w

Author: Gusarich

.github/scripts/create_pod.py

@@ -17,7 +17,6 @@ def main():
         sys.exit(1)
 
     print("Creating RunPod instance...")
-    print(f"SSH public key: {ssh_public_key[:50]}...")
 
     pod = runpod.create_pod(
         name="nanogpt-benchmark",
@@ -26,54 +25,34 @@ def main():
         gpu_count=8,
         volume_in_gb=0,
         container_disk_in_gb=100,
-        ports="22/tcp,8888/http",
+        ports="22/tcp",
         support_public_ip=True,
         env={"SSH_PUBLIC_KEY": ssh_public_key},
     )
 
     pod_id = pod["id"]
     pod_host_id = pod.get("machine", {}).get("podHostId", pod_id)
-    print(f"Pod created: {pod_id}")
-    print(f"SSH host ID: {pod_host_id}")
+    print(f"Pod ID: {pod_id}")
+    print(f"SSH target: {pod_host_id}@ssh.runpod.io")
 
     with open("pod_id.txt", "w") as f:
         f.write(pod_id)
-    with open("ssh_host_id.txt", "w") as f:
+    with open("ssh_target.txt", "w") as f:
         f.write(pod_host_id)
 
-    print("Waiting for pod...")
-    start_time = time.time()
-
-    while True:
+    print("Waiting for pod to be RUNNING...")
+    for _ in range(24):
         pod_info = runpod.get_pod(pod_id)
         status = pod_info.get("desiredStatus")
-
+        print(f"  Status: {status}")
         if status == "RUNNING":
-            public_ip = pod_info.get("publicIp")
-            port_mappings = pod_info.get("portMappings") or {}
-            ssh_port = port_mappings.get("22") or port_mappings.get(22)
-
-            if public_ip and ssh_port:
-                # Direct SSH via public IP
-                print(f"Pod ready (direct SSH): root@{public_ip} -p {ssh_port}")
-                with open("ssh_info.txt", "w") as f:
-                    f.write(f"direct:{public_ip}:{ssh_port}")
-                break
-            elif time.time() - start_time > 30:
-                # Fall back to basic SSH proxy after 30s
-                print(f"Pod ready (proxy SSH): {pod_host_id}@ssh.runpod.io")
-                with open("ssh_info.txt", "w") as f:
-                    f.write(f"proxy:{pod_host_id}")
-                break
-            else:
-                print(f"  Waiting for public IP... ({int(time.time() - start_time)}s)")
-
-        if time.time() - start_time > 120:
-            print("ERROR: Pod timeout (2 min)", file=sys.stderr)
-            sys.exit(1)
-
+            print("Pod is ready!")
+            return
         time.sleep(5)
 
+    print("ERROR: Pod timeout", file=sys.stderr)
+    sys.exit(1)
+
 
 if __name__ == "__main__":
     main()