Merge remote-tracking branch 'origin/main' into features/v3-docs

Author: matt-aitken

.env.example

@@ -55,6 +55,8 @@ COORDINATOR_SECRET=coordinator-secret # generate the actual secret with `openssl
 # ENABLE_REGISTRY_PROXY=true
 # DEPOT_TOKEN=<Depot org token>
 # DEPOT_PROJECT_ID=<Depot project id>
+# DEPLOY_REGISTRY_HOST=${APP_ORIGIN} # This is the host that the deploy CLI will use to push images to the registry
 # CONTAINER_REGISTRY_ORIGIN=<Container registry origin e.g. https://registry.digitalocean.com>
 # CONTAINER_REGISTRY_USERNAME=<Container registry username e.g. Digital ocean email address>
-# CONTAINER_REGISTRY_PASSWORD=<Container registry password e.g. Digital ocean PAT>
+# CONTAINER_REGISTRY_PASSWORD=<Container registry password e.g. Digital ocean PAT>
+# DEV_OTEL_EXPORTER_OTLP_ENDPOINT="http://0.0.0.0:4318"

.github/workflows/e2e.yml

@@ -4,7 +4,7 @@ on:
 jobs:
   e2e:
     name: "🧪 E2E Tests"
-    runs-on: buildjet-4vcpu-ubuntu-2204
+    runs-on: buildjet-16vcpu-ubuntu-2204
     steps:
       - name: 🐳 Login to Docker Hub
         uses: docker/login-action@v2

.github/workflows/publish-dev.yml

@@ -38,7 +38,7 @@ jobs:
     strategy:
       matrix:
         package: [coordinator, kubernetes-provider]
-    runs-on: buildjet-4vcpu-ubuntu-2204
+    runs-on: buildjet-16vcpu-ubuntu-2204
     env:
       DOCKER_BUILDKIT: "1"
     steps:

.github/workflows/publish-docker.yml

@@ -3,21 +3,14 @@ on:
   workflow_call:
 jobs:
   publish:
-    strategy:
-      fail-fast: true # when a job fails, all remaining ones will be cancelled
-      matrix:
-        runs-on: [buildjet-4vcpu-ubuntu-2204, buildjet-4vcpu-ubuntu-2204-arm]
-    name: ${{matrix.runs-on}}
-    runs-on: ${{matrix.runs-on}}
+    runs-on: ubuntu-latest
     outputs:
       version: ${{ steps.get_version.outputs.version }}
       short_sha: ${{ steps.get_commit.outputs.sha_short }}
     steps:
-      - name: 🐳 Login to Docker Hub
-        uses: docker/login-action@v2
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Setup Depot CLI
+        uses: depot/setup-action@v1
+        
       - name: ⬇️ Checkout repo
         uses: actions/checkout@v3
         with:
@@ -43,44 +36,24 @@ jobs:
             echo "Invalid reference: ${GITHUB_REF}"
             exit 1
           fi
-          if [[ ${{matrix.runs-on}} == *-arm ]]; then
-            IMAGE_TAG="${IMAGE_TAG}-arm"
-          fi
           echo "::set-output name=version::${IMAGE_TAG}"
       - name: 🔢 Get the commit hash
         id: get_commit
         run: |
           echo ::set-output name=sha_short::$(echo ${{ github.sha }} | cut -c1-7)
 
-      - name: 🐳 Build Docker Image
-        run: |
-          docker build -t release_build_image -f ./docker/Dockerfile .
-
       - name: 🐙 Login to GitHub Container Registry
         uses: docker/login-action@v2
         with:
           registry: ghcr.io
           username: ${{ github.repository_owner }}
           password: ${{ secrets.GITHUB_TOKEN }}
 
-      - name: 🐙 Push to GitHub Container Registry
-        run: |
-          docker tag release_build_image $REGISTRY/$REPOSITORY:$IMAGE_TAG
-          docker push $REGISTRY/$REPOSITORY:$IMAGE_TAG
-        env:
-          REGISTRY: ghcr.io/triggerdotdev
-          REPOSITORY: trigger.dev
-          IMAGE_TAG: ${{ steps.get_version.outputs.version }}
-
-      - name: 🐙 Push 'latest' to GitHub Container Registry
-        if: startsWith(github.ref, 'refs/tags/v.docker')
-        run: |
-          LATEST=latest
-          if [[ ${{matrix.runs-on}} == *-arm ]]; then
-            LATEST="${LATEST}-arm"
-          fi
-          docker tag release_build_image $REGISTRY/$REPOSITORY:$LATEST
-          docker push $REGISTRY/$REPOSITORY:$LATEST
-        env:
-          REGISTRY: ghcr.io/triggerdotdev
-          REPOSITORY: trigger.dev
+      - name: 🐳 Build image and push to GitHub Container Registry
+        uses: depot/build-push-action@v1
+        with:
+          file: ./docker/Dockerfile
+          platforms: linux/amd64,linux/arm64
+          tags: |
+            ghcr.io/triggerdotdev/trigger.dev:${{ steps.get_version.outputs.version }}
+          push: true

.github/workflows/release.yml

@@ -12,7 +12,7 @@ on:
 jobs:
   release:
     name: 🦋 Changesets Release
-    runs-on: buildjet-4vcpu-ubuntu-2204
+    runs-on: buildjet-16vcpu-ubuntu-2204
     if: |
       github.repository == 'triggerdotdev/trigger.dev'
     outputs:

.github/workflows/typecheck.yml

@@ -3,7 +3,7 @@ on:
   workflow_call:
 jobs:
   typecheck:
-    runs-on: buildjet-4vcpu-ubuntu-2204
+    runs-on: buildjet-16vcpu-ubuntu-2204
 
     steps:
       - name: ⬇️ Checkout repo

.github/workflows/unit-tests.yml

@@ -4,7 +4,7 @@ on:
 jobs:
   unitTests:
     name: "🧪 Unit Tests"
-    runs-on: buildjet-4vcpu-ubuntu-2204
+    runs-on: buildjet-16vcpu-ubuntu-2204
     steps:
       - name: ⬇️ Checkout repo
         uses: actions/checkout@v3

.vscode/launch.json

@@ -33,7 +33,7 @@
       "type": "node-terminal",
       "request": "launch",
       "name": "Debug V3 Dev CLI",
-      "command": "pnpm exec trigger.dev dev",
+      "command": "pnpm exec trigger.dev dev --log-level debug",
       "cwd": "${workspaceFolder}/references/v3-catalog",
       "sourceMaps": true
     },

apps/coordinator/src/index.ts

@@ -349,6 +349,7 @@ class TaskCoordinator {
           setSocketDataFromHeader("podName", "x-pod-name");
           setSocketDataFromHeader("contentHash", "x-trigger-content-hash");
           setSocketDataFromHeader("projectRef", "x-trigger-project-ref");
+          setSocketDataFromHeader("runId", "x-trigger-run-id");
           setSocketDataFromHeader("attemptId", "x-trigger-attempt-id");
           setSocketDataFromHeader("envId", "x-trigger-env-id");
           setSocketDataFromHeader("deploymentId", "x-trigger-deployment-id");
@@ -363,7 +364,7 @@ class TaskCoordinator {
         next();
       },
       onConnection: async (socket, handler, sender) => {
-        const logger = new SimpleLogger(`[task][${socket.id}]`);
+        const logger = new SimpleLogger(`[prod-worker][${socket.id}]`);
 
         this.#platformSocket?.send("LOG", {
           metadata: {
@@ -391,6 +392,7 @@ class TaskCoordinator {
           const executionAck = await this.#platformSocket?.sendWithAck("READY_FOR_EXECUTION", {
             version: "v1",
             attemptId: message.attemptId,
+            runId: message.runId,
           });
 
           if (!executionAck) {
@@ -404,7 +406,7 @@ class TaskCoordinator {
           }
 
           if (!executionAck.success) {
-            logger.error("execution unsuccessful", { attemptId: socket.data.attemptId });
+            logger.error("failed to get execution payload", { attemptId: socket.data.attemptId });
 
             socket.emit("REQUEST_EXIT", {
               version: "v1",
@@ -424,26 +426,87 @@ class TaskCoordinator {
           this.#platformSocket?.send("READY_FOR_RESUME", message);
         });
 
-        socket.on("TASK_RUN_COMPLETED", async (message, callback) => {
-          logger.log("completed task", { completionId: message.completion.id });
+        socket.on("TASK_RUN_COMPLETED", async ({ completion, execution }, callback) => {
+          logger.log("completed task", { completionId: completion.id });
 
-          this.#platformSocket?.send("TASK_RUN_COMPLETED", {
+          const sendCompletionToPlatform = () => {
+            this.#platformSocket?.send("TASK_RUN_COMPLETED", {
+              version: "v1",
+              execution,
+              completion,
+            });
+          };
+
+          const confirmCompletion = ({
+            didCheckpoint,
+            shouldExit,
+          }: {
+            didCheckpoint: boolean;
+            shouldExit: boolean;
+          }) => {
+            sendCompletionToPlatform();
+            callback({ didCheckpoint, shouldExit });
+          };
+
+          if (completion.ok) {
+            confirmCompletion({ didCheckpoint: false, shouldExit: true });
+            return;
+          }
+
+          if (
+            completion.error.type === "INTERNAL_ERROR" &&
+            completion.error.code === "TASK_RUN_CANCELLED"
+          ) {
+            confirmCompletion({ didCheckpoint: false, shouldExit: true });
+            return;
+          }
+
+          if (completion.retry === undefined) {
+            confirmCompletion({ didCheckpoint: false, shouldExit: true });
+            return;
+          }
+
+          const { canCheckpoint, willSimulate } = await this.#checkpointer.initialize();
+
+          const willCheckpointAndRestore = canCheckpoint || willSimulate;
+
+          if (!willCheckpointAndRestore) {
+            confirmCompletion({ didCheckpoint: false, shouldExit: false });
+            return;
+          }
+
+          const checkpoint = await this.#checkpointer.checkpointAndPush(socket.data.podName);
+
+          if (!checkpoint) {
+            logger.error("Failed to checkpoint", { podName: socket.data.podName });
+            confirmCompletion({ didCheckpoint: false, shouldExit: false });
+            return;
+          }
+
+          this.#platformSocket?.send("CHECKPOINT_CREATED", {
             version: "v1",
-            execution: message.execution,
-            completion: message.completion,
+            attemptId: socket.data.attemptId,
+            docker: checkpoint.docker,
+            location: checkpoint.destination,
+            reason: {
+              type: "RETRYING_AFTER_FAILURE",
+              attemptNumber: execution.attempt.number,
+            },
           });
 
-          callback();
+          confirmCompletion({ didCheckpoint: true, shouldExit: false });
         });
 
         socket.on("WAIT_FOR_DURATION", async (message, callback) => {
           logger.log("[WAIT_FOR_DURATION]", message);
 
           const { canCheckpoint, willSimulate } = await this.#checkpointer.initialize();
 
-          callback({ willCheckpointAndRestore: canCheckpoint || willSimulate });
+          const willCheckpointAndRestore = canCheckpoint || willSimulate;
+
+          callback({ willCheckpointAndRestore });
 
-          if (!canCheckpoint) {
+          if (!willCheckpointAndRestore) {
             return;
           }
 
@@ -476,7 +539,13 @@ class TaskCoordinator {
 
           const { canCheckpoint, willSimulate } = await this.#checkpointer.initialize();
 
-          callback({ willCheckpointAndRestore: canCheckpoint || willSimulate });
+          const willCheckpointAndRestore = canCheckpoint || willSimulate;
+
+          callback({ willCheckpointAndRestore });
+
+          if (!willCheckpointAndRestore) {
+            return;
+          }
 
           const checkpoint = await this.#checkpointer.checkpointAndPush(socket.data.podName);
 
@@ -502,7 +571,13 @@ class TaskCoordinator {
 
           const { canCheckpoint, willSimulate } = await this.#checkpointer.initialize();
 
-          callback({ willCheckpointAndRestore: canCheckpoint || willSimulate });
+          const willCheckpointAndRestore = canCheckpoint || willSimulate;
+
+          callback({ willCheckpointAndRestore });
+
+          if (!willCheckpointAndRestore) {
+            return;
+          }
 
           const checkpoint = await this.#checkpointer.checkpointAndPush(socket.data.podName);
 

apps/coordinator/tsconfig.json

@@ -5,6 +5,7 @@
     "target": "es2016",
     "module": "commonjs",
     "esModuleInterop": true,
+    "resolveJsonModule": true,
     "forceConsistentCasingInFileNames": true,
     "strict": true,
     "skipLibCheck": true,