some errors should never be retried

Author: nicktrn

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

@@ -8,6 +8,7 @@ import {
   TaskRunSuccessfulExecutionResult,
   flattenAttributes,
   sanitizeError,
+  shouldRetryError,
 } from "@trigger.dev/core/v3";
 import { $transaction, PrismaClientOrTransaction } from "~/db.server";
 import { AuthenticatedEnvironment } from "~/services/apiAuth.server";
@@ -216,7 +217,11 @@ export class CompleteAttemptService extends BaseService {
 
     const environment = env ?? (await this.#getEnvironment(execution.environment.id));
 
-    if (completion.retry !== undefined && taskRunAttempt.number < MAX_TASK_RUN_ATTEMPTS) {
+    if (
+      shouldRetryError(completion.error) &&
+      completion.retry !== undefined &&
+      taskRunAttempt.number < MAX_TASK_RUN_ATTEMPTS
+    ) {
       return await this.#retryAttempt({
         execution,
         executionRetry: completion.retry,

packages/core/src/v3/errors.ts

@@ -10,6 +10,7 @@ import {
 import { TaskMetadataFailedToParseData } from "./schemas/messages.js";
 import { links } from "./links.js";
 import { ExceptionEventProperties } from "./schemas/openTelemetry.js";
+import { assertExhaustive } from "../utils.js";
 
 export class AbortTaskRunError extends Error {
   constructor(message: string) {
@@ -137,6 +138,55 @@ export function sanitizeError(error: TaskRunError): TaskRunError {
   }
 }
 
+export function shouldRetryError(error: TaskRunError): boolean {
+  switch (error.type) {
+    case "INTERNAL_ERROR": {
+      switch (error.code) {
+        case "COULD_NOT_FIND_EXECUTOR":
+        case "COULD_NOT_FIND_TASK":
+        case "COULD_NOT_IMPORT_TASK":
+        case "CONFIGURED_INCORRECTLY":
+        case "TASK_ALREADY_RUNNING":
+        case "TASK_PROCESS_EXITED_WITH_NON_ZERO_CODE":
+        case "TASK_PROCESS_SIGKILL_TIMEOUT":
+        case "TASK_PROCESS_OOM_KILLED":
+        case "TASK_PROCESS_MAYBE_OOM_KILLED":
+        case "TASK_RUN_CANCELLED":
+        case "TASK_OUTPUT_ERROR":
+        case "MAX_DURATION_EXCEEDED":
+        case "DISK_SPACE_EXCEEDED":
+          return false;
+
+        case "GRACEFUL_EXIT_TIMEOUT":
+        case "HANDLE_ERROR_ERROR":
+        case "IMPORT_PAYLOAD_ERROR":
+        case "POD_EVICTED":
+        case "POD_UNKNOWN_ERROR":
+        case "TASK_EXECUTION_ABORTED":
+        case "TASK_EXECUTION_FAILED":
+        case "TASK_RUN_CRASHED":
+        case "TASK_RUN_HEARTBEAT_TIMEOUT":
+          return true;
+
+        default:
+          assertExhaustive(error.code);
+      }
+    }
+    case "STRING_ERROR": {
+      return true;
+    }
+    case "BUILT_IN_ERROR": {
+      return true;
+    }
+    case "CUSTOM_ERROR": {
+      return true;
+    }
+    default: {
+      assertExhaustive(error);
+    }
+  }
+}
+
 export function correctErrorStackTrace(
   stackTrace: string,
   projectDir?: string,