@runuai/host 0.9.159 → 0.9.160

package/lib/task-files-machine-retirement-outbox.ts

@@ -243,6 +243,77 @@ export function taskMachinePeerProcessIds(): number[] {
     );
 }
 
+/** A process's parent from /proc; null once the process is gone. */
+export function processParentId(pid: number): number | null {
+  let stat: string;
+  try {
+    stat = readFileSync(`/proc/${pid}/stat`, "utf8");
+  } catch (error) {
+    if ((error as NodeJS.ErrnoException).code === "ENOENT") return null;
+    throw error;
+  }
+  // `pid (comm) state ppid ...`: comm may itself hold spaces or parens.
+  const ppid = Number(stat.slice(stat.lastIndexOf(")") + 2).split(" ")[1]);
+  if (!Number.isSafeInteger(ppid) || ppid < 0) {
+    throw new Error(`process ${pid} has an unreadable parent`);
+  }
+  return ppid;
+}
+
+/**
+ * Every process in the sidecar service cgroup that this runner did not start.
+ * The fresh-incarnation proof asks this, not `taskMachinePeerProcessIds`: the
+ * runner's own helpers share its cgroup from boot, tsx's esbuild service above
+ * all (a long-lived child of every runner that compiled a module; live
+ * 2026-10-20, every marked machine stayed pending "not quiescent" on it), and
+ * a pending runner admits no task, so nothing it started is a task process.
+ * What an earlier runner process left behind (KillMode=process) was reparented
+ * when that process ended: it is never this runner's descendant, so it counts.
+ */
+export function taskMachineForeignProcessIds(): number[] {
+  return foreignProcessIdsWith({
+    list: taskMachinePeerProcessIds,
+    parent: processParentId,
+    self: process.pid,
+  });
+}
+
+export interface ForeignProcessDeps {
+  list(): readonly number[];
+  parent(pid: number): number | null;
+  self: number;
+}
+
+/** Exported seam keeps the ancestry walk executable in tests. */
+export function foreignProcessIdsWith(deps: ForeignProcessDeps): number[] {
+  const foreign: number[] = [];
+  for (const pid of deps.list()) {
+    if (pid === deps.self) continue;
+    const seen = new Set<number>();
+    let current: number | null = pid;
+    let ancestry: "ours" | "foreign" | "gone" = "foreign";
+    for (;;) {
+      const parent = deps.parent(current);
+      if (parent === null) {
+        // The process itself is gone: nothing to count. An ancestor gone
+        // mid-walk leaves an orphan the next look sees reparented: fail
+        // closed now, as that look would.
+        if (current === pid) ancestry = "gone";
+        break;
+      }
+      if (parent === deps.self) {
+        ancestry = "ours";
+        break;
+      }
+      if (parent <= 1 || seen.has(parent)) break;
+      seen.add(parent);
+      current = parent;
+    }
+    if (ancestry === "foreign") foreign.push(pid);
+  }
+  return foreign;
+}
+
 /** Kill everything in the sidecar service cgroup except this root runner.
  * The task's agents, editor, terminals and sudo children all descend from the
  * service; KillMode=process is precisely why they otherwise survive. */

package/lib/task-machine-credentials-activation.ts

@@ -26,6 +26,8 @@ export interface TaskMachineCredentialsActivationDeps {
   filesReady(): boolean;
   retirementReady(): boolean;
   taskRows(): readonly { taskId: string }[];
+  /** Processes in the sidecar's service cgroup that this runner did not
+   *  start (its own helpers, such as tsx's esbuild service, never count). */
   peerProcessIds(): readonly number[];
 }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@runuai/host",
-  "version": "0.9.159",
+  "version": "0.9.160",
   "description": "Uai host — runs ephemeral AI tasks in containers on a machine you control.",
   "license": "MIT",
   "author": "Uai Tech <team@runuai.com>",

package/src/main.ts

@@ -428,7 +428,7 @@ import {
   kernelBootId,
   TaskFilesMachineRetirementOutbox,
   TASK_FILES_MACHINE_RETIREMENT_RECORD,
-  taskMachinePeerProcessIds,
+  taskMachineForeignProcessIds,
 } from "../lib/task-files-machine-retirement-outbox";
 import {
   planFreshTaskMachineCredentialsActivation,
@@ -1819,7 +1819,8 @@ function taskMachineCredentialsActivationDeps(): TaskMachineCredentialsActivatio
         .select({ taskId: schema.hostTasks.taskId })
         .from(schema.hostTasks)
         .all(),
-    peerProcessIds: taskMachinePeerProcessIds,
+    // Not the runner's own children (tsx's esbuild service): see the helper.
+    peerProcessIds: taskMachineForeignProcessIds,
   };
 }