Only retry runs when the response is a non-400

apps/webapp/app/services/endpointApi.server.ts

@@ -285,6 +285,7 @@ async function safeFetch(url: string, options: RequestInit) {
   } catch (error) {
     logger.debug("Error while trying to connect to endpoint", {
       url,
+      error,
     });
   }
 }

apps/webapp/app/services/runs/performRunExecution.server.ts

@@ -22,6 +22,7 @@ import { safeJsonZodParse } from "~/utils/json";
 import { EndpointApi } from "../endpointApi.server";
 import { workerQueue } from "../worker.server";
 import { formatError } from "~/utils/formatErrors.server";
+import { logger } from "../logger.server";
 
 type FoundRunExecution = NonNullable<
   Awaited<ReturnType<typeof findRunExecution>>
@@ -290,22 +291,47 @@ export class PerformRunExecutionService {
 
     if (!response) {
       return await this.#failRunExecutionWithRetry(execution, {
-        message: "Could not connect to the endpoint",
+        message: `Connection could not be established to the endpoint (${run.endpoint.url})`,
       });
     }
 
     const rawBody = await response.text();
 
     if (!response.ok) {
+      logger.debug("Endpoint responded with non-200 status code", {
+        status: response.status,
+        runId: run.id,
+        endpoint: run.endpoint.url,
+      });
+
       const errorBody = safeJsonZodParse(errorParser, rawBody);
 
       if (errorBody && errorBody.success) {
-        return await this.#failRunExecutionWithRetry(execution, errorBody.data);
+        // Only retry if the error isn't a 4xx
+        if (response.status >= 400 && response.status <= 499) {
+          return await this.#failRunExecution(
+            this.#prismaClient,
+            execution,
+            errorBody.data
+          );
+        } else {
+          return await this.#failRunExecutionWithRetry(
+            execution,
+            errorBody.data
+          );
+        }
       }
 
-      return await this.#failRunExecutionWithRetry(execution, {
-        message: `Endpoint responded with ${response.status} status code`,
-      });
+      // Only retry if the error isn't a 4xx
+      if (response.status >= 400 && response.status <= 499) {
+        return await this.#failRunExecution(this.#prismaClient, execution, {
+          message: `Endpoint responded with ${response.status} status code`,
+        });
+      } else {
+        return await this.#failRunExecutionWithRetry(execution, {
+          message: `Endpoint responded with ${response.status} status code`,
+        });
+      }
     }
 
     const safeBody = safeJsonZodParse(parser, rawBody);

examples/nextjs-example/src/jobs/edgeCases.ts

@@ -0,0 +1,27 @@
+import { client } from "@/trigger";
+import { Job, eventTrigger } from "@trigger.dev/sdk";
+import { z } from "zod";
+
+new Job(client, {
+  id: "test-long-running-cpu",
+  name: "Test long running CPU",
+  version: "0.0.1",
+  trigger: eventTrigger({
+    name: "test.cpu",
+    schema: z.object({
+      iterations: z.number(),
+      sleepDuration: z.number(),
+    }),
+  }),
+  run: async (payload, io, ctx) => {
+    console.log(`Running run ${ctx.run.id} at ${new Date().toISOString()}`);
+
+    for (let i = 0; i < payload.iterations ?? 1; i++) {
+      await new Promise((resolve) =>
+        setTimeout(resolve, payload.sleepDuration ?? 1000)
+      );
+    }
+
+    console.log(`Finishing run ${ctx.run.id} at ${new Date().toISOString()}`);
+  },
+});

examples/nextjs-example/src/pages/api/trigger.ts

@@ -10,6 +10,7 @@ import "@/jobs/resend";
 import "@/jobs/schedules";
 import "@/jobs/slack";
 import "@/jobs/typeform";
+import "@/jobs/edgeCases";
 
 import { createPagesRoute } from "@trigger.dev/nextjs";