From 4287951d711b94972abd13bac6dcbd7250e0867e Mon Sep 17 00:00:00 2001
From: Derek Higgins <derekh@redhat.com>
Date: Wed, 3 Feb 2021 16:00:57 +0000
Subject: [PATCH] Don't mark an agent as alive if rebooted

If 'agent_url' has been cleared from internal_info
it indicates that the node has been powered off.

Change-Id: Idba486c98e1e92d35fca2e2d156866566acb9e40
Story: 2008583
Task: 41736
---
 ironic/conductor/utils.py                             |  5 +++++
 ironic/tests/unit/conductor/test_utils.py             | 11 ++++++++++-
 .../notes/agent-rebooted-fab20d012fe6cbe8.yaml        |  6 ++++++
 3 files changed, 21 insertions(+), 1 deletion(-)
 create mode 100644 releasenotes/notes/agent-rebooted-fab20d012fe6cbe8.yaml

diff --git a/ironic/conductor/utils.py b/ironic/conductor/utils.py
index d0e2bbd157..f251dbff91 100644
--- a/ironic/conductor/utils.py
+++ b/ironic/conductor/utils.py
@@ -1042,6 +1042,11 @@ def agent_is_alive(node, timeout=None):
     :param node: A node object.
     :param timeout: Heartbeat timeout, defaults to `fast_track_timeout`.
     """
+    # If no agent_url is present then we have powered down since the
+    # last agent heartbeat
+    if not node.driver_internal_info.get('agent_url'):
+        return False
+
     return value_within_timeout(
         node.driver_internal_info.get('agent_last_heartbeat'),
         timeout or CONF.deploy.fast_track_timeout)
diff --git a/ironic/tests/unit/conductor/test_utils.py b/ironic/tests/unit/conductor/test_utils.py
index 8d0bf5f35c..5954ef266a 100644
--- a/ironic/tests/unit/conductor/test_utils.py
+++ b/ironic/tests/unit/conductor/test_utils.py
@@ -1934,7 +1934,8 @@ class FastTrackTestCase(db_base.DbTestCase):
             self.context, driver='fake-hardware',
             uuid=uuidutils.generate_uuid(),
             driver_internal_info={
-                'agent_last_heartbeat': str(timeutils.utcnow().isoformat())})
+                'agent_last_heartbeat': str(timeutils.utcnow().isoformat()),
+                'agent_url': 'a_url'})
         self.config(fast_track=True, group='deploy')
 
     def test_is_fast_track(self, mock_get_power):
@@ -1966,6 +1967,14 @@ class FastTrackTestCase(db_base.DbTestCase):
                 self.context, self.node.uuid, shared=False) as task:
             self.assertFalse(conductor_utils.is_fast_track(task))
 
+    def test_is_fast_track_powered_after_heartbeat(self, mock_get_power):
+        mock_get_power.return_value = states.POWER_ON
+        with task_manager.acquire(
+                self.context, self.node.uuid, shared=False) as task:
+            conductor_utils.node_power_action(task, states.POWER_OFF)
+            conductor_utils.node_power_action(task, states.POWER_ON)
+            self.assertFalse(conductor_utils.is_fast_track(task))
+
     def test_is_fast_track_error_blocks(self, mock_get_power):
         mock_get_power.return_value = states.POWER_ON
         self.node.last_error = "bad things happened"
diff --git a/releasenotes/notes/agent-rebooted-fab20d012fe6cbe8.yaml b/releasenotes/notes/agent-rebooted-fab20d012fe6cbe8.yaml
new file mode 100644
index 0000000000..863eb5191f
--- /dev/null
+++ b/releasenotes/notes/agent-rebooted-fab20d012fe6cbe8.yaml
@@ -0,0 +1,6 @@
+---
+fixes:
+  - |
+    Fixes fast-track to prevent marking the agent as alive if
+    trying to rebuild a node before the fast-track timeout has
+    expired.