@volter/supercode-orchestrator 0.4.11 → 0.4.13

package/bin/orchestrator.mjs

@@ -69,6 +69,9 @@ import * as events from '../events.mjs';
 import { httpJobExecutor } from '../hosted.mjs';
 import { SupercodeHarnessClient } from '@volter/supercode-harness-sdk';
 
+/** How long a starting daemon waits for a stopping predecessor to release the home's lease. */
+const LEASE_HANDOFF_MS = 20_000;
+
 /// Adapter selector -> constructor. A `platforms:` block picks its adapter by
 /// `extra.adapter` when it has one and by the block's own name otherwise, so
 /// `platforms.slack.extra.adapter: chat-sdk` and `platforms.api_server` both
@@ -596,6 +599,13 @@ async function main() {
   if (!args.once) {
     // Serialize lease replacement before loading state. An acquisition guard
     // left by a crash fails closed; inspect it before removing it manually.
+    // A service manager's restart starts this one while the last is still stopping (launchd's kickstart -k did, and
+    // this one exited 'already serving'): a live predecessor gets a moment to let go before it counts as serving.
+    for (let waited = 0; waited < LEASE_HANDOFF_MS; waited += 250) {
+      const previous = readLock(root);
+      if (!previous || previous.pid === process.pid || !leaseIsLive(previous)) break;
+      await new Promise((resolve) => setTimeout(resolve, 250));
+    }
     const guard = join(root, 'orchestrator.acquire');
     mkdirSync(guard);
     try {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@volter/supercode-orchestrator",
-  "version": "0.4.11",
+  "version": "0.4.13",
   "type": "module",
   "description": "The orchestrator runtime over the Volter Harness ontology: one typed operational model whose folder is its serialization, read and written through the harness orchestration doors (docs/ORCHESTRATOR-IR.md)",
   "exports": {

package/reducer.mjs

@@ -549,8 +549,8 @@ function handleTick(state, ev, effects) {
 // ---------------------------------------------------------------- restart / adapter / operator
 
 // §4.7 on a clean stop: the daemon closes every runtime, so each binding that held one keeps it as the runtime to
-// resume, as a crash's pass marks it, and the next line reopens the same session. A turn cut off mid-way is told so
-// on its surface; an idle binding is not.
+// resume, as a crash's pass marks it, and the restart reopens the same session (a clean stop is a restart's). A turn
+// cut off mid-way is told so on its surface; an idle binding is not.
 function handleStopping(state, ev, effects) {
   const rt = ensureRuntime(state);
   for (const profile of Object.values(state.profiles)) {
@@ -558,6 +558,8 @@ function handleStopping(state, ev, effects) {
       if (b.ended_at || ks.includes('#') || !b.residue?.runtime_id) continue;
       b.residue.resume_runtime_id = b.residue.runtime_id;
       delete b.residue.runtime_id;
+      // a stop for a restart: what was live now is live again once the daemon is back (handleRestart)
+      if (b.key.platform !== 'webhook' && b.key.platform !== 'cron') b.residue.reopen_at_start = true;
       if (rt.turns[leaseId(profile.name, ks)] && !b.recurrence && b.key.platform !== 'webhook' && b.key.platform !== 'cron') {
         b.residue.interrupted_at = ev.at;
         reply(state, profile, b.key, REPLY.interrupted, ev.at, effects);
@@ -579,7 +581,16 @@ function handleRestart(state, ev, effects) {
     // attaches where the worker's door allows it and marks the rest
     // interrupted (`binding_interrupted` below).
     for (const [ks, b] of Object.entries(profile.bindings)) {
-      if (b.ended_at || ks.includes('#') || !b.residue?.runtime_id) continue;
+      if (b.ended_at || ks.includes('#')) continue;
+      // one the clean stop closed reopens its own session now, so its conversation stays reachable (a message, a
+      // voice line) rather than waiting for its next inbound
+      if (!b.residue?.runtime_id) {
+        if (!b.residue?.reopen_at_start) continue;
+        delete b.residue.reopen_at_start;
+        live += 1;
+        startWorker(state, profile, ks, effects);
+        continue;
+      }
       live += 1;
       // a message arriving while the attach is in flight queues behind it,
       // rather than starting a fresh runtime that would replace the session