test: fix GPU e2e tests — auto-install runtime and wait for endpoint readiness

Signed-off-by: fredespi <fredrik.espinoza@gmail.com>

Author: fredespi

tests/e2e/libraries/RocmCliLibrary.py

@@ -55,8 +55,15 @@ def setup_standard_rocm(self):
 
     @keyword("a managed runtime is active")
     def setup_active_runtime(self):
-        """Precondition: verify a managed runtime is active."""
+        """Precondition: ensure a managed runtime is active, installing if needed."""
         result = run_rocm("runtimes", "list")
+        if result.contains("installed: none"):
+            logger.info("No runtime found, installing SDK...")
+            install = run_rocm("install", "sdk", timeout=600)
+            assert install.rc == 0, (
+                f"rocm install sdk failed (rc={install.rc}):\n{install.stdout}"
+            )
+            result = run_rocm("runtimes", "list")
         assert not result.contains("installed: none"), (
             f"No managed runtime is active:\n{result.stdout}"
         )

tests/e2e/libraries/ServeLibrary.py

@@ -11,6 +11,8 @@
 from __future__ import annotations
 
 import json
+import time
+import urllib.error
 import urllib.request
 
 from common.cli_runner import CliResult, run_rocm
@@ -90,7 +92,7 @@ def setup_localhost_model(self):
 
     @keyword("a model is being served on GPU")
     def setup_gpu_model(self):
-        """Start a real model server using rocm serve."""
+        """Start a real model server using rocm serve, waiting for readiness."""
         self._serve_result = run_rocm(
             "serve",
             "Qwen/Qwen2.5-1.5B-Instruct",
@@ -104,6 +106,7 @@ def setup_gpu_model(self):
         )
         self._endpoint = "http://127.0.0.1:11435/v1"
         self._model_name = "Qwen/Qwen2.5-1.5B-Instruct"
+        self._wait_for_endpoint()
         logger.info(f"Model served at {self._endpoint}")
 
     @keyword("a model is served in the background")
@@ -170,6 +173,23 @@ def user_serves_default_engine(self):
         self._endpoint = "http://127.0.0.1:11435/v1"
         self._model_name = "Qwen/Qwen2.5-1.5B-Instruct"
 
+    def _wait_for_endpoint(self, timeout: int = 120):
+        """Poll the endpoint until it responds or timeout is reached."""
+        assert self._endpoint is not None, "No endpoint configured"
+        url = f"{self._endpoint}/models"
+        deadline = time.time() + timeout
+        while time.time() < deadline:
+            try:
+                req = urllib.request.Request(url)
+                with urllib.request.urlopen(req, timeout=5) as resp:
+                    if resp.status == 200:
+                        logger.info(f"Endpoint ready at {self._endpoint}")
+                        return
+            except (urllib.error.URLError, OSError):
+                pass
+            time.sleep(5)
+        raise AssertionError(f"Endpoint {self._endpoint} not ready after {timeout}s")
+
     @keyword("the dashboard offers to use the detected endpoint")
     def dashboard_offers_endpoint(self):
         """Simulate the TUI detecting a local endpoint.