLED, EXIT_CODE_CHILD_NONZERO } from "@trigger.dev/core/v3/apps";
 import { setTimeout } from "timers/promises";
 import PQueue from "p-queue";
-import type { Prettify } from "@trigger.dev/core/v3";
+import { TaskRunErrorCodes, type Prettify, type TaskRunInternalError } from "@trigger.dev/core/v3";
 
 type FailureDetails = Prettify<{
   exitCode: number;
   reason: string;
   logs: string;
   overrideCompletion: boolean;
+  errorCode: TaskRunInternalError["code"];
 }>;
 
 type IndexFailureHandler = (deploymentId: string, details: FailureDetails) => Promise<any>;
@@ -160,18 +161,23 @@ export class TaskMonitor {
     let reason = rawReason || "Unknown error";
     let logs = rawLogs || "";
     let overrideCompletion = false;
+    let errorCode: TaskRunInternalError["code"] = TaskRunErrorCodes.POD_UNKNOWN_ERROR;
 
     switch (rawReason) {
       case "Error":
         reason = "Unknown error.";
+        errorCode = TaskRunErrorCodes.POD_UNKNOWN_ERROR;
         break;
       case "Evicted":
         if (message.startsWith("Pod ephemeral local storage usage")) {
           reason = "Storage limit exceeded.";
+          errorCode = TaskRunErrorCodes.DISK_SPACE_EXCEEDED;
         } else if (message) {
           reason = `Evicted: ${message}`;
+          errorCode = TaskRunErrorCodes.POD_EVICTED;
         } else {
           reason = "Evicted for unknown reason.";
+          errorCode = TaskRunErrorCodes.POD_EVICTED;
         }
 
         if (logs.startsWith("failed to try resolving symlinks")) {
@@ -183,6 +189,7 @@ export class TaskMonitor {
         reason = `${
           exitCode === EXIT_CODE_CHILD_NONZERO ? "Child process" : "Parent process"
         } ran out of memory! Try choosing a machine preset with more memory for this task.`;
+        errorCode = TaskRunErrorCodes.TASK_PROCESS_OOM_KILLED;
         break;
       default:
         break;
@@ -193,6 +200,7 @@ export class TaskMonitor {
       reason,
       logs,
       overrideCompletion,
+      errorCode,
     } satisfies FailureDetails;
 
     const app = pod.metadata?.labels?.app;

apps/webapp/app/components/runs/v3/RunInspector.tsx

@@ -1,6 +1,11 @@
 import { CheckIcon, ClockIcon, CloudArrowDownIcon, QueueListIcon } from "@heroicons/react/20/solid";
 import { Link } from "@remix-run/react";
-import { formatDuration, formatDurationMilliseconds, TaskRunError } from "@trigger.dev/core/v3";
+import {
+  formatDuration,
+  formatDurationMilliseconds,
+  TaskRunError,
+  taskRunErrorEnhancer,
+} from "@trigger.dev/core/v3";
 import { useEffect } from "react";
 import { useTypedFetcher } from "remix-typedjson";
 import { ExitIcon } from "~/assets/icons/ExitIcon";
@@ -553,32 +558,39 @@ function RunTimeline({ run }: { run: RawRun }) {
 }
 
 function RunError({ error }: { error: TaskRunError }) {
-  switch (error.type) {
+  const enhancedError = taskRunErrorEnhancer(error);
+
+  switch (enhancedError.type) {
     case "STRING_ERROR":
     case "CUSTOM_ERROR": {
       return (
         <div className="flex flex-col gap-2 rounded-sm border border-rose-500/50 px-3 pb-3 pt-2">
           <CodeBlock
             showCopyButton={false}
             showLineNumbers={false}
-            code={error.raw}
+            code={enhancedError.raw}
             maxLines={20}
           />
         </div>
       );
     }
     case "BUILT_IN_ERROR":
     case "INTERNAL_ERROR": {
-      const name = "name" in error ? error.name : error.code;
+      const name = "name" in enhancedError ? enhancedError.name : enhancedError.code;
       return (
         <div className="flex flex-col gap-2 rounded-sm border border-rose-500/50 px-3 pb-3 pt-2">
           <Header3 className="text-rose-500">{name}</Header3>
-          {error.message && <Callout variant="error">{error.message}</Callout>}
-          {error.stackTrace && (
+          {enhancedError.message && <Callout variant="error">{enhancedError.message}</Callout>}
+          {enhancedError.link && (
+            <Callout variant="docs" to={enhancedError.link.href}>
+              {enhancedError.link.name}
+            </Callout>
+          )}
+          {enhancedError.stackTrace && (
             <CodeBlock
               showCopyButton={false}
               showLineNumbers={false}
-              code={error.stackTrace}
+              code={enhancedError.stackTrace}
               maxLines={20}
             />
           )}

apps/webapp/app/components/runs/v3/SpanEvents.tsx

@@ -1,4 +1,5 @@
 import {
+  exceptionEventEnhancer,
   isExceptionSpanEvent,
   type ExceptionEventProperties,
   type SpanEvent as OtelSpanEvent,
@@ -64,19 +65,26 @@ export function SpanEventError({
   spanEvent: OtelSpanEvent;
   exception: ExceptionEventProperties;
 }) {
+  const enhancedException = exceptionEventEnhancer(exception);
+
   return (
     <div className="flex flex-col gap-2 rounded-sm border border-rose-500/50 px-3 pb-3 pt-2">
       <SpanEventHeader
-        title={exception.type ?? "Error"}
+        title={enhancedException.type ?? "Error"}
         time={spanEvent.time}
         titleClassName="text-rose-500"
       />
-      {exception.message && <Callout variant="error">{exception.message}</Callout>}
-      {exception.stacktrace && (
+      {enhancedException.message && <Callout variant="error">{enhancedException.message}</Callout>}
+      {enhancedException.link && (
+        <Callout variant="docs" to={enhancedException.link.href}>
+          {enhancedException.link.name}
+        </Callout>
+      )}
+      {enhancedException.stacktrace && (
         <CodeBlock
           showCopyButton={false}
           showLineNumbers={false}
-          code={exception.stacktrace}
+          code={enhancedException.stacktrace}
           maxLines={20}
         />
       )}

apps/webapp/app/routes/resources.orgs.$organizationSlug.projects.v3.$projectParam.runs.$runParam.spans.$spanParam/route.tsx

@@ -6,6 +6,7 @@ import {
   formatDurationMilliseconds,
   nanosecondsToMilliseconds,
   TaskRunError,
+  taskRunErrorEnhancer,
 } from "@trigger.dev/core/v3";
 import { ReactNode, useEffect } from "react";
 import { typedjson, useTypedFetcher } from "remix-typedjson";
@@ -933,12 +934,14 @@ function RunTimelineLine({ title, state }: RunTimelineLineProps) {
 }
 
 function RunError({ error }: { error: TaskRunError }) {
-  switch (error.type) {
+  const enhancedError = taskRunErrorEnhancer(error);
+
+  switch (enhancedError.type) {
     case "STRING_ERROR":
       return (
         <div className="flex flex-col gap-2 rounded-sm border border-rose-500/50 px-3 pb-3 pt-2">
           <Header3 className="text-rose-500">Error</Header3>
-          <Callout variant="error">{error.raw}</Callout>
+          <Callout variant="error">{enhancedError.raw}</Callout>
         </div>
       );
     case "CUSTOM_ERROR": {
@@ -947,24 +950,29 @@ function RunError({ error }: { error: TaskRunError }) {
           <CodeBlock
             showCopyButton={false}
             showLineNumbers={false}
-            code={error.raw}
+            code={enhancedError.raw}
             maxLines={20}
           />
         </div>
       );
     }
     case "BUILT_IN_ERROR":
     case "INTERNAL_ERROR": {
-      const name = "name" in error ? error.name : error.code;
+      const name = "name" in enhancedError ? enhancedError.name : enhancedError.code;
       return (
         <div className="flex flex-col gap-2 rounded-sm border border-rose-500/50 px-3 pb-3 pt-2">
           <Header3 className="text-rose-500">{name}</Header3>
-          {error.message && <Callout variant="error">{error.message}</Callout>}
-          {error.stackTrace && (
+          {enhancedError.message && <Callout variant="error">{enhancedError.message}</Callout>}
+          {enhancedError.link && (
+            <Callout variant="docs" to={enhancedError.link.href}>
+              {enhancedError.link.name}
+            </Callout>
+          )}
+          {enhancedError.stackTrace && (
             <CodeBlock
               showCopyButton={false}
               showLineNumbers={false}
-              code={error.stackTrace}
+              code={enhancedError.stackTrace}
               maxLines={20}
             />
           )}

apps/webapp/app/v3/handleSocketIo.server.ts

@@ -188,7 +188,7 @@ function createCoordinatorNamespace(io: Server) {
           const environment = await findEnvironmentById(message.envId);
 
           if (!environment) {
-            logger.error("Environment not found", { id: message.envId });
+            logger.error("CREATE_TASK_RUN_ATTEMPT: Environment not found", message);
             return { success: false, reason: "Environment not found" };
           }
 
@@ -198,16 +198,14 @@ function createCoordinatorNamespace(io: Server) {
           const payload = await sharedQueueTasks.getExecutionPayloadFromAttempt(attempt.id, true);
 
           if (!payload) {
-            logger.error("Failed to retrieve payload after attempt creation", {
-              id: message.envId,
-            });
+            logger.error("Failed to retrieve payload after attempt creation", message);
             return { success: false, reason: "Failed to retrieve payload" };
           }
 
           return { success: true, executionPayload: payload };
         } catch (error) {
           logger.error("Error while creating attempt", {
-            runId: message.runId,
+            ...message,
             error,
           });
           return { success: false };

apps/webapp/app/v3/marqs/sharedQueueConsumer.server.ts

@@ -1160,7 +1160,7 @@ class SharedQueueTasks {
     const environment = await findEnvironmentById(envId);
 
     if (!environment) {
-      logger.error("Environment not found", { id: envId });
+      logger.error("getLazyAttemptPayload: Environment not found", { runId, envId });
       return;
     }
 
@@ -1182,7 +1182,7 @@ class SharedQueueTasks {
     });
 
     if (!run) {
-      logger.error("Run not found", { id: runId, envId });
+      logger.error("getLazyAttemptPayload: Run not found", { runId, envId });
       return;
     }
 

apps/webapp/app/v3/services/crashTaskRun.server.ts

@@ -5,7 +5,7 @@ import { BaseService } from "./baseService.server";
 import { logger } from "~/services/logger.server";
 import { AuthenticatedEnvironment } from "~/services/apiAuth.server";
 import { CRASHABLE_ATTEMPT_STATUSES, isCrashableRunStatus } from "../taskStatus";
-import { sanitizeError } from "@trigger.dev/core/v3";
+import { sanitizeError, TaskRunInternalError } from "@trigger.dev/core/v3";
 import { FinalizeTaskRunService } from "./finalizeTaskRun.server";
 
 export type CrashTaskRunServiceOptions = {
@@ -15,6 +15,7 @@ export type CrashTaskRunServiceOptions = {
   crashAttempts?: boolean;
   crashedAt?: Date;
   overrideCompletion?: boolean;
+  errorCode?: TaskRunInternalError["code"];
 };
 
 export class CrashTaskRunService extends BaseService {
@@ -26,6 +27,8 @@ export class CrashTaskRunService extends BaseService {
       ...options,
     };
 
+    logger.debug("CrashTaskRunService.call", { runId, opts });
+
     const taskRun = await this._prisma.taskRun.findFirst({
       where: {
         id: runId,
@@ -71,7 +74,7 @@ export class CrashTaskRunService extends BaseService {
       attemptStatus: "FAILED",
       error: {
         type: "INTERNAL_ERROR",
-        code: "TASK_RUN_CRASHED",
+        code: opts.errorCode ?? "TASK_RUN_CRASHED",
         message: opts.reason,
         stackTrace: opts.logs,
       },
@@ -129,6 +132,7 @@ export class CrashTaskRunService extends BaseService {
     error: {
       reason: string;
       logs?: string;
+      code?: TaskRunInternalError["code"];
     }
   ) {
     return await this.traceWithEnv("failAttempt()", environment, async (span) => {
@@ -146,7 +150,7 @@ export class CrashTaskRunService extends BaseService {
           completedAt: failedAt,
           error: sanitizeError({
             type: "INTERNAL_ERROR",
-            code: "TASK_RUN_CRASHED",
+            code: error.code ?? "TASK_RUN_CRASHED",
             message: error.reason,
             stackTrace: error.logs,
           }),