Enable custom tags (#6)

Author: Ankcorn

package.json

@@ -12,7 +12,7 @@
 		"deploy": "wrangler deploy",
 		"start": "wrangler dev",
 		"dev": "wrangler dev",
-		"cf-typegen": "wrangler types",
+		"cf-typegen": "wrangler types --strict-vars=false",
 		"typecheck": "tsgo",
 		"lint": "biome check .",
 		"lint:fix": "biome check --write .",

readme.md

@@ -33,13 +33,36 @@ To run GraphQL queries from a specific jurisdiction (closer to the data source),
 
 ```jsonc
 "vars": {
-  "BATCH_SIZE": "5000",
-  "JURISDICTION": "eu" // e.g., "eu", "fedramp"
+  "BATCH_SIZE": 5000,
+  "RETRY_LIMIT": 3,
+  "RETRY_DELAY_SECONDS": 1,
+  "JURISDICTION": "eu", // e.g., "eu", "fedramp"
+  "DATADOG_TAGS": {}
 }
 ```
 
 This uses a Durable Object to proxy requests from the specified jurisdiction.
 
+#### Optional: Custom Datadog Tags
+
+Add custom tags to all metrics by setting the `DATADOG_TAGS` variable in `wrangler.jsonc`:
+
+```jsonc
+"vars": {
+  "BATCH_SIZE": 5000,
+  "RETRY_LIMIT": 3,
+  "RETRY_DELAY_SECONDS": 1,
+  "JURISDICTION": "eu",
+  "DATADOG_TAGS": {
+    "env": "production",
+    "team": "platform",
+    "service": "containers"
+  }
+}
+```
+
+These tags will be added to all health and resource metrics sent to Datadog.
+
 ### Verify
 
 ```bash
@@ -123,11 +146,21 @@ See [Datadog's documentation](https://docs.datadoghq.com/dashboards/configure/#c
 
 ## Workflow Behavior
 
-The exporter runs as a Cloudflare Workflow triggered every minute via cron. Each workflow step uses the default retry configuration:
+The exporter runs as a Cloudflare Workflow triggered every minute via cron. Each workflow step uses configurable retry settings:
 
-- **Retries**: 3 attempts
-- **Delay**: 1 second initial delay
-- **Backoff**: Exponential (1s, 2s, 4s)
+- **Retries**: Configurable via `RETRY_LIMIT` (default: 3 attempts)
+- **Delay**: Configurable via `RETRY_DELAY_SECONDS` (default: 1 second initial delay)
+- **Backoff**: Exponential (e.g., 1s, 2s, 4s)
 
 Steps will automatically retry on transient failures (API errors, network issues).
 
+### Configuration Options
+
+| Variable | Type | Default | Description |
+|----------|------|---------|-------------|
+| `BATCH_SIZE` | number | 5000 | Maximum metrics per Datadog API request |
+| `RETRY_LIMIT` | number | 3 | Number of retry attempts for failed workflow steps |
+| `RETRY_DELAY_SECONDS` | number | 1 | Initial delay in seconds before retry (exponential backoff) |
+| `JURISDICTION` | string | "" | Durable Object jurisdiction for GraphQL queries (e.g., "eu", "fedramp") |
+| `DATADOG_TAGS` | object | {} | Custom tags to add to all metrics |
+

src/metrics.ts

@@ -1,17 +1,29 @@
+import { z } from "zod/v4";
 import type { DatadogMetric } from "./api/datadog";
 import type { Container, MetricsGroup } from "./types";
 
-export interface ContainerWithMetrics {
-	container: Container;
-	metrics: MetricsGroup[];
-}
-
 export interface ContainerInfo {
 	id: string;
 	name: string;
 	version: number;
 }
 
+const DatadogTagsSchema = z.record(z.string(), z.string()).optional();
+
+function parseCustomTags(datadogTags: unknown): string[] {
+	const parsed = DatadogTagsSchema.safeParse(datadogTags);
+	if (!parsed.success) {
+		console.warn("Invalid DATADOG_TAGS format, ignoring custom tags", {
+			error: parsed.error.message,
+		});
+		return [];
+	}
+	if (!parsed.data) {
+		return [];
+	}
+	return Object.entries(parsed.data).map(([key, value]) => `${key}:${value}`);
+}
+
 /**
  * Format metrics for a single container into Datadog metrics
  */
@@ -20,7 +32,9 @@ export function formatMetricsForContainer(
 	container: ContainerInfo,
 	metricsGroups: MetricsGroup[],
 	timestamp?: number,
+	datadogTags?: unknown,
 ): DatadogMetric[] {
+	const customTags = parseCustomTags(datadogTags);
 	const ts = timestamp ?? Math.floor(Date.now() / 1000);
 	const metrics: DatadogMetric[] = [];
 
@@ -32,6 +46,7 @@ export function formatMetricsForContainer(
 			`version:${container.version}`,
 			`instance_id:${group.dimensions.deploymentId}`,
 			`placement_id:${group.dimensions.placementId}`,
+			...customTags,
 		];
 
 		// CPU metrics
@@ -138,36 +153,18 @@ export function formatMetricsForContainer(
 	return metrics;
 }
 
-/**
- * Format container metrics data into Datadog metrics
- */
-export function formatContainerMetrics(
-	accountId: string,
-	containersWithMetrics: ContainerWithMetrics[],
-	timestamp?: number,
-): DatadogMetric[] {
-	const ts = timestamp ?? Math.floor(Date.now() / 1000);
-	const metrics: DatadogMetric[] = [];
-
-	for (const { container, metrics: groups } of containersWithMetrics) {
-		metrics.push(
-			...formatMetricsForContainer(accountId, container, groups, ts),
-		);
-	}
-
-	return metrics;
-}
-
 /**
  * Format container health data into Datadog metrics
  */
 export function formatHealthMetrics(
 	accountId: string,
 	containers: Container[],
 	timestamp?: number,
+	datadogTags?: unknown,
 ): DatadogMetric[] {
+	const customTags = parseCustomTags(datadogTags);
 	const ts = timestamp ?? Math.floor(Date.now() / 1000);
-	const baseTags = [`account_id:${accountId}`];
+	const baseTags = [`account_id:${accountId}`, ...customTags];
 	const metrics: DatadogMetric[] = [];
 
 	const totals = {

src/workflow.ts

@@ -27,17 +27,19 @@ function getMetricsTimeWindow(now: Date = new Date()): {
 	return { start, end };
 }
 
-const STEP_CONFIG = {
-	retries: {
-		limit: 3,
-		delay: "1 second" as const,
-		backoff: "exponential" as const,
-	},
-};
-
 export class MetricsExporterWorkflow extends WorkflowEntrypoint<Env> {
 	async run(_event: WorkflowEvent<unknown>, step: WorkflowStep) {
-		const batchSize = Number.parseInt(this.env.BATCH_SIZE || "5000", 10);
+		const batchSize = this.env.BATCH_SIZE ?? 5000;
+		const retryLimit = this.env.RETRY_LIMIT ?? 3;
+		const retryDelaySeconds = this.env.RETRY_DELAY_SECONDS ?? 1;
+
+		const stepConfig = {
+			retries: {
+				limit: retryLimit,
+				delay: `${retryDelaySeconds} seconds` as const,
+				backoff: "exponential" as const,
+			},
+		};
 
 		// Create a fetcher that proxies requests through a Durable Object in a specific jurisdiction
 		// This ensures GraphQL queries run close to the data source
@@ -70,14 +72,16 @@ export class MetricsExporterWorkflow extends WorkflowEntrypoint<Env> {
 
 		const containers = await step.do(
 			"fetch containers",
-			STEP_CONFIG,
+			stepConfig,
 			async () => {
 				const result = await cloudflare.listContainers();
 				console.log("Fetched containers", { count: result.length });
 
 				const healthMetrics = formatHealthMetrics(
 					this.env.CLOUDFLARE_ACCOUNT_ID,
 					result,
+					undefined,
+					this.env.DATADOG_TAGS,
 				);
 				await datadog.sendMetrics(healthMetrics);
 
@@ -94,7 +98,7 @@ export class MetricsExporterWorkflow extends WorkflowEntrypoint<Env> {
 		for (const container of containers) {
 			const count = await step.do(
 				`Download Metrics: ${container.name}`,
-				STEP_CONFIG,
+				stepConfig,
 				async () => {
 					const metricsGroups = await cloudflare.getContainerMetrics(
 						container.id,
@@ -106,6 +110,8 @@ export class MetricsExporterWorkflow extends WorkflowEntrypoint<Env> {
 						this.env.CLOUDFLARE_ACCOUNT_ID,
 						container,
 						metricsGroups,
+						undefined,
+						this.env.DATADOG_TAGS,
 					);
 
 					const batches = chunk(metrics, batchSize);
@@ -115,7 +121,7 @@ export class MetricsExporterWorkflow extends WorkflowEntrypoint<Env> {
 							(batch, i) => () =>
 								step.do(
 									`Export Metrics: ${container.name} batch ${i + 1}/${batches.length}`,
-									STEP_CONFIG,
+									stepConfig,
 									async () => {
 										await datadog.sendMetrics(batch);
 									},