fix(cli): stop install-service writing a unit that dies on upgrade and reboot

apps/cli/src/commands/__tests__/service.test.ts

@@ -0,0 +1,54 @@
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
+import { describe, expect, it } from "vitest";
+import { stableBinPath } from "../service.js";
+
+const UNIT = readFileSync(
+  join(__dirname, "..", "..", "systemd", "threatcrushd.service"),
+  "utf-8",
+);
+
+describe("stableBinPath", () => {
+  // The bug this file exists for: `install-service` baked the version-stamped
+  // pnpm path into ExecStart, so the next upgrade left the unit pointing at a
+  // directory that no longer existed. On one host that was 6315 failed starts
+  // against a path last valid at 0.2.1.
+  it("rewrites a version-stamped pnpm path to the stable symlink", () => {
+    const root = "/home/ubuntu/.local/share/pnpm/global/5";
+    const versioned = `${root}/.pnpm/@profullstack+threatcrush@0.11.6/node_modules/@profullstack/threatcrush/dist/index.js`;
+
+    // The rewrite only takes effect when the stable path exists on disk, which
+    // it does not here, so we assert the shape of the rewrite itself.
+    const rewritten = versioned.replace(/\/\.pnpm\/[^/]+\/node_modules\//, "/node_modules/");
+    expect(rewritten).toBe(`${root}/node_modules/@profullstack/threatcrush/dist/index.js`);
+    expect(rewritten).not.toMatch(/@\d+\.\d+\.\d+/);
+  });
+
+  it("leaves a path with no .pnpm segment untouched", () => {
+    // npm's global layout is already version-free.
+    const npmPath = "/usr/lib/node_modules/@profullstack/threatcrush/dist/index.js";
+    expect(stableBinPath(npmPath)).toBe(npmPath);
+  });
+
+  it("falls back to the original path when the stable one is absent", () => {
+    const versioned =
+      "/nonexistent/.pnpm/@profullstack+threatcrush@0.11.6/node_modules/@profullstack/threatcrush/dist/index.js";
+    expect(stableBinPath(versioned)).toBe(versioned);
+  });
+});
+
+describe("systemd unit template", () => {
+  // /var/run is a symlink to the /run tmpfs, so a directory created at install
+  // time is gone after the next reboot — and a missing ReadWritePaths entry
+  // fails the unit 226/NAMESPACE on every start.
+  it("does not list a tmpfs path under ReadWritePaths", () => {
+    const readWrite = UNIT.split("\n").find((l) => l.startsWith("ReadWritePaths="));
+    expect(readWrite).toBeDefined();
+    expect(readWrite).not.toContain("/var/run/");
+    expect(readWrite).not.toContain("/run/");
+  });
+
+  it("lets systemd create the runtime directory on each start", () => {
+    expect(UNIT).toMatch(/^RuntimeDirectory=threatcrush$/m);
+  });
+});

apps/cli/src/commands/service.ts

@@ -14,10 +14,25 @@ function resolveTemplate(): string {
   return readFileSync(templatePath, 'utf-8');
 }
 
+// pnpm installs a global package under a version-stamped directory
+// (`.pnpm/@profullstack+threatcrush@0.11.6/node_modules/…`) and points a stable
+// symlink at it from the store root. Baking the version-stamped path into
+// ExecStart means the very next upgrade leaves the unit pointing at a directory
+// that no longer exists — the unit then fails on every start, forever, and
+// `Restart=on-failure` turns that into a permanent loop. Observed in the wild at
+// 6315 restarts against a path last valid at 0.2.1.
+export function stableBinPath(binPath: string): string {
+  const stable = binPath.replace(/\/\.pnpm\/[^/]+\/node_modules\//, '/node_modules/');
+  if (stable === binPath) return binPath;
+  // Only take the rewrite if the symlink is really there; a layout we have not
+  // seen is better served by the path we were actually invoked with.
+  return existsSync(stable) ? stable : binPath;
+}
+
 function resolveBinPath(): string {
   // When installed globally the script path is the CLI bin.
   const arg = process.argv[1];
-  if (arg && existsSync(arg)) return arg;
+  if (arg && existsSync(arg)) return stableBinPath(arg);
   try {
     return execSync('command -v threatcrush', { encoding: 'utf-8' }).trim();
   } catch {
@@ -74,7 +89,9 @@ function ensureSystemDirs(): void {
     { path: '/etc/threatcrush/threatcrushd.conf.d', groupWritable: true },
     { path: '/var/log/threatcrush', groupWritable: false },
     { path: '/var/lib/threatcrush', groupWritable: false },
-    { path: '/var/run/threatcrush', groupWritable: false },
+    // /run/threatcrush is deliberately absent: it lives on a tmpfs, so creating
+    // it here only lasts until the next reboot. The unit's RuntimeDirectory=
+    // recreates it on every start instead.
   ];
   let admGid: number | null = null;
   try {

apps/cli/src/systemd/threatcrushd.service

@@ -17,7 +17,13 @@ TimeoutStopSec=10
 NoNewPrivileges=true
 ProtectSystem=full
 ProtectHome=read-only
-ReadWritePaths=/etc/threatcrush /var/log/threatcrush /var/lib/threatcrush /var/run/threatcrush
+ReadWritePaths=/etc/threatcrush /var/log/threatcrush /var/lib/threatcrush
+# /var/run is a symlink to /run, which is a tmpfs — anything created there at
+# install time is gone after the next reboot. Listing it under ReadWritePaths
+# then makes the unit fail 226/NAMESPACE on every start. RuntimeDirectory has
+# systemd create (and clean up) /run/threatcrush itself, on each start.
+RuntimeDirectory=threatcrush
+RuntimeDirectoryMode=0755
 PrivateTmp=true
 
 # Log directly to journald