Merge "Do not silently ignore exceptions when running next steps"

ironic/conductor/cleaning.py

@@ -126,6 +126,8 @@ def do_node_clean(task, clean_steps=None):
     do_next_clean_step(task, step_index)
 
 
+@utils.fail_on_error(utils.deploying_error_handler,
+                     _("Unexpected error when processing next clean step"))
 @task_manager.require_exclusive_lock
 def do_next_clean_step(task, step_index):
     """Do cleaning, starting from the specified clean step.

ironic/conductor/deployments.py

@@ -204,6 +204,8 @@ def do_node_deploy(task, conductor_id=None, configdrive=None):
     do_next_deploy_step(task, 0, conductor_id)
 
 
+@utils.fail_on_error(utils.deploying_error_handler,
+                     _("Unexpected error when processing next deploy step"))
 @task_manager.require_exclusive_lock
 def do_next_deploy_step(task, step_index, conductor_id):
     """Do deployment, starting from the specified deploy step.

ironic/conductor/utils.py

@@ -16,6 +16,7 @@ import contextlib
 import crypt
 import datetime
 from distutils.version import StrictVersion
+import functools
 import os
 import secrets
 import time
@@ -563,6 +564,21 @@ def deploying_error_handler(task, logmsg, errmsg=None, traceback=False,
     task.process_event('fail')
 
 
+def fail_on_error(error_callback, msg, *error_args, **error_kwargs):
+    """A decorator for failing operation on failure."""
+    def wrapper(func):
+        @functools.wraps(func)
+        def wrapped(task, *args, **kwargs):
+            try:
+                return func(task, *args, **kwargs)
+            except Exception as exc:
+                errmsg = "%s. %s: %s" % (msg, exc.__class__.__name__, exc)
+                error_callback(task, errmsg, *error_args, **error_kwargs)
+
+        return wrapped
+    return wrapper
+
+
 @task_manager.require_exclusive_lock
 def abort_on_conductor_take_over(task):
     """Set node's state when a task was aborted due to conductor take over.

releasenotes/notes/spawn-error-2249f94606388fbd.yaml

@@ -0,0 +1,5 @@
+---
+fixes:
+  - |
+    No longer silently ignores exceptions that happen when trying to run the
+    next clean or deploy step.