Integration test for bluechi-is-online agent --wait

1. Start agent and verify --wait option returns 0 immediately.
2. Stop agent and verify --wait returns 1 after wait time is expired.
3. Stop agent, run 'bluechi-is-online agent --wait' start agent and verify --wait returns 0 before the wait time expires.

Signed-off-by: Dana Orr <[email protected]>

tests/bluechi_test/bluechi_is_online.py

@@ -26,10 +26,15 @@ def run(
             )
         return result, output
 
-    def agent_is_online(self) -> bool:
+    def agent_is_online(self, wait: int = None) -> bool:
+        cmd = ["agent"]
+
+        if wait:
+            cmd.extend(["--wait", str(wait)])
+
         result, output = self.run(
             "Checking if agent is active.",
-            "agent",
+            " ".join(cmd),
             False,
             0,
         )

tests/tests/tier0/bluechi-is-online-agent-wait/main.fmf

@@ -0,0 +1,2 @@
+summary: test bluechi-is-online agent --wait command
+id: afde18db-d923-4596-8a2f-c0eb8fab2a15

tests/tests/tier0/bluechi-is-online-agent-wait/test_bluechi_is_online_agent_wait.py

@@ -0,0 +1,125 @@
+#
+# Copyright Contributors to the Eclipse BlueChi project
+#
+# SPDX-License-Identifier: LGPL-2.1-or-later
+
+import logging
+import threading
+import time
+from typing import Dict
+
+from bluechi_test.bluechi_is_online import BluechiIsOnline
+from bluechi_test.config import BluechiAgentConfig, BluechiControllerConfig
+from bluechi_test.machine import BluechiAgentMachine, BluechiControllerMachine
+from bluechi_test.test import BluechiTest
+from bluechi_test.util import Timeout, get_test_env_value_int
+
+LOGGER = logging.getLogger(__name__)
+
+NODE_FOO = "node-foo"
+IMMEDIATE_RETURN_TIMEOUT = get_test_env_value_int("IMMEDIATE_RETURN_TIMEOUT", 1000)
+WAIT_PARAM_VALUE = get_test_env_value_int("WAIT_PARAM_VALUE", 5000)
+SLEEP_DURATION = get_test_env_value_int("SLEEP_DURATION", 2)
+
+
+class ResultFuture:
+    def __init__(self):
+        self.result = None
+        self.output = ""
+
+
+def check_agent(
+    bluechi_is_online: BluechiIsOnline,
+    future: ResultFuture,
+    wait_time: int = 0,
+):
+    result = bluechi_is_online.agent_is_online(
+        wait_time,
+    )
+    future.result = result
+
+
+def exec(ctrl: BluechiControllerMachine, nodes: Dict[str, BluechiAgentMachine]):
+
+    node_foo = nodes[NODE_FOO]
+
+    # Test 1: Start agent and verify --wait option returns 0 immediately
+    LOGGER.debug("Starting test number 1 - agent should be online.")
+    with Timeout(
+        IMMEDIATE_RETURN_TIMEOUT, "bluechi-is-online didn't return immediately"
+    ):
+        assert node_foo.bluechi_is_online.agent_is_online(wait=WAIT_PARAM_VALUE)
+
+    # Test 2: Stop agent and verify --wait returns 1 after wait time is expired
+    node_foo.systemctl.stop_unit("bluechi-agent")
+    assert node_foo.wait_for_unit_state_to_be("bluechi-agent", "inactive")
+    LOGGER.debug("Starting test number 2 - agent should remain offline.")
+
+    start_time = time.time()
+    result = node_foo.bluechi_is_online.agent_is_online(wait=WAIT_PARAM_VALUE)
+
+    assert (
+        not result
+    ), f"Expected bluechi-is-online with --wait={WAIT_PARAM_VALUE} to exit false"
+    assert (
+        time.time() - start_time <= WAIT_PARAM_VALUE + 1
+    ), "Expected around 5 second for bluechi-is-online to exit"
+
+    # Test 3: Stop agent, run 'bluechi-is-online agent --wait', start agent and verify --wait returns 0 before the
+    # wait time expires
+    LOGGER.debug(
+        "Starting test number 3, ensure agent is inactive before starting `bluechi-is-online`."
+    )
+    with Timeout(WAIT_PARAM_VALUE, "Timeout during Test 3"):
+        assert node_foo.wait_for_unit_state_to_be("bluechi-agent", "inactive")
+        LOGGER.debug("Agent confirmed inactive after stopping.")
+
+        result_future_wait = ResultFuture()
+        start_time = time.time()
+        LOGGER.debug(
+            "Starting `bluechi-is-online` thread with wait time of 20 seconds."
+        )
+        check_thread_wait = threading.Thread(
+            target=check_agent,
+            args=(
+                node_foo.bluechi_is_online,
+                result_future_wait,
+                WAIT_PARAM_VALUE,
+            ),
+        )
+        check_thread_wait.start()
+        time.sleep(SLEEP_DURATION)
+        node_foo.systemctl.start_unit("bluechi-agent")
+        assert node_foo.wait_for_unit_state_to_be("bluechi-agent", "active")
+        LOGGER.debug("Agent confirmed active after starting.")
+
+        # Wait for the thread to complete
+        with Timeout(2, "Thread didn't stop within 2s of starting bluechi-agent"):
+            check_thread_wait.join()
+
+        elapsed_time = time.time() - start_time
+        LOGGER.debug(
+            f"Test 3 result: {result_future_wait.result}, Elapsed time: {elapsed_time:.2f} seconds"
+        )
+        assert (
+            result_future_wait.result
+        ), "Expected agent to come online before wait expired"
+        assert (
+            2 < elapsed_time < 3
+        ), f"Command took too long to complete: {elapsed_time:.2f} seconds"
+
+
+def test_bluechi_is_online_agent_wait(
+    bluechi_test: BluechiTest,
+    bluechi_node_default_config: BluechiAgentConfig,
+    bluechi_ctrl_default_config: BluechiControllerConfig,
+):
+    node_bar_cfg = bluechi_node_default_config.deep_copy()
+    node_bar_cfg.node_name = NODE_FOO
+
+    bluechi_ctrl_default_config.allowed_node_names = [NODE_FOO]
+
+    bluechi_test.set_bluechi_controller_config(bluechi_ctrl_default_config)
+    bluechi_test.add_bluechi_agent_config(node_bar_cfg)
+
+    bluechi_test.run(exec)