starlingx
/
config
Code Issues Proposed changes

Merge "AppFwk: remove -f prevents remove-failed state"

This commit is contained in:
Zuul 2022-09-14 13:57:29 +00:00 committed by Gerrit Code Review
parent 22555270af 7f8a5fb242
commit 7782f877a0
5 changed files with 44 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
sysinv/sysinv/sysinv/sysinv/api/controllers/v1/kube_app.py
View File

@ -378,7 +378,8 @@ class KubeAppController(rest.RestController):
lifecycle_hook_info.mode = constants.APP_LIFECYCLE_MODE_MANUAL
pecan.request.rpcapi.perform_app_remove(pecan.request.context,
db_app, lifecycle_hook_info=lifecycle_hook_info)
db_app, lifecycle_hook_info=lifecycle_hook_info,
force=force)
else:
if db_app.status not in [constants.APP_APPLY_IN_PROGRESS,
constants.APP_UPDATE_IN_PROGRESS,

4
sysinv/sysinv/sysinv/sysinv/common/constants.py
View File

@ -1844,7 +1844,7 @@ APP_EVALUATE_REAPPLY_TRIGGER_TO_METADATA_MAP = {
}
# Progress constants
APP_PROGRESS_ABORTED = 'operation aborted, check logs for detail'
APP_PROGRESS_ABORTED = 'operation aborted, check system inventory logs for details'
APP_PROGRESS_ABORTED_BY_USER = 'operation aborted by user'
APP_PROGRESS_APPLY_MANIFEST = 'applying application manifest'
APP_PROGRESS_COMPLETED = 'completed'
@ -1870,6 +1870,8 @@ APP_PROGRESS_UPDATE_FAILED_SKIP_RECOVERY = "Application {} update from " \
APP_PROGRESS_UPDATE_FAILED_ARMADA_TO_FLUXCD = "Application {} update from " \
"version {} to version {} failed and recovery skipped " \
"because recovering between Armada and FluxCD is not allowed"
APP_PROGRESS_REMOVE_FAILED_WARNING = "Application remove failed. Status forced to '{}'. " \
"Use native helm commands to clean up application helm releases."
# Auto-recovery limits
APP_AUTO_RECOVERY_MAX_COUNT = 5

45
sysinv/sysinv/sysinv/sysinv/conductor/kube_app.py
View File

@ -331,7 +331,8 @@ class AppOperator(object):
def _abort_operation(self, app, operation,
progress=constants.APP_PROGRESS_ABORTED,
user_initiated=False, reset_status=False):
user_initiated=False, reset_status=False,
forced_operation=False):
if user_initiated:
progress = constants.APP_PROGRESS_ABORTED_BY_USER
@ -377,17 +378,28 @@ class AppOperator(object):
"please check system inventory log for cause."),
True)
elif app.status == constants.APP_REMOVE_IN_PROGRESS:
new_status = constants.APP_REMOVE_FAILURE
op = 'application-remove'
self._raise_app_alarm(
app.name, constants.APP_REMOVE_FAILURE,
fm_constants.FM_ALARM_ID_APPLICATION_REMOVE_FAILED,
fm_constants.FM_ALARM_SEVERITY_MAJOR,
_("Application Remove Failure"),
fm_constants.FM_ALARM_TYPE_3,
_("Retry removing the application. If the issue persists, "
"please check system inventory log for cause."),
True)
if not forced_operation:
new_status = constants.APP_REMOVE_FAILURE
self._raise_app_alarm(
app.name, constants.APP_REMOVE_FAILURE,
fm_constants.FM_ALARM_ID_APPLICATION_REMOVE_FAILED,
fm_constants.FM_ALARM_SEVERITY_MAJOR,
_("Application Remove Failure"),
fm_constants.FM_ALARM_TYPE_3,
_("Retry removing the application. If the issue persists, "
"please check system inventory log for cause. "
"Using --force will set the app status to 'uploaded' "
"in case the error persists."),
True)
else:
# In case there is an existing alarm for previous remove failure
self._clear_app_alarm(app.name)
new_status = constants.APP_UPLOAD_SUCCESS
progress = constants.APP_PROGRESS_REMOVE_FAILED_WARNING.format(new_status)
LOG.warning(progress)
else:
# Should not get here, perhaps a new status was introduced?
LOG.error("No abort handling code for app status = '%s'!" % app.status)
@ -3060,7 +3072,7 @@ class AppOperator(object):
self._clear_app_alarm(to_app.name)
return True
def perform_app_remove(self, rpc_app, lifecycle_hook_info_app_remove):
def perform_app_remove(self, rpc_app, lifecycle_hook_info_app_remove, force=False):
"""Process application remove request
This method invokes Armada to delete the application manifest.
@ -3068,6 +3080,8 @@ class AppOperator(object):
:param rpc_app: application object in the RPC request
:param lifecycle_hook_info_app_remove: LifecycleHookInfo object
:param force: If set to True, will set the app state to 'uploaded'
instead of 'remove-failed' in case of an error
:return boolean: whether application remove was successful
"""
@ -3110,7 +3124,7 @@ class AppOperator(object):
self.app_lifecycle_actions(None, None, rpc_app, lifecycle_hook_info_app_remove)
except Exception as e:
self._abort_operation(app, constants.APP_REMOVE_OP)
self._abort_operation(app, constants.APP_REMOVE_OP, forced_operation=force)
LOG.exception(e)
self._deregister_app_abort(app.name)
return False
@ -3122,10 +3136,11 @@ class AppOperator(object):
LOG.info("Application (%s) remove completed." % app.name)
else:
if AppOperator.is_app_aborted(app.name):
# App is always set to APP_REMOVE_FAILURE if abort is initiated by user.
self._abort_operation(app, constants.APP_REMOVE_OP,
user_initiated=True)
user_initiated=True, forced_operation=False)
else:
self._abort_operation(app, constants.APP_REMOVE_OP)
self._abort_operation(app, constants.APP_REMOVE_OP, forced_operation=force)
rc = False
self._deregister_app_abort(app.name)

6
sysinv/sysinv/sysinv/sysinv/conductor/manager.py
View File

@ -14167,12 +14167,14 @@ class ConductorManager(service.PeriodicService):
self._app.perform_app_update(from_rpc_app, to_rpc_app, tarfile,
operation, lifecycle_hook_info_app_update, reuse_user_overrides)
def perform_app_remove(self, context, rpc_app, lifecycle_hook_info_app_remove):
def perform_app_remove(self, context, rpc_app, lifecycle_hook_info_app_remove, force=False):
"""Handling of application removal request (via AppOperator)
:param context: request context.
:param rpc_app: data object provided in the rpc request
:param lifecycle_hook_info_app_remove: LifecycleHookInfo object
:param force: If set to True, will set the app state to 'uploaded'
instead of 'remove-failed' in case of an error
"""
lifecycle_hook_info_app_remove.operation = constants.APP_REMOVE_OP
@ -14190,7 +14192,7 @@ class ConductorManager(service.PeriodicService):
LOG.error("Error performing app_lifecycle_actions %s" % str(e))
app_removed = self._app.perform_app_remove(
rpc_app, lifecycle_hook_info_app_remove)
rpc_app, lifecycle_hook_info_app_remove, force)
lifecycle_hook_info_app_remove[LifecycleConstants.EXTRA][LifecycleConstants.APP_REMOVED] = app_removed
# Perform post remove operation actions

7
sysinv/sysinv/sysinv/sysinv/conductor/rpcapi.py
View File

@ -1793,18 +1793,21 @@ class ConductorAPI(sysinv.openstack.common.rpc.proxy.RpcProxy):
lifecycle_hook_info_app_update=lifecycle_hook_info,
reuse_user_overrides=reuse_user_overrides))
def perform_app_remove(self, context, rpc_app, lifecycle_hook_info):
def perform_app_remove(self, context, rpc_app, lifecycle_hook_info, force=False):
"""Handle application remove request
:param context: request context.
:param rpc_app: data object provided in the rpc request
:param lifecycle_hook_info: LifecycleHookInfo object
:param force: If set to True, will set the app state to 'uploaded'
instead of 'remove-failed' in case of an error
"""
return self.cast(context,
self.make_msg('perform_app_remove',
rpc_app=rpc_app,
lifecycle_hook_info_app_remove=lifecycle_hook_info))
lifecycle_hook_info_app_remove=lifecycle_hook_info,
force=force))
def perform_app_abort(self, context, rpc_app, lifecycle_hook_info):
"""Handle application abort request