@runuai/host 0.9.105 → 0.9.106

package/lib/machine-provider-aws.ts

@@ -279,10 +279,19 @@ async function hostEgressIp(): Promise<string> {
   );
 }
 
+/** How long a resolved payload AMI is trusted before the newest published
+ * image is looked up again. Long enough that a burst of launches shares one
+ * describe-images; short enough that a freshly published image reaches every
+ * RUNNING host — shared and BYO — on its next task without a host restart
+ * (before 0.9.106 the answer was kept for the process lifetime). */
+export const PAYLOAD_AMI_TTL_MS = 5 * 60_000;
+
 export function createAwsMachineProvider(
   config: AwsMachineConfig,
   runner: Runner = defaultRunner(config.region),
+  deps: { now?: () => number } = {},
 ): MachineProvider {
+  const now = deps.now ?? (() => Date.now());
   const taskIdOf = (machineId: string): string => {
     if (!machineId.startsWith("uai-machine-")) {
       throw new Error(`not a uai machine id: ${machineId}`);
@@ -332,10 +341,12 @@ export function createAwsMachineProvider(
   // orchestrator can never reach.
   // -------------------------------------------------------------------------
 
-  let resolvedImage: string | null = null;
+  let resolvedImage: { id: string; at: number } | null = null;
   async function resolveImage(specImage: string): Promise<string> {
     if (specImage.startsWith("ami-")) return specImage;
-    if (resolvedImage) return resolvedImage;
+    if (resolvedImage && now() - resolvedImage.at < PAYLOAD_AMI_TTL_MS) {
+      return resolvedImage.id;
+    }
     const res = await runner([
       "ec2",
       "describe-images",
@@ -356,12 +367,24 @@ export function createAwsMachineProvider(
       }
     }
     if (typeof image !== "string" || !image.startsWith("ami-")) {
+      // A stale-but-known image beats a failed launch when the lookup itself
+      // hiccups; only a host that never resolved one has nothing to offer.
+      if (resolvedImage) return resolvedImage.id;
       throw new Error(
         `no uai machine payload AMI is published in ${config.region} — set ` +
           "UAI_MACHINE_IMAGE to an AMI id, or ask Uai to publish this region",
       );
     }
-    resolvedImage = image;
+    if (resolvedImage?.id !== image) {
+      // The journal line that tells an operator which payload a task got —
+      // an exit 127 for a missing CLI is otherwise the only symptom of a
+      // region that never received the latest publish.
+      console.log(
+        `[machine-aws] payload image for ${config.region}: ${image}` +
+          (resolvedImage ? ` (was ${resolvedImage.id})` : ""),
+      );
+    }
+    resolvedImage = { id: image, at: now() };
     return image;
   }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@runuai/host",
-  "version": "0.9.105",
+  "version": "0.9.106",
   "description": "Uai host — runs ephemeral AI tasks in containers on a machine you control.",
   "license": "MIT",
   "author": "Uai Tech <team@runuai.com>",