gitdone-agent 0.6.14 → 0.6.16

package/index.js

@@ -27,7 +27,7 @@ import { randomUUID } from 'node:crypto'
 // Reported to the server on every sync so the web UI can flag outdated agents.
 // Keep in lockstep with packages/agent/package.json "version" AND
 // src/lib/agentVersion.ts LATEST_AGENT_VERSION.
-const AGENT_VERSION = '0.6.14'
+const AGENT_VERSION = '0.6.16'
 
 const AGENT_DIR = join(homedir(), '.gitdone-agent')
 const CONFIG_PATH = join(AGENT_DIR, 'config.json')
@@ -186,10 +186,52 @@ function findClaude() {
   return { path: 'claude', shell: true, found: false }
 }
 
+// ─── Single instance (gd-407) ─────────────────────────────────────────────────
+// Two agents on one machine (an old one surviving an update, or a manual
+// `npx gitdone-agent` next to the autostarted one) flap the machine's online
+// state and race dispatches. The running agent claims the machine via a pid
+// file; a newcomer that finds a LIVE agent behind it exits with EXIT_DUPLICATE,
+// which the supervisor treats as "do not restart me". First one wins.
+// 86 is outside node's own exit-code range (1-13) — keep in lockstep with the
+// run-agent.cmd template in installStartup().
+const EXIT_DUPLICATE = 86
+const PID_PATH = join(AGENT_DIR, 'agent.pid')
+
+// Command line of a live process, '' when unreadable (dead process, no rights).
+function processCmdline(pid) {
+  try {
+    if (process.platform === 'win32') {
+      return execSync(
+        `powershell -NoProfile -NonInteractive -Command "(Get-CimInstance Win32_Process -Filter 'ProcessId=${pid}').CommandLine"`,
+        { encoding: 'utf8' },
+      ).trim()
+    }
+    return execSync(`ps -p ${pid} -o command=`, { encoding: 'utf8' }).trim()
+  } catch { return '' }
+}
+
+function ensureSingleInstance() {
+  try {
+    const pid = parseInt(readFileSync(PID_PATH, 'utf8'), 10)
+    if (pid && pid !== process.pid) {
+      process.kill(pid, 0) // throws when that pid is no longer alive
+      // PID reuse guard: only defer to a process that really looks like an
+      // agent (stable copy agent.mjs, or any gitdone-agent npx/global run).
+      if (/gitdone-agent|agent\.mjs/i.test(processCmdline(pid))) {
+        log(`✗ друг gitdone-agent вече върви (PID ${pid}) — този процес излиза, за да няма два агента на машината`)
+        process.exit(EXIT_DUPLICATE)
+      }
+    }
+  } catch { /* no/stale pid file → the machine is free, we take over */ }
+  try { ensureAgentDir(); writeFileSync(PID_PATH, String(process.pid)) } catch { /* best-effort */ }
+}
+
 // Kill any previously-running agent processes so an update applies WITHOUT a
-// Windows re-login. Targets node processes launched from the agent dir
-// (~/.gitdone-agent), excluding ourselves — the npx installer runs from the
-// npx cache, so it won't match. Best-effort; failures are non-fatal.
+// Windows re-login. Targets node processes that look like a gitdone agent —
+// both the stable copy in ~/.gitdone-agent AND one running straight from the
+// npx cache / global install (its command line contains "gitdone-agent") —
+// excluding ourselves and our parent (the npx wrapper that launched us).
+// Best-effort; failures are non-fatal.
 function stopRunningAgents() {
   // Kill the supervisor loops (cmd.exe running run-agent.cmd) FIRST, then the
   // agents — the other order lets a still-alive supervisor immediately respawn
@@ -200,7 +242,7 @@ function stopRunningAgents() {
     "  Where-Object { $_.CommandLine -like '*run-agent.cmd*' } |",
     '  ForEach-Object { Stop-Process -Id $_.ProcessId -Force }',
     "Get-CimInstance Win32_Process -Filter \"Name='node.exe'\" |",
-    `  Where-Object { $_.CommandLine -like '*.gitdone-agent*' -and $_.ProcessId -ne ${process.pid} } |`,
+    `  Where-Object { ($_.CommandLine -like '*gitdone-agent*' -or $_.CommandLine -like '*agent.mjs*') -and $_.ProcessId -ne ${process.pid} -and $_.ProcessId -ne ${process.ppid || 0} } |`,
     '  ForEach-Object { Stop-Process -Id $_.ProcessId -Force }',
   ].join('\n')
   try {
@@ -236,6 +278,14 @@ function installStartup() {
   // node-level failures (corrupt file, bad path) finally leave a trace. If the
   // node path recorded at install time vanished (node upgraded/moved), fall
   // back to whatever `node` is on PATH.
+  //
+  // Restart forever on normal crashes, but STOP on process-START failures
+  // (0xc0000142 STATUS_DLL_INIT_FAILED, 0xc0000135 missing DLL): those mean the
+  // login session itself is broken (e.g. after a hardware crash), every attempt
+  // pops a blocking "Application Error" dialog, and a retry can never succeed —
+  // the loop turned into an endless dialog storm (gd-405). Same for the
+  // ping-as-sleep: if even ping can't start, bail instead of spinning with no
+  // delay at all. A Windows re-login restarts the supervisor cleanly.
   const crashLog = join(AGENT_DIR, 'agent-crash.log')
   const cmd = [
     '@echo off',
@@ -244,9 +294,18 @@ function installStartup() {
     'if not exist "%NODE%" set "NODE=node"',
     ':loop',
     `"%NODE%" "${agentPath}" 2>> "${crashLog}"`,
-    `echo [%date% %time%] agent exited (code %errorlevel%) - restart in 10s >> "${crashLog}"`,
-    'ping -n 11 127.0.0.1 >nul',
+    'set CODE=%errorlevel%',
+    `echo [%date% %time%] agent exited (code %CODE%) - restart in 10s >> "${crashLog}"`,
+    'if "%CODE%"=="86" goto duplicate',
+    'if "%CODE%"=="-1073741502" goto dead',
+    'if "%CODE%"=="-1073741515" goto dead',
+    'ping -n 11 127.0.0.1 >nul 2>nul || goto dead',
     'goto loop',
+    ':dead',
+    `echo [%date% %time%] node cannot start (code %CODE%) - supervisor giving up until next login >> "${crashLog}"`,
+    'exit /b',
+    ':duplicate',
+    `echo [%date% %time%] another agent already runs this machine (code 86) - this supervisor exits >> "${crashLog}"`,
   ].join('\r\n')
   const cmdPath = join(AGENT_DIR, 'run-agent.cmd')
   writeFileSync(cmdPath, cmd, 'utf8')
@@ -1590,6 +1649,7 @@ async function streamCommands(cfg) {
 // ─── Main ────────────────────────────────────────────────────────────────────
 
 async function runLoop(cfg) {
+  ensureSingleInstance()
   log(`gitdone-agent started — machine: ${cfg.hostname}, server: ${cfg.url}, interval: ${cfg.interval}s`)
   log(`  roots: ${cfg.roots.join(', ') || '(none — add one with --root)'}`)
 

package/package.json

@@ -1,6 +1,6 @@
 {
     "name":  "gitdone-agent",
-    "version":  "0.6.14",
+    "version":  "0.6.16",
     "description":  "Local git agent for gitdone — watches a local repo and sends snapshots to gitdone.eu",
     "type":  "module",
     "bin":  {