@runuai/host 0.9.91 → 0.9.93

package/lib/machine-provider-aws.ts

@@ -101,11 +101,18 @@ export function awsMachineName(taskId: string): string {
   return `uai-machine-${taskId.toLowerCase()}`;
 }
 
-/** Smallest Graviton type satisfying both dimensions. */
+/** Smallest Graviton type satisfying both dimensions.
+ *
+ * FLOORS AT 16GiB (owner decision 2026-08-31, priced at $0.25/machine-up
+ * hour): an 8GiB machine thrashed to death twice in one day under the
+ * ordinary heavy shape — a 5-agent crew + Turbopack dev server + editor +
+ * VNC browser — and a memory wall reads as "the product is broken", which
+ * costs more than the larger instance. m7g (sustained) over t4g (burstable)
+ * deliberately: compile bursts on a drained CPU-credit balance are the same
+ * wall wearing a different mask. The floor is AWS-only — local VM sizing
+ * has its own constraints and keeps the spec-driven path. */
 export function defaultInstanceType(spec: MachineSpec): string {
   const ladder: Array<[string, number, number]> = [
-    ["m7g.medium", 1, 4096],
-    ["m7g.large", 2, 8192],
     ["m7g.xlarge", 4, 16384],
     ["m7g.2xlarge", 8, 32768],
     ["m7g.4xlarge", 16, 65536],
@@ -136,6 +143,24 @@ export function awsUserData(authorizedPublicKey: string): string {
     '  chmod 700 "$home/.ssh"',
     '  chmod 600 "$home/.ssh/authorized_keys"',
     "done",
+    // Swap BEFORE any workload: an Opus agent + a bundler + Chromium on a
+    // swapless machine hits the memory wall as a thrash (load 14, RAM 100%,
+    // sshd/code-server accept loops starved — live 2026-08-31 on the first
+    // real machine task; the HOSTS got their swapfile in 0.9.85, the
+    // machines never did). Best-effort: a machine without fallocate still
+    // boots.
+    // 16G (owner decision 2026-08-31): the graceful-degradation floor has to
+    // absorb the whole heavy shape, not soften its first half — 4G filled
+    // and the wall moved instead of vanishing. Disk is already paid for.
+    "(fallocate -l 16G /swapfile && chmod 600 /swapfile && mkswap /swapfile && swapon /swapfile && echo '/swapfile none swap sw 0 0' >> /etc/fstab) || true",
+    // The task OWNS its machine (ADR-121 — per-customer kernel isolation is
+    // the whole point); denying sudo just breaks project setups that
+    // container tasks handle fine (live 2026-08-31: a Hightop machine task
+    // could not install the Infisical CLI its repo's config expects).
+    "echo 'node ALL=(ALL) NOPASSWD:ALL' > /etc/sudoers.d/node && chmod 440 /etc/sudoers.d/node",
+    // Infisical CLI: baseline parity with the standard container image —
+    // pnpm-secrets-style project setups expect it on PATH. Best-effort.
+    "(V=0.43.110; case \$(uname -m) in aarch64|arm64) T=arm64;; *) T=amd64;; esac; curl -fsSL -o /tmp/inf.tgz https://github.com/Infisical/cli/releases/download/v\${V}/cli_\${V}_linux_\${T}.tar.gz && tar -xzf /tmp/inf.tgz -C /tmp infisical && install -m 0755 /tmp/infisical /usr/local/bin/infisical) || true",
     "",
   ].join("\n");
   // RAW script, not base64: AWS CLI v2 base64-encodes --user-data ITSELF.

package/lib/vnode-pressure.ts

@@ -30,12 +30,25 @@ const exec = promisify(execFile);
 const ADMISSION_THRESHOLD = 0.9;
 /** Warn-log above this fraction. */
 const WARN_THRESHOLD = 0.8;
+/**
+ * The vnode counter ALONE cannot refuse: `kern.num_vnodes` includes the
+ * reclaimable vnode cache, and a healthy long-uptime Mac converges to
+ * num_vnodes == maxvnodes forever (live false-positive, 2026-08-31: the
+ * Studio refused every task at "100%" with ZERO VMs running and the file
+ * table at 1.2%). Real virtiofs pinning holds OPEN FILES — the Aug-24
+ * incident read 96% of `kern.num_files/kern.maxfiles`. So refusal
+ * requires the vnode ratio AND genuine file-table pressure together.
+ */
+const FILE_TABLE_CORROBORATION_THRESHOLD = 0.5;
 const CACHE_MS = 15_000;
 
 export interface VnodePressure {
   current: number;
   max: number;
   ratio: number;
+  filesCurrent: number;
+  filesMax: number;
+  filesRatio: number;
 }
 
 let cached: { at: number; value: VnodePressure | null } | null = null;
@@ -47,14 +60,30 @@ export async function readVnodePressure(): Promise<VnodePressure | null> {
   try {
     const { stdout } = await exec(
       "/usr/sbin/sysctl",
-      ["-n", "kern.num_vnodes", "kern.maxvnodes"],
+      ["-n", "kern.num_vnodes", "kern.maxvnodes", "kern.num_files", "kern.maxfiles"],
       { timeout: 5_000 },
     );
     const parts = stdout.trim().split(/\s+/).map(Number);
     const current = parts[0] ?? Number.NaN;
     const max = parts[1] ?? Number.NaN;
-    if (Number.isFinite(current) && Number.isFinite(max) && max > 0) {
-      value = { current, max, ratio: current / max };
+    const filesCurrent = parts[2] ?? Number.NaN;
+    const filesMax = parts[3] ?? Number.NaN;
+    if (
+      Number.isFinite(current) &&
+      Number.isFinite(max) &&
+      max > 0 &&
+      Number.isFinite(filesCurrent) &&
+      Number.isFinite(filesMax) &&
+      filesMax > 0
+    ) {
+      value = {
+        current,
+        max,
+        ratio: current / max,
+        filesCurrent,
+        filesMax,
+        filesRatio: filesCurrent / filesMax,
+      };
     }
   } catch {
     // Sensor failure is never a task failure; admission simply has no data.
@@ -67,11 +96,15 @@ export async function readVnodePressure(): Promise<VnodePressure | null> {
 export async function vnodeAdmissionProblem(): Promise<string | null> {
   const pressure = await readVnodePressure();
   if (!pressure || pressure.ratio < ADMISSION_THRESHOLD) return null;
+  // A full vnode table with an idle file table is a warm cache, not danger.
+  if (pressure.filesRatio < FILE_TABLE_CORROBORATION_THRESHOLD) return null;
   return (
     `the macOS vnode table is at ${Math.round(pressure.ratio * 100)}% ` +
-    `(${pressure.current}/${pressure.max}); starting another task VM risks ` +
-    "system-wide file-table exhaustion. Close a running task, or raise the " +
-    "ceiling once with: sudo uai-host tune"
+    `(${pressure.current}/${pressure.max}) with the file table at ` +
+    `${Math.round(pressure.filesRatio * 100)}% ` +
+    `(${pressure.filesCurrent}/${pressure.filesMax}); starting another task ` +
+    "VM risks system-wide file-table exhaustion. Close a running task, or " +
+    "raise the ceiling once with: sudo uai-host tune"
   );
 }
 
@@ -82,15 +115,19 @@ let lastLoggedBand: "ok" | "warn" | "critical" = "ok";
 export async function logVnodePressure(): Promise<void> {
   const pressure = await readVnodePressure();
   if (!pressure) return;
-  const band =
-    pressure.ratio >= ADMISSION_THRESHOLD
-      ? "critical"
-      : pressure.ratio >= WARN_THRESHOLD
-        ? "warn"
-        : "ok";
+  const refusing =
+    pressure.ratio >= ADMISSION_THRESHOLD &&
+    pressure.filesRatio >= FILE_TABLE_CORROBORATION_THRESHOLD;
+  const band = refusing
+    ? "critical"
+    : pressure.filesRatio >= WARN_THRESHOLD || pressure.ratio >= WARN_THRESHOLD
+      ? "warn"
+      : "ok";
   if (band === lastLoggedBand) return;
   lastLoggedBand = band;
-  const detail = `${pressure.current}/${pressure.max} (${Math.round(pressure.ratio * 100)}%)`;
+  const detail =
+    `vnodes ${pressure.current}/${pressure.max} (${Math.round(pressure.ratio * 100)}%), ` +
+    `files ${pressure.filesCurrent}/${pressure.filesMax} (${Math.round(pressure.filesRatio * 100)}%)`;
   if (band === "critical") {
     console.warn(
       `[vnode] table at ${detail} — new apple task VMs will be refused; run: sudo uai-host tune`,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@runuai/host",
-  "version": "0.9.91",
+  "version": "0.9.93",
   "description": "Uai host — runs ephemeral AI tasks in containers on a machine you control.",
   "license": "MIT",
   "author": "Uai Tech <team@runuai.com>",