Cleaned simple kubernetes manager

Dockerfile

@@ -4,15 +4,20 @@ FROM python:3.11 AS builder
 WORKDIR /build
 COPY . .
 
-RUN pip install .
+RUN apt-get update && apt-get install -y --no-install-recommends \
+ build-essential gcc python3-dev pkg-config && \
+ rm -rf /var/lib/apt/lists/* && \
+ pip install --upgrade pip setuptools wheel build
 
-RUN python3 setup.py sdist && \
- pip install dist/*.tar.gz --target /install
+RUN python -m build && \
+ pip install dist/*.tar.gz --prefix=/install
 
-# minimal runtime image
+# runtime stage
 FROM python:3.11-slim AS runtime
 
-COPY --from=builder /install /usr/local/lib/python3.11/site-packages
+# copy installed Python libs + CLI scripts
+COPY --from=builder /install /usr/local
+
 COPY etc/*.json etc/*.yaml /etc/villas/controller/
 COPY villas-controller.service /etc/systemd/system/
 

etc/config_simplekub.yaml

@@ -0,0 +1,14 @@
+---
+components:
+- type: generic
+ category: manager
+ name: Generic Manager
+ location: VM Iris
+ uuid: eddb51a0-557b-4848-ac7a-faccc7c51fa3
+
+- category: manager
+ type: kubernetes-simple
+ name: Simple Kubernetes Manager
+ location: VM Iris
+ uuid: 4f8fb73e-7e74-11eb-8f63-f3ccc3ab82f6
+ namespace: villas-controller

etc/params_k8s_dpsim.yaml

@@ -13,9 +13,9 @@ properties:
  name: dpsim
  spec:
  suspend: true
- activeDeadlineSeconds: 120 # kill the Job after 1h
+ activeDeadlineSeconds: 3600 # kill the Job after 1h
  backoffLimit: 0 # only try to run pod once, no retries
- ttlSecondsAfterFinished: 120 # delete the Job resources 1h after completion
+ ttlSecondsAfterFinished: 3600 # delete the Job resources 1h after completion
  template:
  spec:
  restartPolicy: Never

villas/controller/component.py

@@ -107,7 +107,11 @@ def load_schema(self):
  fo = resources.open_text(pkg, res)
  loadedschema = yaml.load(fo, yaml.SafeLoader)
 
- schema[name] = Draft202012Validator(loadedschema)
+ try:
+ Draft202012Validator.check_schema(loadedschema)
+ schema[name] = loadedschema
+ except jsonschema.exceptions.SchemaError:
+ self.logger.warning("Schema is invalid!")
 
  return schema
 
@@ -148,7 +152,7 @@ def status(self):
  **self.headers
  },
  'schema': {
- name: v.schema for name, v in self.schema.items()
+ name: v for name, v in self.schema.items()
  }
  }
 
@@ -277,12 +281,15 @@ def from_dict(dict):
 
  def publish_status(self):
  if not self.mixin:
+ self.logger.warn('No mixin!')
  return
 
  self.mixin.publish(self.status, headers=self.headers)
 
  def publish_status_periodically(self):
- self.logger.info('Start state publish thread')
+ self.logger.info('Start state publish thread, initial status: %s',
+ self.status)
+ self.publish_status() # publish the first update immediately
 
  while not self.publish_status_thread_stop.wait(
  self.publish_status_interval):

villas/controller/components/manager.py

@@ -28,6 +28,9 @@ def from_dict(dict):
  if type == 'kubernetes':
  from villas.controller.components.managers import kubernetes
  return kubernetes.KubernetesManager(**dict)
+ if type == 'kubernetes-simple':
+ from villas.controller.components.managers import kubernetes_simple
+ return kubernetes_simple.KubernetesManagerSimple(**dict)
  if type == 'villas-node':
  from villas.controller.components.managers import villas_node # noqa E501
  return villas_node.VILLASnodeManager(**dict)
@@ -43,7 +46,6 @@ def from_dict(dict):
  def add_component(self, comp):
  if comp.uuid in self.mixin.components:
  existing_comp = self.mixin.components[comp.uuid]
-
  raise SimulationException(self, 'Component with same UUID ' +
  'already exists!',
  component=existing_comp)

villas/controller/components/managers/kubernetes.py

@@ -22,30 +22,30 @@ class KubernetesManager(Manager):
  def __init__(self, **args):
  super().__init__(**args)
 
- self.thread_stop = threading.Event()
-
- self.pod_watcher_thread = threading.Thread(
- target=self._run_pod_watcher)
- self.job_watcher_thread = threading.Thread(
- target=self._run_job_watcher)
- self.event_watcher_thread = threading.Thread(
- target=self._run_event_watcher)
-
  if os.environ.get('KUBECONFIG'):
  k8s.config.load_kube_config()
  else:
  k8s.config.load_incluster_config()
 
- self.namespace = args.get('namespace', 'default')
+ # the namespace in which to create the jobs
+ # and to watch for events
+ self.namespace = os.environ.get('NAMESPACE')
+ if self.namespace:
+ self.namespace = ''.join([self.namespace, '-controller'])
+ else:
+ self.namespace = 'villas-controller'
+
+ self._check_namespace(self.namespace)
 
- self.my_namespace = os.environ.get('NAMESPACE')
+ # name and UID of the pod in which this controller is running
+ # used in kubernetes simulator to set the owner reference
  self.my_pod_name = os.environ.get('POD_NAME')
  self.my_pod_uid = os.environ.get('POD_UID')
 
- self._check_namespace(self.namespace)
+ self.thread_stop = threading.Event()
 
-# self.pod_watcher_thread.start()
-# self.job_watcher_thread.start()
+ self.event_watcher_thread = threading.Thread(
+ target=self._run_event_watcher)
  self.event_watcher_thread.setDaemon(True)
  self.event_watcher_thread.start()
 
@@ -59,28 +59,6 @@ def _check_namespace(self, ns):
 
  raise RuntimeError(f'Namespace {ns} does not exist')
 
- def _run_pod_watcher(self):
- w = k8s.watch.Watch()
- c = k8s.client.CoreV1Api()
-
- for sts in w.stream(c.list_namespaced_pod,
- namespace=self.namespace):
- stso = sts.get('object')
- typ = sts.get('type')
-
- self.logger.info('%s Pod: %s', typ, stso.metadata.name)
-
- def _run_job_watcher(self):
- w = k8s.watch.Watch()
- b = k8s.client.BatchV1Api()
-
- for sts in w.stream(b.list_namespaced_job,
- namespace=self.namespace):
- stso = sts.get('object')
- typ = sts.get('type')
-
- self.logger.info('%s Job: %s', typ, stso.metadata.name)
-
  def _run_event_watcher(self):
  while not self.thread_stop.is_set():
  w = k8s.watch.Watch()
@@ -107,6 +85,10 @@ def _run_event_watcher(self):
 
  if _match(comp.job.metadata.name,
  eo.involved_object.name):
+ if comp._state == 'stopping':
+ # incoming events are old repetitions
+ continue
+
  if eo.reason == 'Completed':
  comp.change_state('stopping', True)
  elif eo.reason == 'Started':

villas/controller/components/managers/kubernetes_simple.py

@@ -0,0 +1,79 @@
+from villas.controller.components.managers.kubernetes import KubernetesManager
+from villas.controller.components.simulators.kubernetes import KubernetesJob
+
+
+def build_parameters(sim_name, jobname, image, adls=3600,
+ privileged=False, name=None, uuid=None):
+ parameters = {
+ 'type': 'kubernetes',
+ 'category': 'simulator',
+ 'uuid': uuid,
+ 'name': name or sim_name,
+ 'properties': {
+ 'job': {
+ 'apiVersion': 'batch/v1',
+ 'kind': 'Job',
+ 'metadata': {
+ 'name': jobname
+ },
+ 'spec': {
+ 'activeDeadlineSeconds': adls,
+ 'backoffLimit': 2,
+ 'template': {
+ 'spec': {
+ 'restartPolicy': 'Never',
+ 'containers': [
+ {
+ 'image': image,
+ 'imagePullPolicy': 'Always',
+ 'name': 'jobcontainer',
+ 'securityContext': {
+ 'privileged': privileged
+ }
+ }
+ ]
+ }
+ }
+ }
+ }
+ }
+ }
+ return parameters
+
+
+class KubernetesManagerSimple(KubernetesManager):
+
+ def __init__(self, **args):
+ super().__init__(**args)
+
+ def create(self, payload):
+ params = payload.get('parameters', {})
+ sim_name = payload.get('name', 'Kubernetes Simulator')
+ jobname = params.get('jobname', 'noname')
+ adls = params.get('activeDeadlineSeconds', 3600)
+ if type(adls) is str:
+ adls = int(adls)
+ image = params.get('image')
+ name = params.get('name')
+ privileged = params.get('privileged', False)
+ uuid = params.get('uuid')
+ self.logger.info('uuid:')
+ self.logger.info(uuid)
+
+ if image is None:
+ self.logger.error('No image given, will try super.create')
+ super().create(payload)
+ return
+
+ parameters = build_parameters(
+ sim_name=sim_name,
+ jobname=jobname,
+ image=image,
+ adls=adls,
+ privileged=privileged,
+ name=name,
+ uuid=uuid
+ )
+
+ comp = KubernetesJob(self, **parameters)
+ self.add_component(comp)

villas/controller/components/simulators/kubernetes.py

@@ -106,8 +106,8 @@ def _prepare_job(self, job, payload):
  job.metadata.generate_name = name + '-'
  job.metadata.name = None
 
- if o := self._owner():
- job.metadata.owner_references = [o]
+# if o := self._owner():
+# job.metadata.owner_references = [o]
 
  if job.metadata.labels is None:
  job.metadata.labels = {}
@@ -144,8 +144,8 @@ def _create_config_map(self, payload):
  }
  )
 
- if o := self._owner():
- self.cm.metadata.owner_references = [o]
+# if o := self._owner():
+# self.cm.metadata.owner_references = [o]
 
  return c.create_namespaced_config_map(
  namespace=self.manager.namespace,
@@ -165,8 +165,12 @@ def _delete_job(self):
  name=self.job.metadata.name,
  body=body)
  except k8s.client.exceptions.ApiException as e:
- raise SimulationException(self, 'Kubernetes API error',
- error=str(e))
+ if e.status == 404:
+ # Job does not exist, treat as already deleted
+ return
+ else:
+ raise SimulationException(self, 'Kubernetes API exception',
+ error=str(e))
 
  self.pods.clear()
 
@@ -194,9 +198,9 @@ def start(self, payload):
  self.properties['job_name'] = self.job.metadata.name
  self.properties['namespace'] = self.manager.namespace
 
- def stop(self, payload):
+ def stop(self, message):
+ self.change_state('stopping', True)
  self._delete_job()
-
  self.change_state('idle')
 
  def _send_signal(self, sig):
@@ -227,6 +231,7 @@ def resume(self, payload):
  self.change_state('running')
 
  def reset(self, payload):
+ self.change_state('resetting', True)
  self._delete_job()
  super().reset(payload)
 

villas/controller/controller.py

@@ -84,6 +84,8 @@ def _drain_publish_queue(self):
  self.producer.publish(body, **kwargs)
  except queue.Empty:
  pass
+ except TimeoutError:
+ LOGGER.warn('TimeoutError, let kombu reconnect..')
 
  def on_iteration(self):
  # Drain publish queue

villas/controller/schemas/manager/kubernetes-simple/create.yaml

@@ -0,0 +1,33 @@
+---
+$schema: http://json-schema.org/draft-04/schema#
+
+type: object
+title: 'Simple Kubernetes Job'
+required:
+ - image
+properties:
+ name:
+ type: string
+ title: 'Simulator Name'
+ default: 'Kubernetes Simulator'
+ uuid:
+ type: string
+ title: UUID
+ default: 8dfd03b2-1c78-11ec-9621-0242ac130002
+ jobname:
+ type: string
+ title: 'Jobname'
+ default: myjob
+ activeDeadlineSeconds:
+ type: number
+ title: activeDeadlineSeconds
+ default: 3600
+ image:
+ type: string
+ title: Image
+ default: perl
+ privileged:
+ type: boolean
+ title: Privileged
+ default: false
+ description: 'WARNING: If true, the container has root privileges on the host'