fix: bot wakes worker after job insert, increase idle timeout to 15min (#22)

The Telegram bot now pings the worker's /health endpoint after every
insertJob call, triggering Fly.io auto-start if the machine is stopped.
Also increases worker idle shutdown from 5 to 15 minutes.

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: abrichr

apps/bot/src/index.ts

@@ -35,6 +35,24 @@ if (!GITHUB_TOKEN) {
   process.exit(1)
 }
 
+const WORKER_URL = process.env.WORKER_URL || 'https://wright-worker.fly.dev'
+
+/**
+ * Wake the worker by hitting its health endpoint.
+ * This triggers Fly.io auto-start if the machine is stopped.
+ * Errors are silently ignored — the request just needs to reach Fly's proxy.
+ */
+async function wakeWorker(): Promise<void> {
+  try {
+    await fetch(`${WORKER_URL}/health`, { signal: AbortSignal.timeout(5000) })
+    console.log('[wake] Worker pinged successfully')
+  } catch {
+    // Ignore — the request just needs to trigger Fly.io auto-start.
+    // The machine may take a few seconds to boot, so a timeout is expected.
+    console.log('[wake] Worker ping sent (may be booting)')
+  }
+}
+
 // ---------------------------------------------------------------------------
 // Bot setup
 // ---------------------------------------------------------------------------
@@ -349,6 +367,9 @@ bot.command('revise', async (ctx: Context) => {
       parentJobId: originalJob.id,
     })
 
+    // Wake the worker so it picks up the revision job
+    wakeWorker()
+
     await ctx.reply(
       [
         '\u{1F504} <b>Revision queued!</b>',
@@ -441,6 +462,9 @@ bot.command('task', async (ctx: Context) => {
         parentJobId,
       })
 
+      // Wake the worker so it picks up the PR revision job
+      wakeWorker()
+
       await ctx.reply(
         [
           '\u{1F504} <b>PR revision queued!</b>',
@@ -495,6 +519,9 @@ bot.command('task', async (ctx: Context) => {
       githubToken: GITHUB_TOKEN,
     })
 
+    // Wake the worker so it picks up the new job
+    wakeWorker()
+
     await ctx.reply(
       [
         '\u{2705} <b>Job queued!</b>',
@@ -666,6 +693,9 @@ bot.on('message:text', async (ctx, next) => {
       parentJobId: originalJob.id,
     })
 
+    // Wake the worker so it picks up the revision job
+    wakeWorker()
+
     await ctx.reply(
       [
         '\u{1F504} <b>Revision queued from reply!</b>',

apps/worker/src/index.ts

@@ -19,15 +19,15 @@ const FLY_APP_NAME = process.env.FLY_APP_NAME || 'wright-worker'
 // Track active jobs with AbortControllers for cancellation
 const runningJobs = new Map<string, AbortController>()
 let activeJobs = 0
-const IDLE_SHUTDOWN_MS = 5 * 60 * 1000 // 5 min idle → exit (scale-to-zero)
+const IDLE_SHUTDOWN_MS = 15 * 60 * 1000 // 15 min idle → exit (scale-to-zero)
 let idleTimer: ReturnType<typeof setTimeout> | null = null
 let keepAliveInterval: ReturnType<typeof setInterval> | null = null
 
 function resetIdleTimer() {
   if (idleTimer) clearTimeout(idleTimer)
   if (activeJobs === 0) {
     idleTimer = setTimeout(() => {
-      console.log('No active jobs for 5 minutes, shutting down')
+      console.log('No active jobs for 15 minutes, shutting down')
       process.exit(0)
     }, IDLE_SHUTDOWN_MS)
   }