@zgeoff/atc 0.1.3 → 0.1.5

package/README.md

@@ -121,3 +121,8 @@ live fleet (name, cwd, Claude session id) to its SQLite store. If the daemon its
 SIGKILL, reboot — the child claude processes die with it, but every session's transcript is already
 on disk. Start atc and press `R`: the whole fleet respawns via `claude --resume`. Only deliberate
 kills (`K`, `Y` eject) remove entries from the fleet, so it stays restorable.
+
+Restoring revives one session at a time: the next `claude --resume` starts only once the previous
+one has reported it is up (its `SessionStart` hook), so bringing back a dozen sessions no longer
+launches a dozen Claude processes at the same instant and pins the machine. The list fills in live
+as each comes back.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@zgeoff/atc",
-  "version": "0.1.3",
+  "version": "0.1.5",
   "description": "Terminal control tower for Claude Code sessions",
   "homepage": "https://github.com/zgeoff/atc#readme",
   "bugs": "https://github.com/zgeoff/atc/issues",

package/src/cli.ts

@@ -50,16 +50,26 @@ const main = defineCommand({
           // Test harnesses shrink the outbound queue to force overflow
           // deterministically; unset means the production default.
           const queueBytes = Number(process.env['ATC_QUEUE_BYTES']);
+          const cfg = config.loadConfig();
+
+          // Cap on how long a fleet restore waits for one revived session to
+          // report it has booted before moving to the next. Tests pin it to
+          // keep timing deterministic; unset means the production default.
+          const capOverride = Number(process.env['ATC_RESTORE_BOOT_TIMEOUT_MS']);
+
+          const restoreBootTimeoutMs =
+            Number.isFinite(capOverride) && capOverride >= 0 ? capOverride : 15_000;
 
           const handle = daemon.startDaemon({
             headlessRunner: (runOpts, hooks) => headless.startHeadlessRun(runOpts, hooks),
             socketPath: config.daemonSocketPath,
             reporterSocketPath: config.socketPath,
             build: getBuild(),
-            adapter: new claude.ClaudeAdapter(config.loadConfig()),
+            adapter: new claude.ClaudeAdapter(cfg),
             dbPath: config.dbFile,
             legacyFleetPath: config.legacyFleetFile,
             pidPath: config.daemonPidFile,
+            restoreBootTimeoutMs,
             ...(Number.isFinite(queueBytes) && queueBytes > 0 ? { queueBytes } : {}),
             onQuit: () => process.exit(0),
           });

package/src/daemon.ts

@@ -10,7 +10,7 @@ import { MAX_CHUNK, PROTOCOL_V } from './protocol';
 import type { EventMsg } from './protocol';
 import { ScreenModel } from './screen-model';
 import { SessionManager } from './sessions';
-import type { SessionDescriptor, SessionState } from './sessions';
+import type { FleetEntry, Session, SessionDescriptor, SessionState } from './sessions';
 import { StateStore } from './state-store';
 
 export interface DaemonOptions {
@@ -44,6 +44,13 @@ export interface DaemonOptions {
   // before starting the headless run anyway.
   readonly ejectSettleMs?: number;
 
+  // A fleet-wide restore revives one session at a time, waiting for each to
+  // report it has booted before starting the next so the machine is not
+  // buried under a dozen simultaneous agent boots. This caps how long a
+  // single revive waits for that signal before moving on regardless, so a
+  // session that never reports cannot stall the rest. Zero waits forever.
+  readonly restoreBootTimeoutMs?: number;
+
   // Called after a client-requested quit has stopped the daemon; the real
   // entrypoint exits the process, tests leave it unset.
   readonly onQuit?: () => void;
@@ -126,6 +133,37 @@ export function startDaemon(opts: DaemonOptions): DaemonHandle {
   // (or a settle timeout) before the headless run starts.
   const pendingEjects = new Map<string, () => void>();
 
+  // A staggered fleet restore parks a resolver here while it waits for the
+  // session it just revived to report it has booted; the reporter fires it on
+  // SessionStart, and a dying revive fires it too so a failed resume does not
+  // hold up the rest. Resolving a settled promise again is a no-op.
+  const bootWaiters = new Map<string, () => void>();
+
+  // Blocks until the session reports SessionStart, it dies, or the cap
+  // elapses (a positive cap only; zero waits on the signal alone).
+  const waitForBoot = (sessionID: string, capMs: number): Promise<void> => {
+    const settled = Promise.withResolvers<void>();
+    const timer = capMs > 0 ? setTimeout(settled.resolve, capMs) : undefined;
+
+    bootWaiters.set(sessionID, settled.resolve);
+
+    if (timer !== undefined) {
+      restoreTimers.add(timer);
+    }
+
+    return (async () => {
+      await settled.promise;
+
+      bootWaiters.delete(sessionID);
+
+      if (timer !== undefined) {
+        clearTimeout(timer);
+
+        restoreTimers.delete(timer);
+      }
+    })();
+  };
+
   const startHeadlessTurn = (sessionID: string, prompt: string): boolean => {
     const runner = opts.headlessRunner;
     const s = mgr.sessions.find((x) => x.id === sessionID);
@@ -172,6 +210,7 @@ export function startDaemon(opts: DaemonOptions): DaemonHandle {
   const screens = new Map<string, ScreenModel>();
   const resizeTimers = new Map<string, ReturnType<typeof setTimeout>>();
   const detectTimers = new Map<string, ReturnType<typeof setTimeout>>();
+  const restoreTimers = new Set<ReturnType<typeof setTimeout>>();
 
   // The screen tier of the detector stack: once a session's output has
   // quiesced, judge the serialized screen and flip running/needs_you.
@@ -345,6 +384,12 @@ export function startDaemon(opts: DaemonOptions): DaemonHandle {
   mgr.onEvent = (kind, s) => {
     recordAttention(kind, s);
 
+    // A revive that dies before it ever announces itself must still release
+    // the staggered restore, or a failed resume would hold up the fleet.
+    if (kind === 'removed' || (kind === 'state' && s.pty === null)) {
+      bootWaiters.get(s.id)?.();
+    }
+
     if (kind === 'removed') {
       attachments.removeSession(s.id);
       seqs.delete(s.id);
@@ -397,6 +442,12 @@ export function startDaemon(opts: DaemonOptions): DaemonHandle {
       pendingEjects.get(e.atcId)?.();
     }
 
+    // A revived session announcing itself is the cue a staggered restore
+    // waits on before booting the next one.
+    if (e.event === 'SessionStart') {
+      bootWaiters.get(e.atcId)?.();
+    }
+
     mgr.applyHook(e);
   }, opts.reporterSocketPath);
 
@@ -563,13 +614,12 @@ export function startDaemon(opts: DaemonOptions): DaemonHandle {
     getEffectiveDims: (sessionID) =>
       attachments.findEffectiveDims(sessionID) ?? ptyDims.get(sessionID) ?? { cols: 80, rows: 24 },
     restoreFleet: (cols, rows) => {
-      let restored = 0;
+      const isLive = (claudeId: string | undefined) =>
+        mgr.sessions.some((s) => s.pty !== null && s.claudeId === claudeId);
 
-      for (const entry of store.loadFleet()) {
-        const live = mgr.sessions.some((s) => s.pty !== null && s.claudeId === entry.claudeId);
-
-        if (live) {
-          continue;
+      const restoreEntry = (entry: FleetEntry): Session | null => {
+        if (isLive(entry.claudeId)) {
+          return null;
         }
 
         const s = mgr.spawn(
@@ -586,10 +636,37 @@ export function startDaemon(opts: DaemonOptions): DaemonHandle {
         ptyDims.set(s.id, { cols, rows });
         screens.set(s.id, new ScreenModel(cols, rows));
 
-        restored++;
+        return s;
+      };
+
+      const [first, ...rest] = store.loadFleet().filter((entry) => !isLive(entry.claudeId));
+
+      if (first === undefined) {
+        return 0;
       }
 
-      return restored;
+      // The first revive is synchronous so a caller can attach at once; each
+      // later revive waits for the previous session to report it has booted,
+      // so a heavy fleet comes up one process at a time instead of all at
+      // once. A per-session cap keeps a session that never reports from
+      // stalling the rest.
+      const cap = opts.restoreBootTimeoutMs ?? 0;
+
+      const restoreRest = async (previous: Session | null) => {
+        let prev = previous;
+
+        for (const entry of rest) {
+          if (prev !== null) {
+            await waitForBoot(prev.id, cap);
+          }
+
+          prev = restoreEntry(entry);
+        }
+      };
+
+      void restoreRest(restoreEntry(first));
+
+      return 1 + rest.length;
     },
   };
 
@@ -628,6 +705,10 @@ export function startDaemon(opts: DaemonOptions): DaemonHandle {
       clearTimeout(timer);
     }
 
+    for (const timer of restoreTimers) {
+      clearTimeout(timer);
+    }
+
     server.stop(true);
     reporter.stop(true);
     mgr.killAll();

package/src/index.ts

@@ -4,7 +4,7 @@ import { loadConfig } from './config';
 import { collectDirs, findFuzzyScore, formatDir, pickMatches } from './dirs';
 import type { EventMsg } from './protocol';
 import { isRecord } from './report';
-import { countSessionStates, sortSessionViews } from './sessions';
+import { countSessionStates, sortGroupedSessionViews, sortSessionViews } from './sessions';
 import type { SessionState } from './sessions';
 import { ansi, cols, drawHelp, drawHome, drawOverlay, drawPicker, drawStatusBar, rows } from './ui';
 
@@ -195,7 +195,7 @@ function applyHelpKey(buf: Buffer) {
 let overlayFilter: string | null = null;
 
 function pickOverlaySessions(): MirrorSession[] {
-  const sorted = sortSessionViews(fleet);
+  const sorted = sortGroupedSessionViews(fleet);
 
   if (overlayFilter === null || overlayFilter === '') {
     return sorted;
@@ -229,7 +229,7 @@ function openOverlay() {
   confirmKill = false;
   overlayFilter = null;
 
-  const focusedIndex = sortSessionViews(fleet).findIndex((s) => s.id === focusedID);
+  const focusedIndex = sortGroupedSessionViews(fleet).findIndex((s) => s.id === focusedID);
 
   overlaySelected = Math.max(0, focusedIndex);
 

package/src/sessions.ts

@@ -613,3 +613,30 @@ export function sortSessionViews<
 
   return [...list].toSorted((a, b) => rank[a.state] - rank[b.state] || a.createdAt - b.createdAt);
 }
+
+// Overlay display order: the urgency sort with each group's sessions pulled
+// together at the position of its most urgent member, so the renderer's
+// adjacency-based headers appear once per group and urgent groups still lead.
+export function sortGroupedSessionViews<
+  T extends {
+    readonly state: SessionState;
+    readonly createdAt: number;
+    readonly group?: string;
+    readonly cwd: string;
+  },
+>(list: readonly T[]): T[] {
+  const buckets = new Map<string, T[]>();
+
+  for (const s of sortSessionViews(list)) {
+    const key = s.group ?? s.cwd;
+    const bucket = buckets.get(key);
+
+    if (bucket === undefined) {
+      buckets.set(key, [s]);
+    } else {
+      bucket.push(s);
+    }
+  }
+
+  return [...buckets.values()].flat();
+}