django-orchestra/orchestra/contrib/resources/tasks.py

53 lines
1.9 KiB
Python

from celery import shared_task
from orchestra.contrib.orchestration import Operation
from orchestra.models.utils import get_model_field_path
from .backends import ServiceMonitor
@shared_task(name='resources.Monitor')
def monitor(resource_id, ids=None, async=True):
from .models import ResourceData, Resource
resource = Resource.objects.get(pk=resource_id)
resource_model = resource.content_type.model_class()
logs = []
# Execute monitors
for monitor_name in resource.monitors:
backend = ServiceMonitor.get_backend(monitor_name)
model = backend.model_class()
kwargs = {}
if ids:
path = get_model_field_path(model, resource_model)
path = '%s__in' % ('__'.join(path) or 'id')
kwargs = {
path: ids
}
# Execute monitor
monitorings = []
for obj in model.objects.filter(**kwargs):
op = Operation(backend, obj, Operation.MONITOR)
monitorings.append(op)
# TODO async=True only when running with celery
logs += Operation.execute(monitorings, async=async)
kwargs = {'id__in': ids} if ids else {}
# Update used resources and trigger resource exceeded and revovery
triggers = []
model = resource.content_type.model_class()
for obj in model.objects.filter(**kwargs):
data, __ = ResourceData.get_or_create(obj, resource)
data.update()
if not resource.disable_trigger:
a = data.used
b = data.allocated
if data.used > (data.allocated or 0):
op = Operation(backend, obj, Operation.EXCEEDED)
triggers.append(op)
elif data.used < (data.allocated or 0):
op = Operation(backend, obj, Operation.RECOVERY)
triggers.append(op)
Operation.execute(triggers)
return logs