ci: Reduce sleeping/retries by using unikraft instance wait

_testlib/http_client.py

@@ -43,7 +43,7 @@ def _request_with_retries(
     *,
     timeout: float = 10.0,
     retries: int = 10,
-    backoff: float = 2.0,
+    backoff: float = 1.0,
     expected_status: int | None = 200,
     verify: bool | str | None = None,
     **kwargs,
@@ -95,7 +95,7 @@ def http_get(
     *,
     timeout: float = 10.0,
     retries: int = 10,
-    backoff: float = 2.0,
+    backoff: float = 1.0,
     expected_status: int | None = 200,
     verify: bool | str | None = None,
     **kwargs,
@@ -113,7 +113,7 @@ def http_post(
     *,
     timeout: float = 10.0,
     retries: int = 10,
-    backoff: float = 2.0,
+    backoff: float = 1.0,
     expected_status: int | None = 200,
     verify: bool | str | None = None,
     **kwargs,

_testlib/unikraft.py

@@ -150,6 +150,29 @@ def get_instance(self, target: str) -> dict[str, Any]:
 
         return _parse_json(proc.stdout)
 
+    def wait_instance(
+        self, target: str, state: str, *, timeout: float | None = 150
+    ) -> dict[str, Any]:
+        """Wait until ``target`` reaches the given ``state``.
+
+        Wraps ``unikraft instances wait <target> --until state==<state>``.
+        Returns the parsed JSON description of the instance once it reaches
+        the desired state.
+        """
+        proc = self.run(
+            [
+                "instances",
+                "wait",
+                target,
+                "--until",
+                f"state=={state}",
+                "--output",
+                "json",
+            ],
+            timeout=timeout,
+        )
+        return _parse_json(proc.stdout)
+
     def delete_instance(self, target: str) -> None:
         try:
             proc =self.run(["instances", "delete", target], check=False)

build-environments/test_build-environments.py

@@ -61,7 +61,6 @@ def _cleanup_base():
 
     # 1. Build the base image.
     unikraft.build(context, base_tag)
-    time.sleep(3)
 
     # 2. Run the base image; it auto-converts into a template.
     unikraft.run_instance(
@@ -85,13 +84,12 @@ def _cleanup_base():
     # 4. Build ROM images.
     unikraft.build(context / "rom1", rom1_tag)
     unikraft.build(context / "rom2", rom2_tag)
-    time.sleep(3)
 
     return template_name, rom1_tag, rom2_tag
 
 
 def test_build_environments_rom1(
-    _template_and_roms, request, unikraft, test_run_id, http
+    _template_and_roms, request, unikraft, test_run_id, http, wait_instance
 ):
     """Run an instance from the template with ROM1 and verify response."""
     template_name, rom1_tag, _ = _template_and_roms
@@ -112,13 +110,14 @@ def test_build_environments_rom1(
     url = extract_instance_url(instance)
     assert url, f"could not determine instance URL from: {instance!r}"
 
+    wait_instance(instance_name, "standby")
     resp = http(url)
     assert resp.status_code == 200
     assert "Bye, World!" in resp.text
 
 
 def test_build_environments_rom2(
-    _template_and_roms, request, unikraft, test_run_id, http
+    _template_and_roms, request, unikraft, test_run_id, http, wait_instance
 ):
     """Run an instance from the template with ROM2 and verify response."""
     template_name, _, rom2_tag = _template_and_roms
@@ -139,6 +138,7 @@ def test_build_environments_rom2(
     url = extract_instance_url(instance)
     assert url, f"could not determine instance URL from: {instance!r}"
 
+    wait_instance(instance_name, "standby")
     resp = http(url)
     assert resp.status_code == 200
     assert "Auf Wiedersehen!" in resp.text

caddy2.7-go1.21/test_caddy2.7-go1.21.py

@@ -9,10 +9,10 @@
 
 from __future__ import annotations
 
-from _testlib.unikraft import extract_instance_url
+from _testlib.unikraft import extract_instance_name, extract_instance_url
 
 
-def test_caddy_serves_welcome_page(build_image, run_instance, http):
+def test_caddy_serves_welcome_page(build_image, run_instance, http, wait_instance):
     image = build_image("caddy2.7-go1.21", "caddy2.7-go1.21")
 
     instance = run_instance(
@@ -24,6 +24,8 @@ def test_caddy_serves_welcome_page(build_image, run_instance, http):
     url = extract_instance_url(instance)
     assert url, f"could not determine instance URL from: {instance!r}"
 
+    wait_instance(extract_instance_name(instance), "running")
+
     resp = http(url)
     assert resp.status_code == 200
     assert "Hello, world!" in resp.text

chromium-cdp/test_chromium_cdp.py

@@ -9,10 +9,10 @@
 
 from __future__ import annotations
 
-from _testlib.unikraft import extract_instance_url
+from _testlib.unikraft import extract_instance_name, extract_instance_url
 
 
-def test_chromium_version(build_image, run_instance, http):
+def test_chromium_version(build_image, run_instance, http, wait_instance):
     image = build_image("chromium-cdp", "chromium-cdp")
 
     instance = run_instance(
@@ -24,6 +24,8 @@ def test_chromium_version(build_image, run_instance, http):
     url = extract_instance_url(instance)
     assert url, f"could not determine instance URL from: {instance!r}"
 
+    wait_instance(extract_instance_name(instance), "running")
+
     resp = http(f"{url}/json/version")
     assert resp.status_code == 200
     assert "Browser" in resp.text

conftest.py

@@ -32,6 +32,7 @@
 from _testlib.socat import SocatTunnel
 from _testlib.unikraft import (
     UnikraftCLI,
+    extract_instance_name,
     extract_instance_url,
 )
 
@@ -138,10 +139,6 @@ def _build(example_dir: str, image_name: str) -> str:
 
         unikraft.build(context, tag)
 
-        # TODO: drop this once the platform exposes a way to wait until a
-        # freshly-built image is fully available for `unikraft run`.
-        time.sleep(3)
-
         return tag
 
     return _build
@@ -228,6 +225,31 @@ def _cleanup() -> None:
     return _run
 
 
+# ---------------------------------------------------------------------------
+# Instance wait fixture
+# ---------------------------------------------------------------------------
+
+
+WaitInstance = Callable[[str, str], dict[str, Any]]
+
+
+@pytest.fixture
+def wait_instance(unikraft: UnikraftCLI) -> WaitInstance:
+    """Wait until a named instance reaches the given state.
+
+    Usage::
+
+        def test_foo(run_instance, wait_instance):
+            instance = run_instance(image, ...)
+            wait_instance(extract_instance_name(instance), "running")
+    """
+
+    def _wait(target: str, state: str) -> dict[str, Any]:
+        return unikraft.wait_instance(target, state)
+
+    return _wait
+
+
 # ---------------------------------------------------------------------------
 # HTTP helper
 # ---------------------------------------------------------------------------
@@ -284,4 +306,5 @@ def _create(host: str, port: int, local_port: int) -> SocatTunnel:
     "run_instance",
     "socat_tunnel",
     "unikraft",
+    "wait_instance",
 ]

debian-ssh/test_debian-ssh.py

@@ -12,12 +12,11 @@
 from __future__ import annotations
 
 import subprocess
-import time
 from pathlib import Path
 
 import pytest
 
-from _testlib.unikraft import extract_instance_fqdn
+from _testlib.unikraft import extract_instance_fqdn, extract_instance_name
 
 SSH_PORT = 2222
 
@@ -36,7 +35,7 @@ def _generate_test_keypair(tmp_path: Path) -> tuple[Path, str]:
     return key_path, pub_key
 
 
-def test_debian_ssh(build_image, run_instance, socat_tunnel, tmp_path):
+def test_debian_ssh(build_image, run_instance, socat_tunnel, wait_instance, tmp_path):
     """Build, deploy, and SSH into a Debian instance."""
     private_key, public_key = _generate_test_keypair(tmp_path)
 
@@ -58,37 +57,24 @@ def test_debian_ssh(build_image, run_instance, socat_tunnel, tmp_path):
     # Set up a socat TLS tunnel to the SSH port.
     tunnel = socat_tunnel(host, SSH_PORT, SSH_PORT)
 
-    # Retry SSH connection with back-off while the instance boots.
-    last_err = None
-    for _ in range(10):
-        try:
-            result = subprocess.run(
-                [
-                    "ssh",
-                    "-o", "StrictHostKeyChecking=no",
-                    "-o", "UserKnownHostsFile=/dev/null",
-                    "-o", "ConnectTimeout=10",
-                    "-i", str(private_key),
-                    "-p", str(tunnel.local_port),
-                    "-l", "root",
-                    "127.0.0.1",
-                    "echo hello-from-pytest",
-                ],
-                capture_output=True,
-                text=True,
-                timeout=30,
-            )
-            if result.returncode == 0:
-                assert "hello-from-pytest" in result.stdout
-                return
-            last_err = RuntimeError(
-                f"ssh exited {result.returncode}: {result.stderr}"
-            )
-        except (subprocess.TimeoutExpired, OSError) as exc:
-            last_err = exc
-
-        time.sleep(5)
-
-    raise AssertionError(
-        f"could not connect via SSH after retries: {last_err}"
-    ) from last_err
+    wait_instance(extract_instance_name(instance), "running")
+    result = subprocess.run(
+        [
+            "ssh",
+            "-o", "StrictHostKeyChecking=no",
+            "-o", "UserKnownHostsFile=/dev/null",
+            "-o", "ConnectTimeout=10",
+            "-i", str(private_key),
+            "-p", str(tunnel.local_port),
+            "-l", "root",
+            "127.0.0.1",
+            "echo hello-from-pytest",
+        ],
+        capture_output=True,
+        text=True,
+        timeout=30,
+    )
+    assert result.returncode == 0, (
+        f"ssh failed (exit={result.returncode})\nstdout: {result.stdout}\nstderr: {result.stderr}"
+    )
+    assert "hello-from-pytest" in result.stdout

dragonflydb/test_dragonflydb.py

@@ -15,10 +15,10 @@
 
 from __future__ import annotations
 
-from _testlib.unikraft import extract_instance_url
+from _testlib.unikraft import extract_instance_name, extract_instance_url
 
 
-def test_dragonflydb(build_image, run_instance, http):
+def test_dragonflydb(build_image, run_instance, http, wait_instance):
     """Build, deploy, and exercise a DragonflyDB instance."""
     image = build_image("dragonflydb", "dragonflydb")
 
@@ -31,6 +31,8 @@ def test_dragonflydb(build_image, run_instance, http):
     url = extract_instance_url(instance)
     assert url, f"could not determine instance URL from: {instance!r}"
 
+    wait_instance(extract_instance_name(instance), "running")
+
     # http() already retries with back-off, so no manual sleep needed.
     # ------------------------------------------------------------------
     # 1. HTTP status page — matches existing CI workflow (curl check).

duckdb-go1.21/test_duckdb-go1.21.py

@@ -9,10 +9,10 @@
 
 from __future__ import annotations
 
-from _testlib.unikraft import extract_instance_url
+from _testlib.unikraft import extract_instance_name, extract_instance_url
 
 
-def test_duckdb_go_serves_query_result(build_image, run_instance, http):
+def test_duckdb_go_serves_query_result(build_image, run_instance, http, wait_instance):
     image = build_image("duckdb-go1.21", "duckdb-go1.21")
 
     instance = run_instance(
@@ -24,6 +24,8 @@ def test_duckdb_go_serves_query_result(build_image, run_instance, http):
     url = extract_instance_url(instance)
     assert url, f"could not determine instance URL from: {instance!r}"
 
+    wait_instance(extract_instance_name(instance), "running")
+
     resp = http(url)
     assert resp.status_code == 200
     assert "42" in resp.text

feature-change-instance-cmd/test_feature-change-instance-cmd.py

@@ -9,10 +9,10 @@
 
 from __future__ import annotations
 
-from _testlib.unikraft import extract_instance_url
+from _testlib.unikraft import extract_instance_name, extract_instance_url
 
 
-def test_httpserver_serves_hello(build_image, run_instance, http):
+def test_httpserver_serves_hello(build_image, run_instance, http, wait_instance):
     image = build_image("feature-change-instance-cmd", "feature-change-instance-cmd")
 
     instance = run_instance(
@@ -24,6 +24,8 @@ def test_httpserver_serves_hello(build_image, run_instance, http):
     url = extract_instance_url(instance)
     assert url, f"could not determine instance URL from: {instance!r}"
 
+    wait_instance(extract_instance_name(instance), "running")
+
     resp = http(url)
     assert resp.status_code == 200
     assert "Hello, World!" in resp.text

github-webhook-node/test_github-webhook-node.py

@@ -9,10 +9,10 @@
 
 from __future__ import annotations
 
-from _testlib.unikraft import extract_instance_url
+from _testlib.unikraft import extract_instance_name, extract_instance_url
 
 
-def test_github_webhook_health(build_image, run_instance, http):
+def test_github_webhook_health(build_image, run_instance, http, wait_instance):
     image = build_image("github-webhook-node", "github-webhook-node")
 
     instance = run_instance(
@@ -27,6 +27,8 @@ def test_github_webhook_health(build_image, run_instance, http):
     url = extract_instance_url(instance)
     assert url, f"could not determine instance URL from: {instance!r}"
 
+    wait_instance(extract_instance_name(instance), "running")
+
     resp = http(f"{url}/health")
     assert resp.status_code == 200
 

grafana/test_grafana.py

@@ -10,10 +10,10 @@
 
 from __future__ import annotations
 
-from _testlib.unikraft import extract_instance_url
+from _testlib.unikraft import extract_instance_name, extract_instance_url
 
 
-def test_grafana(build_image, run_instance, http):
+def test_grafana(build_image, run_instance, http, wait_instance):
     """Build, deploy, and verify Grafana serves its login page."""
     image = build_image("grafana", "grafana")
 
@@ -26,6 +26,8 @@ def test_grafana(build_image, run_instance, http):
     url = extract_instance_url(instance)
     assert url, f"could not determine instance URL from: {instance!r}"
 
+    wait_instance(extract_instance_name(instance), "running")
+
     resp = http(url)
     assert resp.status_code == 200
     body = resp.text.lower()