fix(ci): wait for code-server in healthcheck

.vscode-test.mjs

@@ -27,6 +27,9 @@ export default defineConfig({
     "--disable-extension=ritwickdey.liveserver",
     "--disable-extension=streetsidesoftware.code-spell-checker",
   ],
+  download: {
+    timeout: 60_000,
+  },
   mocha: {
     color: true,
     ui: "bdd",

docker-compose.yml

@@ -4,7 +4,7 @@
 # cspell: ignore healthcheck
 services:
   selenium-vscode:
-    image: ghcr.io/ansible/selenium-adt:main
+    image: ghcr.io/ansible/selenium-adt:26.4.0
     pull_policy: missing
     container_name: selenium-vscode
     shm_size: "10g"
@@ -13,18 +13,19 @@ services:
     ports:
       - "4444:4444"
       - "5999:5999"
+      - "127.0.0.1:8080:8080"
     volumes:
       # apparently podman does not want to mount file as container root
       - ${VSX_FILE:-./out/data/ansible-latest.vsix}:/data/ansible-latest.vsix:ro
       - ./.vscode-test/user-data/User/settings.json:/home/selenium/.local/share/code-server/User/settings.json:rw
       - ./out/ui/logs:/home/selenium/.local/share/code-server/logs:rw
     restart: "no"
     healthcheck:
-      test: ["CMD", "curl", "-f", "http://localhost:4444"]
+      test: ["CMD-SHELL", "curl -sf --max-time 3 http://localhost:4444 && curl -sf --max-time 3 http://localhost:8080"]
       interval: 10s
       timeout: 5s
-      retries: 10
-      start_period: 10s
+      retries: 30
+      start_period: 30s
     deploy:
       resources:
         limits:

package.json

@@ -1150,6 +1150,9 @@
           "mocha": "*"
         }
       }
+    },
+    "patchedDependencies": {
+      "@vscode/test-cli@0.0.12": "patches/@vscode__test-cli@0.0.12.patch"
     }
   },
   "scripts": {

patches/@vscode__test-cli@0.0.12.patch

@@ -0,0 +1,14 @@
+diff --git a/out/cli/platform/desktop.mjs b/out/cli/platform/desktop.mjs
+index b901849b00f8072284a225ab703460e21425fcfc..50b5bcc3d7b7eb4c6f8ecb6ab91dd124af79109a 100644
+--- a/out/cli/platform/desktop.mjs
++++ b/out/cli/platform/desktop.mjs
+@@ -126,7 +126,8 @@ class PreparedDesktopRun {
+             return;
+         }
+         const opts = this.baseCliOptions();
+-        const vscodePath = await electron.downloadAndUnzipVSCode(opts.version, opts.platform, opts.reporter);
++        // https://github.com/microsoft/vscode-test-cli/issues/106
++        const vscodePath = await electron.downloadAndUnzipVSCode(opts);
+         const [cli, ...cliArgs] = electron.resolveCliArgsFromVSCodeExecutablePath(vscodePath, opts);
+         for (const extension of exts.value) {
+             cliArgs.push('--install-extension', extension);

test/ui/fixtures/ui_fixtures.py

@@ -62,7 +62,7 @@ def browser_setup(
         "capturemanager"
     )  # type: ignore[name-defined]
     try:
-        if not is_container_healthy() or True:
+        if not is_container_healthy():
             subprocess.run(
                 f"podman rm -f {CONTAINER_NAME} 2>/dev/null || true",
                 shell=True,
@@ -80,14 +80,43 @@ def browser_setup(
                     shell=True,
                     cwd=_PROJECT_ROOT,
                 )
+            health_timeout_raw = os.environ.get(
+                "UI_CONTAINER_HEALTH_TIMEOUT",
+                "360",
+            )
+            try:
+                health_timeout = int(health_timeout_raw)
+            except ValueError:
+                pytest.fail(
+                    f"Invalid UI_CONTAINER_HEALTH_TIMEOUT={health_timeout_raw!r}; "
+                    "expected an integer number of seconds",
+                )
+            start = time.time()
+            deadline = start + health_timeout
             count = 0
-            while True:
+            while time.time() < deadline:
                 if is_container_healthy():
+                    elapsed = int(time.time() - start)
+                    log.info(
+                        "Container %s healthy after %ss (%s checks)",
+                        CONTAINER_NAME,
+                        elapsed,
+                        count + 1,
+                    )
                     break
                 count += 1
-                time.sleep(1)
+                elapsed = int(time.time() - start)
                 log.info(
-                    "Waiting for container %s to be healthy: %s", CONTAINER_NAME, count
+                    "Waiting for container %s to be healthy: check %s (%ss elapsed)",
+                    CONTAINER_NAME,
+                    count,
+                    elapsed,
+                )
+                time.sleep(2)
+            else:
+                pytest.fail(
+                    f"container {CONTAINER_NAME} did not become healthy "
+                    f"within {health_timeout}s",
                 )
 
         browser = os.environ.get("BROWSER_TYPE")
@@ -103,7 +132,12 @@ def browser_setup(
             command_executor="http://localhost:4444/wd/hub",
             options=options,
         )
+        driver.set_page_load_timeout(120)
+        driver.set_script_timeout(60)
         driver.maximize_window()
+        log.info(
+            "Browser connected, starting UI tests (no output until completion)",
+        )
 
         yield driver, "https://stage.ai.ansible.redhat.com/login"
         close_all_tabs(driver)
@@ -138,6 +172,8 @@ def new_browser() -> Generator[tuple[WebDriver | None, str, None], None, None]:
         driver = WebDriver(
             command_executor="http://localhost:4444/wd/hub", options=options
         )
+        driver.set_page_load_timeout(120)
+        driver.set_script_timeout(60)
         driver.maximize_window()
         yield driver, "https://stage.ai.ansible.redhat.com/login", None
     finally:

test/ui/utils/ui_utils.py

@@ -2,6 +2,7 @@
 
 # pylint: disable=E0401
 import contextlib
+import logging
 import os
 import time
 from collections.abc import Generator
@@ -28,6 +29,8 @@
 from selenium.webdriver.support import expected_conditions
 from selenium.webdriver.support.ui import WebDriverWait
 
+log = logging.getLogger(__name__)
+
 LIGHTSPEED_USER = os.environ.get("LIGHTSPEED_USER", "")
 LIGHTSPEED_PASSWORD = os.environ.get("LIGHTSPEED_PASSWORD", "")
 
@@ -77,14 +80,20 @@ def ensure_vscode_ready(driver: WebDriver, timeout: int = 120) -> None:
     """
     driver.switch_to.default_content()
     if "127.0.0.1:8080" not in driver.current_url:
+        log.info("Navigating to code-server at http://127.0.0.1:8080")
         driver.get("http://127.0.0.1:8080")
+        log.info("Page loaded successfully")
+    log.info("Waiting for Ansible sidebar icon (up to 60s)")
     wait_displayed(driver, "//a[@aria-label='Ansible']", timeout=60)
+    log.info("Sidebar icon found, focusing ADT view")
     vscode_run_command(driver, ">Ansible: Focus on Ansible Development Tools View")
+    log.info("Waiting for extension activation (welcome page, up to %ss)", timeout)
     find_element_across_iframes(
         driver,
         "//a[contains(@title, 'Ansible Development Tools welcome page')]",
         retries=timeout,
     )
+    log.info("Extension ready")
 
 
 def click_and_wait(
@@ -1021,42 +1030,37 @@ def vscode_run_command(
     if not command.startswith(">"):
         command = ">" + command
 
-    # Dismiss any leftover palette/dialog from a previous call to avoid
-    # blocking the command-center click with an overlay.
     ActionChains(driver).send_keys(Keys.ESCAPE).perform()
     time.sleep(0.3)
 
-    # click the command box
-    max_attempts = 6  # we seen timeout issue on GHA with 4 seconds
+    log.info("Opening command palette via F1 for: %s", command)
+    max_attempts = 3
     command_input = None
     for i in range(max_attempts):
+        ActionChains(driver).send_keys(Keys.F1).perform()
         try:
-            command_box = find_element_across_iframes(
-                driver,
-                "//li[@class='action-item command-center-center']",
-            )
-            command_input = click_and_wait(
-                driver,
-                command_box,
-                "//input[@aria-controls='quickInput_list']",
-                timeout=2,
+            command_input = WebDriverWait(driver, timeout=5).until(
+                expected_conditions.visibility_of_element_located(
+                    (By.XPATH, "//input[@aria-controls='quickInput_list']"),
+                ),
             )
             break
-        except (
-            TimeoutException,
-            TimeOutError,
-            NoSuchElementException,
-            StaleElementReferenceException,
-        ):  # pragma: no cover
-            if i == max_attempts - 1:
-                raise
-            time.sleep(1)
+        except TimeoutException:  # pragma: no cover
+            log.warning(
+                "Command palette did not open on attempt %s/%s",
+                i + 1,
+                max_attempts,
+            )
+            ActionChains(driver).send_keys(Keys.ESCAPE).perform()
+            time.sleep(0.5)
 
-    if command_input:
-        command_input.send_keys(command)
-        # Let the palette finish filtering results before pressing Enter,
-        # otherwise the keystroke can arrive before a match is highlighted.
-        time.sleep(0.5)
+    if not command_input:
+        msg = f"Command palette failed to open after {max_attempts} attempts"
+        raise TimeoutException(msg)
+
+    command_input.send_keys(command)
+    time.sleep(0.5)
+    log.info("Command entered, pressing Enter")
     actions = ActionChains(driver)
     actions.send_keys(Keys.ENTER).perform()
     if command_param: