Add Lambda rollout with CodeDeploy traffic shifting

Author: anatolzak

examples/aws-lambda-rollout/infra/alarms.ts

@@ -0,0 +1,79 @@
+export function createCanaryAlarms(
+  name: string,
+  opts: {
+    fn: sst.aws.Function;
+    alertsTopic: sst.aws.SnsTopic;
+  },
+) {
+  const { fn, alertsTopic } = opts;
+
+  // These alarms track a specific Lambda version via the ExecutedVersion
+  // dimension. When we deploy a new version, we replace the alarm so it starts
+  // fresh — no leftover state from the previous version. The old alarm is
+  // deleted after the new one is created so the deployment is not interrupted
+  // by the removal of the alarm mid-deploy.
+  const pulumiOpts: $util.CustomResourceOptions = {
+    replaceOnChanges: ["dimensions"],
+    deleteBeforeReplace: false,
+  };
+
+  // Triggers if any errors occur in the deployed version within a 5-minute window.
+  const errorAlarm = new aws.cloudwatch.MetricAlarm(
+    `${name}ErrorAlarm`,
+    {
+      alarmActions: [alertsTopic.arn],
+      namespace: "AWS/Lambda",
+      metricName: "Errors",
+      dimensions: {
+        FunctionName: fn.name,
+        Resource: getFunctionResource(fn.targetArn),
+        ExecutedVersion: fn.nodes.function.version,
+      },
+      statistic: "Sum",
+      period: 300,
+      evaluationPeriods: 1,
+      threshold: 1,
+      comparisonOperator: "GreaterThanOrEqualToThreshold",
+      treatMissingData: "notBreaching",
+    },
+    pulumiOpts,
+  );
+
+  // Triggers if average latency exceeds 2 seconds in a 5-minute window.
+  const latencyAlarm = new aws.cloudwatch.MetricAlarm(
+    `${name}LatencyAlarm`,
+    {
+      alarmActions: [alertsTopic.arn],
+      namespace: "AWS/Lambda",
+      metricName: "Duration",
+      dimensions: {
+        FunctionName: fn.name,
+        Resource: getFunctionResource(fn.targetArn),
+        ExecutedVersion: fn.nodes.function.version,
+      },
+      statistic: "Average",
+      period: 300,
+      evaluationPeriods: 1,
+      threshold: 2000,
+      comparisonOperator: "GreaterThanOrEqualToThreshold",
+      treatMissingData: "notBreaching",
+    },
+    pulumiOpts,
+  );
+
+  return { errorAlarm, latencyAlarm };
+}
+
+/**
+ * Extracts the `FunctionName:Alias` resource identifier from a Lambda alias ARN.
+ * CloudWatch metrics use this format for the `Resource` dimension when tracking
+ * a specific alias.
+ *
+ * For example, given `arn:aws:lambda:us-east-1:123456789:function:my-fn:live`,
+ * this returns `my-fn:live`.
+ */
+function getFunctionResource(targetArn: $util.Input<string>) {
+  return aws
+    .getArnOutput({ arn: targetArn })
+    .resource.apply((r) => r.split(":").slice(1).join(":"));
+}

examples/aws-lambda-rollout/infra/topics.ts

@@ -0,0 +1,36 @@
+export function createTopics(opts?: { email?: string }) {
+  const alertsTopic = new sst.aws.SnsTopic("Alerts");
+
+  allowCloudWatchPublish("Alerts", alertsTopic);
+
+  if (opts?.email) {
+    subscribeEmail("Alerts", alertsTopic, opts.email);
+  }
+
+  return { alertsTopic };
+}
+
+function subscribeEmail(name: string, topic: sst.aws.SnsTopic, email: string) {
+  new aws.sns.TopicSubscription(`${name}Email`, {
+    topic: topic.arn,
+    protocol: "email",
+    endpoint: email,
+  });
+}
+
+function allowCloudWatchPublish(name: string, topic: sst.aws.SnsTopic) {
+  new aws.sns.TopicPolicy(`${name}CloudWatchPolicy`, {
+    arn: topic.arn,
+    policy: aws.iam.getPolicyDocumentOutput({
+      statements: [
+        {
+          actions: ["sns:Publish"],
+          principals: [
+            { type: "Service", identifiers: ["cloudwatch.amazonaws.com"] },
+          ],
+          resources: [topic.arn],
+        },
+      ],
+    }).json,
+  });
+}

examples/aws-lambda-rollout/package.json

@@ -0,0 +1,5 @@
+{
+  "name": "aws-lambda-rollout",
+  "private": true,
+  "type": "module"
+}

examples/aws-lambda-rollout/scripts/test.ts

@@ -0,0 +1,64 @@
+interface Result {
+  status: number;
+  latency: number;
+  version: string;
+  error: boolean;
+}
+
+export async function loadTest(url: string, count = 50) {
+  console.log(`URL: ${url}\n`);
+  console.log(`Sending ${count} concurrent requests...\n`);
+
+  const results: Result[] = await Promise.all(
+    Array.from({ length: count }, async () => {
+      const start = Date.now();
+      try {
+        const res = await fetch(url);
+        const body = (await res.json()) as { version?: string };
+        return {
+          status: res.status,
+          latency: Date.now() - start,
+          version: body.version ?? "unknown",
+          error: false,
+        };
+      } catch {
+        return {
+          status: 0,
+          latency: Date.now() - start,
+          version: "unknown",
+          error: true,
+        };
+      }
+    }),
+  );
+
+  const byVersion = new Map<string, Result[]>();
+  for (const r of results) {
+    if (!byVersion.has(r.version)) byVersion.set(r.version, []);
+    byVersion.get(r.version)!.push(r);
+  }
+
+  for (const [version, vResults] of byVersion) {
+    const succeeded = vResults.filter(
+      (r) => !r.error && r.status >= 200 && r.status < 300,
+    );
+    const failed = vResults.filter((r) => r.error || r.status >= 400);
+    const avgLatency =
+      vResults.reduce((sum, r) => sum + r.latency, 0) / vResults.length;
+
+    console.log(`Version ${version}:`);
+    console.log(`  Requests:    ${vResults.length}`);
+    console.log(`  Succeeded:   ${succeeded.length}`);
+    console.log(`  Failed:      ${failed.length}`);
+    console.log(`  Avg latency: ${Math.round(avgLatency)}ms`);
+
+    if (failed.length > 0) {
+      const statusCounts: Record<number, number> = {};
+      for (const r of failed) {
+        statusCounts[r.status] = (statusCounts[r.status] || 0) + 1;
+      }
+      console.log(`  Status codes:`, statusCounts);
+    }
+    console.log();
+  }
+}

examples/aws-lambda-rollout/src/api.ts

@@ -0,0 +1,9 @@
+export async function handler() {
+  return {
+    statusCode: 200,
+    body: JSON.stringify({
+      message: "Hello from the API",
+      version: process.env.AWS_LAMBDA_FUNCTION_VERSION ?? "unknown",
+    }),
+  };
+}

examples/aws-lambda-rollout/sst.config.ts

@@ -0,0 +1,84 @@
+/// <reference path="./.sst/platform/config.d.ts" />
+
+/**
+ * ## AWS Lambda Rollout
+ *
+ * Deploys a Lambda function with a canary rollout. Each deploy publishes a new version
+ * and uses CodeDeploy to gradually shift traffic — 10% for 10 minutes, then 100%.
+ *
+ * CloudWatch alarms monitor the error rate and latency during the rollout. If either
+ * alarm fires, CodeDeploy automatically rolls back to the previous version.
+ *
+ * An SNS topic sends notifications on failures, rollbacks, and stops.
+ */
+export default $config({
+  app(input) {
+    return {
+      name: "aws-lambda-rollout",
+      home: "aws",
+      removal: input?.stage === "production" ? "retain" : "remove",
+    };
+  },
+  async run() {
+    const { createTopics } = await import("./infra/topics");
+    const { createCanaryAlarms } = await import("./infra/alarms");
+
+    const { alertsTopic } = createTopics({
+      // email: EMAIL,
+    });
+
+    const fn = new sst.aws.Function("Function", {
+      handler: "src/api.handler",
+      rollout: { latestUrl: true },
+      url: true,
+      // Rollout only runs when function code changes. Set to false to deploy
+      // actual code since sst dev deploys a stub that never changes.
+      dev: false,
+    });
+
+    const { errorAlarm: canaryErrorAlarm, latencyAlarm: canaryLatencyAlarm } =
+      createCanaryAlarms("Function", {
+        fn,
+        alertsTopic,
+      });
+
+    fn.addRollout({
+      type: "canary",
+      percentage: 10,
+      duration: "10 minutes",
+      wait: true,
+      alarms: [canaryErrorAlarm.name, canaryLatencyAlarm.name],
+      notifications: [
+        {
+          name: "Alerts",
+          events: ["failure", "rollback", "stop"],
+          topic: alertsTopic.arn,
+        },
+      ],
+    });
+
+    $util
+      .all([
+        fn.url,
+        fn.nodes.function.version,
+        fn.nodes.rolloutDeployment?.apply(
+          (deployment) => deployment?.deploymentId,
+        ),
+      ])
+      .apply(async ([url, version, deploymentId]) => {
+        // wait for CodeDeploy to update the lambda alias
+        await new Promise((r) => setTimeout(r, 10_000));
+
+        console.log(
+          `\nDeployed version ${version} (deployment: ${deploymentId})\n`,
+        );
+        const { loadTest } = await import("./scripts/test");
+        await loadTest(url);
+      });
+
+    return {
+      url: fn.url,
+      latestUrl: fn.latestUrl,
+    };
+  },
+});

examples/aws-lambda-rollout/tsconfig.json

@@ -0,0 +1,11 @@
+{
+  "compilerOptions": {
+    "target": "ESNext",
+    "module": "ESNext",
+    "moduleResolution": "bundler",
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "forceConsistentCasingInFileNames": true
+  }
+}

examples/aws-lambda-smoke-test-function-url/package.json

@@ -0,0 +1,7 @@
+{
+  "name": "aws-lambda-smoke-test-function-url",
+  "private": true,
+  "dependencies": {
+    "sst": "^4"
+  }
+}

examples/aws-lambda-smoke-test-function-url/src/api.ts

@@ -0,0 +1,19 @@
+export async function handler(event: any) {
+  if (event.type === "health-check") {
+    return {
+      statusCode: 200,
+      body: JSON.stringify({ status: "healthy" }),
+    };
+  }
+
+  return {
+    statusCode: 200,
+    body: JSON.stringify({
+      message: "Hello from the API",
+      version: process.env.AWS_LAMBDA_FUNCTION_VERSION,
+    }),
+    headers: {
+      "content-type": "application/json",
+    },
+  };
+}

examples/aws-lambda-smoke-test-function-url/src/before-traffic.ts

@@ -0,0 +1,24 @@
+import { Resource } from "sst";
+import { rollout } from "sst/aws/rollout";
+
+export const handler = rollout.handler(async (event) => {
+  const status = await validate();
+  await rollout.report(event, status);
+});
+
+async function validate(): Promise<"Succeeded" | "Failed"> {
+  try {
+    const resp = await fetch(Resource.Function.latestUrl);
+    const payload = await resp.text();
+
+    if (resp.ok) {
+      console.log("Health check passed:", payload);
+      return "Succeeded";
+    }
+    console.log("Health check failed:", resp.status, payload);
+    return "Failed";
+  } catch (err) {
+    console.error("Validation failed:", err);
+    return "Failed";
+  }
+}