@deliciousmonster/datadog-agent-binary 7.82.1-next.3 → 7.82.1-next.6

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@deliciousmonster/datadog-agent-binary",
-	"version": "7.82.1-next.3",
+	"version": "7.82.1-next.6",
 	"description": "Harper component that runs the Datadog trace-agent and core-agent alongside a node, with pre-built binaries installed per platform",
 	"type": "module",
 	"scripts": {
@@ -38,7 +38,7 @@
 	"author": "",
 	"license": "Apache-2.0",
 	"dependencies": {
-		"@deliciousmonster/harper-process-guard": "0.1.0-next.2"
+		"@deliciousmonster/harper-process-guard": "0.1.0-next.3"
 	},
 	"devDependencies": {
 		"@types/node": "^26.0.0",
@@ -68,10 +68,10 @@
 	},
 	"homepage": "https://github.com/deliciousmonster/datadog-agent-binary#readme",
 	"optionalDependencies": {
-		"@deliciousmonster/datadog-agent-binary-linux-x86_64": "7.82.1-next.3",
-		"@deliciousmonster/datadog-agent-binary-linux-arm64": "7.82.1-next.3",
-		"@deliciousmonster/datadog-agent-binary-macos-arm64": "7.82.1-next.3",
-		"@deliciousmonster/datadog-agent-binary-windows-x86_64": "7.82.1-next.3"
+		"@deliciousmonster/datadog-agent-binary-linux-x86_64": "7.82.1-next.6",
+		"@deliciousmonster/datadog-agent-binary-linux-arm64": "7.82.1-next.6",
+		"@deliciousmonster/datadog-agent-binary-macos-arm64": "7.82.1-next.6",
+		"@deliciousmonster/datadog-agent-binary-windows-x86_64": "7.82.1-next.6"
 	},
 	"lint-staged": {
 		"*.{ts,js,json}": [

package/runtime/config.js

@@ -78,6 +78,12 @@ function renderDatadogYaml(paths, ports) {
 		"# Pinned because the core-agent verify reads expvar off this port. Measured on 7.82.1: the environment",
 		"# outranks this file, so a set DD_EXPVAR_PORT wins and this only pins the port against a default that moves.",
 		`expvar_port: ${ports.expvar}`,
+		"# Live Processes: every process on the node with its CPU, memory and command line, Harper and the agents",
+		"# included. Go, so it runs in this Python-free build; DD_PROCESS_CONFIG_PROCESS_COLLECTION_ENABLED=false",
+		"# turns it off, since the environment outranks this file.",
+		"process_config:",
+		"  process_collection:",
+		"    enabled: true",
 		"apm_config:",
 		"  enabled: true",
 		`  receiver_port: ${ports.receiver}`,
@@ -94,16 +100,27 @@ function renderDatadogYaml(paths, ports) {
 
 const HARPER_LOG_CHECK = "harper.d";
 
-/** A logs source for Harper's own log file: `logging.root` is `log` under the root path unless a node moved it. */
-function renderHarperLogSource(logFile) {
+/**
+ * Log sources for what runs on this node: every log Harper writes under its log directory (`logging.root`
+ * is `log` under the root path unless a node moved it; the HTTP request log `http.logging` enables lands
+ * there too, under its own name), and the two agents' and the reaper's, which are the first thing an
+ * operator wants when the node stops reporting. Tailed only while logs are enabled (DD_LOGS_ENABLED=true).
+ */
+function renderLogSources(harperLog, paths) {
+	const source = (path, service, source) => [
+		"  - type: file",
+		`    path: ${yamlString(path)}`,
+		`    service: ${service}`,
+		`    source: ${source}`,
+	];
 	return [
 		"# GENERATED by resources.js on every Harper worker start. Edits are overwritten.",
 		"# Tailed only while logs are enabled (DD_LOGS_ENABLED=true).",
 		"logs:",
-		"  - type: file",
-		`    path: ${yamlString(logFile)}`,
-		"    service: harper",
-		"    source: harper",
+		...source(join(dirname(harperLog), "*.log"), "harper", "harper"),
+		...source(paths.coreLog, "datadog-agent", "datadog-agent"),
+		...source(paths.traceLog, "datadog-trace-agent", "datadog-agent"),
+		...source(paths.reaperLog, "datadog-agent-reaper", "harper-process-guard"),
 		"",
 	].join("\n");
 }
@@ -179,11 +196,12 @@ export function prepareRuntime(componentDir, { ports, log }) {
 				check.body;
 		}
 		const owned = new Set(checks.map((check) => check.dir));
-		// Harper's own log as a log source, written whenever the root is known. The agent tails it only
-		// when logs are on, which is DD_LOGS_ENABLED=true in the environment; off, this file costs nothing.
+		// Harper's own log and the agents' as log sources, written whenever the root is known. The agent
+		// tails them only when logs are on, which is DD_LOGS_ENABLED=true in the environment; off, this
+		// file costs nothing.
 		if (root) {
 			configFiles[join(paths.confd, HARPER_LOG_CHECK, "conf.yaml.default")] =
-				renderHarperLogSource(join(root, "log", "hdb.log"));
+				renderLogSources(join(root, "log", "hdb.log"), paths);
 			owned.add(HARPER_LOG_CHECK);
 		}
 		removeStaleDefaults(paths.confd, owned);

package/runtime/supervisor.js

@@ -1,11 +1,17 @@
 // Who holds the agents up. Released Harper has no sidecar API, so the bundled guard is the other half; one
 // agent per node comes from a PID lock either way and only the holder changes.
 
+import { readFileSync, unlinkSync } from "node:fs";
 import { join } from "node:path";
 
 // Pinned to one exact version, never a range: Harper runs `npm install` when it installs a component, so a
 // caret here would let a customer's node resolve a guard no test in this repo has run against.
-import { fingerprint, guard } from "@deliciousmonster/harper-process-guard";
+import {
+	argvOf,
+	fingerprint,
+	guard,
+	identify as identifyPid,
+} from "@deliciousmonster/harper-process-guard";
 import { describeSpawnFailure } from "./agent-exit.js";
 import { writeConfigFiles } from "./config.js";
 
@@ -99,12 +105,66 @@ const harperSupervisor = (scope, log) => ({
 // directory would collide on; this one names the package rather than taking the guard's generic default.
 const REAPER_NAME = "datadog-agent-reaper";
 
+/**
+ * Harper's own spawn keeps a pid file per process name under <root>/pids and, when the file names a pid
+ * that answers kill(pid, 0), returns that pid instead of spawning. After a restart the kernel reissues
+ * pids, and a thread of Harper itself answers for one, so the file has to go before the guard asks.
+ * Removing it signals nothing. A file naming the real process, or a dead one, is Harper's to keep.
+ *
+ * @param {string | null} root @param {Array<{ name: string; argv?: readonly string[]; script?: string }>} named
+ * @param {import("./log.js").Log} log
+ */
+export function clearStaleHarperPidFiles(root, named, log) {
+	if (!root) return;
+	for (const { name, argv, script } of named) {
+		const file = join(root, "pids", `${name}.pid`);
+		let pid;
+		try {
+			pid = Number.parseInt(readFileSync(file, "utf-8"), 10);
+		} catch {
+			continue;
+		}
+		if (!Number.isInteger(pid) || pid <= 0) continue;
+		const running = argvOf(pid);
+		if (running === null) continue;
+		const ours = argv
+			? identifyPid(pid, argv) === "match"
+			: running.some((argument) => argument.endsWith(script ?? "\u0000"));
+		if (ours) continue;
+		try {
+			unlinkSync(file);
+			log.warn(
+				`Datadog supervisor: removed ${file}, Harper's own pid file for ${name}: it named pid ${pid}, which ` +
+					`is running \`${running.join(" ")}\`, and Harper would have handed that pid back as the ${name} ` +
+					`instead of starting one.`
+			);
+		} catch (error) {
+			log.error(
+				`Datadog supervisor: could not remove ${file}, which names pid ${pid} running something else: ` +
+					`${error.message}. Harper will hand that pid back as the ${name} rather than start one.`
+			);
+		}
+	}
+}
+
 /** The bundled guard, one call for both agents. `spawn` is the entry module's own, which is the one Harper constrains. */
 const guardSupervisor = (log, spawn) => ({
 	kind: "guard",
 	async start(agents, { runtime, configFiles, fingerprintParts }) {
 		// Harper's start() writes these itself; on this path nothing else will, and both agents read them.
 		writeConfigFiles(configFiles, log);
+		clearStaleHarperPidFiles(
+			runtime.root,
+			[
+				...agents.map((agent) => ({
+					name: agent.name,
+					argv: [agent.command, ...agent.args],
+				})),
+				// The guard builds the reaper's own argv; its script name is the one stable thing to match on.
+				{ name: REAPER_NAME, script: "/reaper.js" },
+			],
+			log
+		);
 		let result;
 		try {
 			result = await guard({