Merge pull request #110 from nolar/pykube

Switch to pykube-ng to handle arbitrary resources (pods, jobs, etc)

.travis.yml

@@ -16,9 +16,10 @@ env:
     - MINIKUBE_HOME=$HOME
     - MINIKUBE_VERSION=1.0.1
   matrix:
-    - KUBERNETES_VERSION=1.14.0
-    - KUBERNETES_VERSION=1.13.0
-    - KUBERNETES_VERSION=1.12.0
+    - KUBERNETES_VERSION=1.14.0 CLIENT=yes  # only one "yes" is enough
+    - KUBERNETES_VERSION=1.14.0 CLIENT=no
+    - KUBERNETES_VERSION=1.13.0 CLIENT=no
+    - KUBERNETES_VERSION=1.12.0 CLIENT=no
 #    - KUBERNETES_VERSION=1.11.10  # Minikube fails on CRI preflight checks
 #    - KUBERNETES_VERSION=1.10.13  # CRDs require spec.version, which fails on 1.14
 
@@ -32,6 +33,7 @@ matrix:
 
 before_script:
   - tools/minikube-for-travis.sh
+  - tools/kubernetes-client.sh
 
 script:
   - pytest -v

examples/09-testing/test_example_09.py

@@ -13,18 +13,25 @@
 
 @pytest.fixture(autouse=True)
 def crd_exists():
-    subprocess.run(f"kubectl apply -f {crd_yaml}", shell=True, check=True)
+    subprocess.run(f"kubectl apply -f {crd_yaml}",
+                   check=True, timeout=10, capture_output=True, shell=True)
 
 
 @pytest.fixture(autouse=True)
 def obj_absent():
-    subprocess.run(f"kubectl delete -f {obj_yaml}", shell=True, check=False)
+    # Operator is not running in fixtures, so we need a force-delete (or this patch).
+    subprocess.run(['kubectl', 'patch', '-f', obj_yaml,
+                    '-p', '{"metadata":{"finalizers":[]}}',
+                    '--type', 'merge'],
+                   check=False, timeout=10, capture_output=True)
+    subprocess.run(f"kubectl delete -f {obj_yaml}",
+                   check=False, timeout=10, capture_output=True, shell=True)
 
 
 def test_resource_lifecycle(mocker):
 
     # To prevent lengthy threads in the loop executor when the process exits.
-    mocker.patch('kopf.clients.watching.DEFAULT_STREAM_TIMEOUT', 10)
+    mocker.patch('kopf.config.WatchersConfig.default_stream_timeout', 10)
 
     # Run an operator and simulate some activity with the operated resource.
     with kopf.testing.KopfRunner(['run', '--verbose', '--standalone', example_py]) as runner:

examples/10-builtins/README.md

@@ -0,0 +1,39 @@
+# Kopf example for built-in resources
+
+Kopf can also handle the built-in resources, such as Pods, Jobs, etc.
+
+In this example, we take control all over the pods (namespaced/cluster-wide),
+and allow the pods to exist for no longer than 30 seconds --
+either after creation or after the operator restart.
+
+For no specific reason, just for fun. Maybe, as a way of Chaos Engineering
+to force making the resilient applications (tolerant to pod killing).
+
+However, the system namespaces (kube-system, etc) are explicitly protected --
+to prevent killing the cluster itself.
+
+Start the operator:
+
+```bash
+kopf run example.py --verbose
+```
+
+Start a sample pod:
+
+```bash
+kubectl run -it --image=ubuntu expr1 -- bash -i
+# wait for 30s
+```
+
+Since `kubectl run` creates a Deployment, not just a Pod,
+a new pod will be created every 30 seconds. Observe with:
+
+```bash
+kubectl get pods --watch
+```
+
+Cleanup in the end:
+
+```bash
+$ kubectl delete deployment expr1
+```

examples/10-builtins/example.py

@@ -0,0 +1,47 @@
+import asyncio
+
+import kopf
+import pykube
+
+tasks = {}  # dict{namespace: dict{name: asyncio.Task}}
+
+try:
+    cfg = pykube.KubeConfig.from_service_account()
+except FileNotFoundError:
+    cfg = pykube.KubeConfig.from_file()
+api = pykube.HTTPClient(cfg)
+
+
+@kopf.on.resume('', 'v1', 'pods')
+@kopf.on.create('', 'v1', 'pods')
+async def pod_in_sight(namespace, name, logger, **kwargs):
+    if namespace.startswith('kube-'):
+        return
+    else:
+        task = asyncio.create_task(pod_killer(namespace, name, logger))
+        tasks.setdefault(namespace, {})
+        tasks[namespace][name] = task
+
+
+@kopf.on.delete('', 'v1', 'pods')
+async def pod_deleted(namespace, name, **kwargs):
+    if namespace in tasks and name in tasks[namespace]:
+        task = tasks[namespace][name]
+        task.cancel()  # it will also remove from `tasks`
+
+
+async def pod_killer(namespace, name, logger, timeout=30):
+    try:
+        logger.info(f"=== Pod killing happens in {timeout}s.")
+        await asyncio.sleep(timeout)
+        logger.info(f"=== Pod killing happens NOW!")
+
+        pod = pykube.Pod.objects(api, namespace=namespace).get_by_name(name)
+        pod.delete()
+
+    except asyncio.CancelledError:
+        logger.info(f"=== Pod killing is cancelled!")
+
+    finally:
+        if namespace in tasks and name in tasks[namespace]:
+            del tasks[namespace][name]

examples/10-builtins/test_example_10.py

@@ -0,0 +1,46 @@
+import os.path
+import time
+
+import pykube
+
+import kopf.testing
+
+
+def test_pods_reacted():
+
+    example_py = os.path.join(os.path.dirname(__file__), 'example.py')
+    with kopf.testing.KopfRunner(['run', '--verbose', example_py]) as runner:
+        _create_pod()
+        time.sleep(5)  # give it some time to react
+        _delete_pod()
+        time.sleep(1)  # give it some time to react
+
+    assert runner.exception is None
+    assert runner.exit_code == 0
+
+    assert '[default/kopf-pod-1] Creation event:' in runner.stdout
+    assert '[default/kopf-pod-1] === Pod killing happens in 30s.' in runner.stdout
+    assert '[default/kopf-pod-1] Deletion event:' in runner.stdout
+    assert '[default/kopf-pod-1] === Pod killing is cancelled!' in runner.stdout
+
+
+def _create_pod():
+    api = pykube.HTTPClient(pykube.KubeConfig.from_file())
+    pod = pykube.Pod(api, {
+        'apiVersion': 'v1',
+        'kind': 'Pod',
+        'metadata': {'name': 'kopf-pod-1', 'namespace': 'default'},
+        'spec': {
+            'containers': [{
+                'name': 'the-only-one',
+                'image': 'busybox',
+                'command': ["sh", "-x", "-c", "sleep 1"],
+            }]},
+    })
+    pod.create()
+
+
+def _delete_pod():
+    api = pykube.HTTPClient(pykube.KubeConfig.from_file())
+    pod = pykube.Pod.objects(api, namespace='default').get_by_name('kopf-pod-1')
+    pod.delete()

examples/requirements.txt

@@ -1,3 +1,4 @@
 kopf
 kubernetes
+pykube-ng
 pyyaml

kopf/cli.py

@@ -12,9 +12,11 @@
 
 def cli_login():
     try:
-        auth.login()
+        auth.login(verify=True)
     except auth.LoginError as e:
         raise click.ClickException(str(e))
+    except auth.AccessError as e:
+        raise click.ClickException(str(e))
 
 
 def logging_options(fn):

kopf/clients/auth.py

@@ -1,42 +1,138 @@
 import logging
+from typing import Optional
 
-import kubernetes
+import pykube
+import requests
 import urllib3.exceptions
 
 logger = logging.getLogger(__name__)
 
+# Set in login(), consumed in get_pykube_cfg() and all API calls.
+_pykube_cfg: Optional[pykube.KubeConfig] = None
+
 
 class LoginError(Exception):
     """ Raised when the operator cannot login to the API. """
 
 
-def login():
+class AccessError(Exception):
+    """ Raised when the operator cannot access the cluster API. """
+
+
+def login(verify=False):
     """
-    Login the the Kubernetes cluster, locally or remotely.
+    Login to Kubernetes cluster, locally or remotely.
+
+    Keep the logged in state or config object in the global variables,
+    so that it can be available for future calls via the same function call.
+
+    Automatic refresh/reload of the tokens or objects also should be done here.
     """
 
-    # Configure the default client credentials for all possible environments.
+    # Pykube login is mandatory. If it fails, the framework will not run at all.
+    try:
+        import pykube
+    except ImportError:
+        raise  # mandatory
+    else:
+        login_pykube(verify=verify)
+
+    # We keep the official client library auto-login only because it was
+    # an implied behavior before switching to pykube -- to keep it so (implied).
+    try:
+        import kubernetes
+    except ImportError:
+        pass  # optional
+    else:
+        login_client(verify=verify)
+
+
+def login_pykube(verify=False):
+    global _pykube_cfg
+    try:
+        _pykube_cfg = pykube.KubeConfig.from_service_account()
+        logger.debug("Pykube is configured in cluster with service account.")
+    except FileNotFoundError:
+        try:
+            _pykube_cfg = pykube.KubeConfig.from_file()
+            logger.debug("Pykube is configured via kubeconfig file.")
+        except (pykube.PyKubeError, FileNotFoundError):
+            raise LoginError(f"Cannot authenticate pykube neither in-cluster, nor via kubeconfig.")
+
+    if verify:
+        verify_pykube()
+
+
+def login_client(verify=False):
+    import kubernetes.client
     try:
         kubernetes.config.load_incluster_config()  # cluster env vars
-        logger.debug("configured in cluster with service account")
+        logger.debug("Client is configured in cluster with service account.")
     except kubernetes.config.ConfigException as e1:
         try:
             kubernetes.config.load_kube_config()  # developer's config files
-            logger.debug("configured via kubeconfig file")
+            logger.debug("Client is configured via kubeconfig file.")
         except kubernetes.config.ConfigException as e2:
-            raise LoginError(f"Cannot authenticate neither in-cluster, nor via kubeconfig.")
+            raise LoginError(f"Cannot authenticate client neither in-cluster, nor via kubeconfig.")
 
-    # Make a sample API call to ensure the login is successful,
-    # and convert some of the known exceptions to the CLI hints.
+    if verify:
+        verify_client()
+
+
+def verify_pykube():
+    """
+    Verify if login has succeeded, and the access configuration is still valid.
+
+    All other errors (e.g. 403, 404) are ignored: it means, the host and port
+    are configured and are reachable, the authentication token is accepted,
+    and the rest are authorization or configuration errors (not a showstopper).
+    """
+    try:
+        api = get_pykube_api()
+        rsp = api.get(version="", base="/")
+        rsp.raise_for_status()
+        api.raise_for_status(rsp)  # replaces requests's HTTPError with its own.
+    except requests.exceptions.ConnectionError as e:
+        raise AccessError("Cannot connect to the Kubernetes API. "
+                          "Please configure the cluster access.")
+    except pykube.exceptions.HTTPError as e:
+        if e.code == 401:
+            raise AccessError("Cannot authenticate to the Kubernetes API. "
+                              "Please login or configure the tokens.")
+    except requests.exceptions.HTTPError as e:
+        if e.response.status_code == 401:
+            raise AccessError("Cannot authenticate to the Kubernetes API. "
+                              "Please login or configure the tokens.")
+
+
+def verify_client():
+    """
+    Verify if login has succeeded, and the access configuration is still valid.
+
+    All other errors (e.g. 403, 404) are ignored: it means, the host and port
+    are configured and are reachable, the authentication token is accepted,
+    and the rest are authorization or configuration errors (not a showstopper).
+    """
+    import kubernetes.client.rest
     try:
         api = kubernetes.client.CoreApi()
         api.get_api_versions()
     except urllib3.exceptions.HTTPError as e:
-        raise LoginError("Cannot connect to the Kubernetes API. "
-                         "Please configure the cluster access.")
+        raise AccessError("Cannot connect to the Kubernetes API. "
+                          "Please configure the cluster access.")
     except kubernetes.client.rest.ApiException as e:
         if e.status == 401:
-            raise LoginError("Cannot authenticate to the Kubernetes API. "
-                             "Please login or configure the tokens.")
-        else:
-            raise
+            raise AccessError("Cannot authenticate to the Kubernetes API. "
+                              "Please login or configure the tokens.")
+
+
+def get_pykube_cfg() -> pykube.KubeConfig:
+    if _pykube_cfg is None:
+        raise LoginError("Not logged in with PyKube.")
+    return _pykube_cfg
+
+
+# TODO: add some caching, but keep kwargs in mind. Maybe add a key= for purpose/use-place?
+def get_pykube_api(timeout=None) -> pykube.HTTPClient:
+    kwargs = dict(timeout=timeout) if timeout is not None else dict()
+    return pykube.HTTPClient(get_pykube_cfg(), **kwargs)

kopf/clients/classes.py

@@ -0,0 +1,23 @@
+from typing import Type
+
+import pykube
+
+from kopf.clients import auth
+
+
+def _make_cls(resource) -> Type[pykube.objects.APIObject]:
+    api = auth.get_pykube_api()
+    api_resources = api.resource_list(resource.api_version)['resources']
+    resource_kind = next((r['kind'] for r in api_resources if r['name'] == resource.plural), None)
+    is_namespaced = next((r['namespaced'] for r in api_resources if r['name'] == resource.plural), None)
+    if not resource_kind:
+        raise pykube.ObjectDoesNotExist(f"No such CRD: {resource.name}")
+
+    cls_name = resource.plural
+    cls_base = pykube.objects.NamespacedAPIObject if is_namespaced else pykube.objects.APIObject
+    cls = type(cls_name, (cls_base,), {
+        'version': resource.api_version,
+        'endpoint': resource.plural,
+        'kind': resource_kind,
+    })
+    return cls