@nylorun/runtime 0.53.2-beta → 0.53.3-beta

package/CHANGELOG.md

@@ -1,5 +1,14 @@
 # Changelog
 
+## 0.53.3-beta
+
+### Patch Changes
+
+- f2ca0f1: A running computer gets the sweep's reconcile each minute its row rests, as a pod whose engine is
+  not connected does: a drain that moved it to another worker is recorded (`agent.suspended`,
+  reason `drain`) while a harness still holds its ticket, and a computer deleted outside the
+  Runtime is found lost.
+
 ## 0.53.2-beta
 
 ### Patch Changes

package/dist/management-openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.2.0",
   "info": {
     "title": "Nylorun Management API",
-    "version": "0.53.2-beta",
+    "version": "0.53.3-beta",
     "x-nylorun-protocol": {
       "min": 16,
       "max": 18,

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.2.0",
   "info": {
     "title": "Nylorun Runtime API",
-    "version": "0.53.2-beta",
+    "version": "0.53.3-beta",
     "x-nylorun-protocol": {
       "min": 16,
       "max": 18,

package/dist/tenant/sweep.d.ts

@@ -1,4 +1,4 @@
-import type { PendingRequest } from "../store/types.js";
+import type { PendingRequest, SandboxPodState } from "../store/types.js";
 import type { TenantContext } from "./context.js";
 import type { OutboxRequest } from "../store/types.js";
 export declare function sweep(ctx: TenantContext): Promise<void>;
@@ -28,3 +28,21 @@ export declare function wakeOrphanedSessions(ctx: Pick<TenantContext, "store" |
     woken: string[];
     next?: string;
 }>;
+/**
+ * Whether the sweep asks for a reconcile of `row`'s pod (`reconcileStalePods`): its row has not
+ * moved for `staleMs`, and its change is unfinished, its engine is not connected, or it runs
+ * past its idle limit. A running computer (ADR-149) always qualifies: nothing else looks at it
+ * while a harness holds a ticket, so the sweep is what finds it drained to another worker
+ * (`drainedAt`, recorded once) or deleted outside the Runtime.
+ */
+export declare function podWantsReconcile(row: {
+    readonly updatedAt: string;
+    readonly createdAt: string;
+    readonly pod?: SandboxPodState;
+}, facts: {
+    now: number;
+    staleMs: number;
+    idleMs: number;
+    computer: boolean;
+    hosting: boolean;
+}): boolean;

package/dist/tenant/sweep.js

@@ -316,28 +316,41 @@ export async function wakeOrphanedSessions(ctx, now = new Date(), after, budgetM
 async function reconcileStalePods(ctx, now) {
     if (!ctx.pods)
         return;
+    const computer = ctx.pods.computer !== undefined;
     const stale = await ctx.store.tx(async (t) => {
         const config = await readSandboxConfig(t);
-        return (await t.listEnvironments()).filter((row) => {
-            const pod = row.pod;
-            if (!pod || now.getTime() - Date.parse(row.updatedAt) < hostTuning().sweep.stalePodMs)
-                return false;
-            return (pod.desired === "deleted" ||
-                pod.retiring !== undefined ||
-                pod.observed === "creating" ||
-                pod.observed === "deleting" ||
-                (pod.desired === "suspended" && pod.observed !== "suspended") ||
-                // Running, but its engine is not connected: the pod may be gone, or its volume.
-                // (A computer has no engine connection, ADR-149.)
-                (pod.desired === "running" && pod.observed === "running" && !ctx.pods?.computer && !ctx.harness.hosting(row.id)) ||
-                // Running past its idle limit: its idle timer is missing (or a turn holds it, which the
-                // reconcile leaves alone).
-                (pod.desired === "running" &&
-                    pod.observed === "running" &&
-                    now.getTime() - Date.parse(pod.lastActiveAt ?? row.createdAt) >
-                        idleMsOf(config, row) + hostTuning().sweep.stalePodMs));
-        });
+        return (await t.listEnvironments()).filter((row) => podWantsReconcile(row, {
+            now: now.getTime(),
+            staleMs: hostTuning().sweep.stalePodMs,
+            idleMs: idleMsOf(config, row),
+            computer,
+            hosting: ctx.harness.hosting(row.id),
+        }));
     });
     for (const row of stale)
         await ctx.sandboxSignal(row.id, { kind: "reconcile" });
 }
+/**
+ * Whether the sweep asks for a reconcile of `row`'s pod (`reconcileStalePods`): its row has not
+ * moved for `staleMs`, and its change is unfinished, its engine is not connected, or it runs
+ * past its idle limit. A running computer (ADR-149) always qualifies: nothing else looks at it
+ * while a harness holds a ticket, so the sweep is what finds it drained to another worker
+ * (`drainedAt`, recorded once) or deleted outside the Runtime.
+ */
+export function podWantsReconcile(row, facts) {
+    const pod = row.pod;
+    if (!pod || facts.now - Date.parse(row.updatedAt) < facts.staleMs)
+        return false;
+    const running = pod.desired === "running" && pod.observed === "running";
+    return (pod.desired === "deleted" ||
+        pod.retiring !== undefined ||
+        pod.observed === "creating" ||
+        pod.observed === "deleting" ||
+        (pod.desired === "suspended" && pod.observed !== "suspended") ||
+        // Running, but its engine is not connected: the pod may be gone, or its volume. A computer
+        // has no engine connection: look at it every time.
+        (running && (facts.computer || !facts.hosting)) ||
+        // Running past its idle limit: its idle timer is missing (or a turn holds it, which the
+        // reconcile leaves alone).
+        (running && facts.now - Date.parse(pod.lastActiveAt ?? row.createdAt) > facts.idleMs + facts.staleMs));
+}

package/dist/version.d.ts

@@ -1 +1 @@
-export declare const RUNTIME_VERSION = "0.53.2-beta";
+export declare const RUNTIME_VERSION = "0.53.3-beta";

package/dist/version.js

@@ -1,3 +1,3 @@
 // Kept in step with package.json by runtime/server/scripts/check-package.mjs, which runs in npm run check.
 // release:prepare also rewrites this when it bumps @nylorun/runtime.
-export const RUNTIME_VERSION = "0.53.2-beta";
+export const RUNTIME_VERSION = "0.53.3-beta";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nylorun/runtime",
-  "version": "0.53.2-beta",
+  "version": "0.53.3-beta",
   "description": "Nylorun Runtime: the agent execution Host on Postgres, Restate and S2, and model providers.",
   "type": "module",
   "license": "Apache-2.0",