Add manager for simple Kubernetes job creation

.gitlab-ci.yml

@@ -20,8 +20,6 @@ build:precommit:
   - pip3 install -r requirements-dev.txt
   script:
   - pre-commit run --all-files
-  tags:
-    - docker
 
 build:test:
   stage: build
@@ -33,8 +31,6 @@ build:test:
   - pip3 install -r requirements-dev.txt
   script:
   - pytest -v
-  tags:
-  - docker
 
 build:dist:
   stage: build

.pre-commit-config.yaml

@@ -18,8 +18,8 @@ repos:
   rev: 6.0.0
   hooks:
   - id: flake8
-- repo: https://github.com/sirosen/check-jsonschema
-  rev: 0.22.0
+- repo: https://github.com/python-jsonschema/check-jsonschema
+  rev: 0.32.1
   hooks:
   - id: check-jsonschema
     name: "Check schemas"
@@ -32,4 +32,4 @@ repos:
     language: python
     files: ^doc/openapi.yaml$
     types: [yaml]
-    args: ["--schemafile", "https://raw.githubusercontent.com/OAI/OpenAPI-Specification/main/schemas/v3.1/schema.json"]
+    args: ["--schemafile", "https://spec.openapis.org/oas/3.1/schema/2025-02-13"]

etc/config_simplekub.yaml

@@ -0,0 +1,14 @@
+---
+broker:
+  url: amqp://villas:Haegiethu0rohtee@kubernetes-master-1.os-cloud.eonerc.rwth-aachen.de:30809/%2F
+
+components:
+- type: generic
+  category: manager
+  uuid: ebbbbba0-557b-4848-ac7a-faa3e7c51fa3
+
+- category: manager
+  type: kubernetes-simple
+  uuid: 4bbbb73e-7e74-11eb-8f63-f3a5b3ab82f6
+
+  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

setup.py

@@ -1,14 +1,30 @@
 from setuptools import setup, find_namespace_packages
 from glob import glob
 
-from villas.controller import __version__ as version
+import os
+import re
+
+
+def get_version():
+    here = os.path.abspath(os.path.dirname(__file__))
+    init_file = os.path.join(here, "villas", "controller", "__init__.py")
+
+    with open(init_file, "r") as f:
+        content = f.read()
+
+    match = re.search(r"^__version__ = ['\"]([^'\"]*)['\"]", content, re.M)
+    if match:
+        return match.group(1)
+
+    raise RuntimeError("Version not found")
+
 
 with open('README.md') as f:
     long_description = f.read()
 
 setup(
     name='villas-controller',
-    version=version,
+    version=get_version(),
     description='A controller/orchestration API for real-time '
                 'power system simulators',
     long_description=long_description,
@@ -20,7 +36,7 @@
     keywords='simulation controller villas',
     classifiers=[
         'Development Status :: 3 - Alpha',
-        'License :: OSI Approved :: Apache Software License'
+        'License :: OSI Approved :: Apache Software License',
         'Programming Language :: Python :: 3'
     ],
     packages=find_namespace_packages(include=['villas.*']),

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
 
@@ -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

@@ -22,42 +22,36 @@ 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')
+        self.namespace = ''.join([self.namespace, '-controller'])
 
-        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 = 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)
 
-        # self.pod_watcher_thread.start()
-        # self.job_watcher_thread.start()
         self.event_watcher_thread.setDaemon(True)
         self.event_watcher_thread.start()
 
-    def _check_namespace(self, ns):
-        c = k8s.client.CoreV1Api()
-
-        namespaces = c.list_namespace()
-        for namespace in namespaces.items:
-            if namespace.metadata.name == ns:
-                return
-
-        raise RuntimeError(f'Namespace {ns} does not exist')
+        # Not used yet, can support more complex logic
+        # self.pod_watcher_thread.start()
+        # self.job_watcher_thread.start()
 
     def _run_pod_watcher(self):
         w = k8s.watch.Watch()
@@ -107,6 +101,10 @@ 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 +1,82 @@
+from villas.controller.components.managers.kubernetes import KubernetesManager
+from villas.controller.components.simulators.kubernetes import KubernetesJob
+
+parameters_simple = {
+    'type': 'kubernetes',
+    'category': 'simulator',
+    'uuid': None,
+    'name': '',
+    'properties': {
+        'job': {
+            'apiVersion': 'batch/v1',
+            'kind': 'Job',
+            'metadata': {
+                'name': ''
+            },
+            'spec': {
+                'activeDeadlineSeconds': 3600,
+                'backoffLimit': 2,
+                'template': {
+                    'spec': {
+                        'restartPolicy': 'Never',
+                        'containers': [
+                            {
+                                'image': '',
+                                'imagePullPolicy': 'Always',
+                                'name': 'jobcontainer',
+                                'securityContext': {
+                                    'privileged': False
+                                }
+                            }
+                        ]
+                    }
+                }
+            }
+        }
+    }
+}
+
+
+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 = parameters_simple
+        parameters['name'] = sim_name
+        job = parameters['properties']['job']
+        job['metadata']['name'] = jobname
+        job['spec']['activeDeadlineSeconds'] = adls
+        job_container = job['spec']['template']['spec']['containers'][0]
+        job_container['image'] = image
+        job_container['securityContext']['privileged'] = privileged
+
+        parameters['job'] = job
+
+        if name:
+            parameters['name'] = name
+
+        if uuid:
+            parameters['uuid'] = uuid
+
+        comp = KubernetesJob(self, **parameters)
+        self.add_component(comp)

villas/controller/components/simulators/kubernetes.py

@@ -2,6 +2,7 @@
 from typing import TYPE_CHECKING
 import json
 import signal
+import time
 
 import kubernetes as k8s
 
@@ -28,6 +29,7 @@ def __init__(self, manager: KubernetesManager, **args):
 
         self.job = None
         self.pods = set()
+        self.cm_name = ''
 
         self.custom_schema = props.get('schema', {})
 
@@ -65,6 +67,7 @@ def _owner(self):
     def _prepare_job(self, job, payload):
         # Create config map
         cm = self._create_config_map(payload)
+        self.cm_name = cm.metadata.name
 
         # Create volumes
         v = k8s.client.V1Volume(
@@ -173,6 +176,9 @@ def _delete_job(self):
         self.job = None
         self.properties['job_name'] = None
         self.properties['pod_names'] = []
+        # job isn't immediately deleted
+        # let the user see that something is happening
+        time.sleep(7)
 
     def start(self, payload):
         # Delete prior job
@@ -194,9 +200,9 @@ 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 +233,8 @@ def resume(self, payload):
         self.change_state('running')
 
     def reset(self, payload):
+        self.change_state('resetting', True)
+        self.mixin.drain_publish_queue()
         self._delete_job()
         super().reset(payload)
 

villas/controller/controller.py

@@ -72,7 +72,7 @@ def add_managers(self):
     def publish(self, body, **kwargs):
         self.publish_queue.put((body, kwargs))
 
-    def _drain_publish_queue(self):
+    def drain_publish_queue(self):
         try:
             while msg := self.publish_queue.get(False):
                 body = msg[0]
@@ -84,10 +84,12 @@ 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
-        self._drain_publish_queue()
+        self.drain_publish_queue()
 
         # Update components
         added = self.components.keys() - self.active_components.keys()