1
0
mirror of https://github.com/ansible/awx.git synced 2024-11-01 16:51:11 +03:00

Add ordering and various filtering for job events as requested by AC-1349

This commit is contained in:
Matthew Jones 2014-06-25 14:26:35 -04:00
parent e4fd79f653
commit 053bb4eeb0
2 changed files with 37 additions and 3 deletions

View File

@ -29,14 +29,21 @@ def paginated(method):
limit = min(api_settings.MAX_PAGINATE_BY, limit)
limit = int(limit)
# Get the order parameter if it's given
if request.QUERY_PARAMS.get("ordering", False):
ordering = request.QUERY_PARAMS[ordering]
else:
ordering = None
# What page are we on?
page = int(request.QUERY_PARAMS.get('page', 1))
offset = (page - 1) * limit
# Add the limit, offset, and page variables to the keyword arguments
# Add the limit, offset, page, and order variables to the keyword arguments
# being sent to the underlying method.
kwargs['limit'] = limit
kwargs['offset'] = offset
kwargs['ordering'] = ordering
# Okay, call the underlying method.
results, count = method(self, request, *args, **kwargs)
@ -64,3 +71,4 @@ def paginated(method):
# Okay, we're done; return response data.
return Response(answer)
return func

View File

@ -1486,12 +1486,26 @@ class JobJobPlaysList(BaseJobEventsList):
new_in_150 = True
@paginated
def get(self, request, limit, offset, *args, **kwargs):
def get(self, request, limit, offset, ordering, *args, **kwargs):
all_plays = []
job = get_object_or_404(self.parent_model, pk=self.kwargs['pk'])
# Put together a queryset for relevant job events.
qs = job.job_events.filter(event='playbook_on_play_start')
if ordering is not None:
qs = qs.order_by(ordering)
# This is a bit of a special case for id filtering requested by the UI
# doing this here for the moment until/unless we need to implement more
# complex filtering (since we aren't under a serializer)
if "id__in" in request.QUERY_PARAMS:
qs = qs.filter(id__in=[int(filter_id) for filter_id in request.QUERY_PARAMS["id__in"].split(",")])
elif "id__gt" in request.QUERY_PARAMS:
qs = qs.filter(id__gt=request.QUERY_PARAMS['id__gt'])
if "failed" in request.QUERY_PARAMS:
qs = qs.filter(failed=(request.QUERY_PARAMS['failed'].lower() == 'true'))
count = qs.count()
# Iterate over the relevant play events and get the details.
@ -1540,7 +1554,7 @@ class JobJobTasksList(BaseJobEventsList):
new_in_150 = True
@paginated
def get(self, request, limit, offset, *args, **kwargs):
def get(self, request, limit, offset, ordering, *args, **kwargs):
"""Return aggregate data about each of the job tasks that is:
- an immediate child of the job event
- corresponding to the spinning up of a new task or playbook
@ -1570,6 +1584,18 @@ class JobJobTasksList(BaseJobEventsList):
.values('parent__id', 'event', 'changed')
.annotate(num=Count('event'))
.order_by('parent__id'))
# This is a bit of a special case for id filtering requested by the UI
# doing this here for the moment until/unless we need to implement more
# complex filtering (since we aren't under a serializer)
if "id__in" in request.QUERY_PARAMS:
qs = qs.filter(id__in=[int(filter_id) for filter_id in request.QUERY_PARAMS["id__in"].split(",")])
elif "id__gt" in request.QUERY_PARAMS:
qs = qs.filter(id__gt=request.QUERY_PARAMS['id__gt'])
if "failed" in request.QUERY_PARAMS:
qs = qs.filter(failed=(request.QUERY_PARAMS['failed'].lower() == 'true'))
count = queryset.count()
# The data above will come back in a list, but we are going to