diff options
author | Lukasz Rajewski <lukasz.rajewski@t-mobile.pl> | 2023-06-20 18:18:39 +0000 |
---|---|---|
committer | Lukasz Rajewski <lukasz.rajewski@t-mobile.pl> | 2023-06-21 14:52:08 +0000 |
commit | d553ee4c89f602263f7a973654244f61b07fbca5 (patch) | |
tree | 1b4ff17e9cb6cba19adbba9dc2550fe1d2129346 /src/onaptests/steps | |
parent | 66cb2f9168c8a8f0c19e662491019d01f16de146 (diff) |
Namespace status verification 2.0
- separate steps for resource type
- optional version check
- new file with details of failed resources
Issue-ID: INT-2246
Signed-off-by: Lukasz Rajewski <lukasz.rajewski@t-mobile.pl>
Change-Id: I538f877ab4d0824fd8dd5e65255c7ecf9545a26b
Diffstat (limited to 'src/onaptests/steps')
-rw-r--r-- | src/onaptests/steps/cloud/check_status.py | 1012 |
1 files changed, 525 insertions, 487 deletions
diff --git a/src/onaptests/steps/cloud/check_status.py b/src/onaptests/steps/cloud/check_status.py index 0bfce35..8b7ac46 100644 --- a/src/onaptests/steps/cloud/check_status.py +++ b/src/onaptests/steps/cloud/check_status.py @@ -2,333 +2,394 @@ # -*- coding: utf-8 -*- # SPDX-License-Identifier: Apache-2.0 import json -import os import logging +import os import re -from onapsdk.configuration import settings -from natural.date import delta -from xtesting.core import testcase +from pathlib import Path + +from jinja2 import Environment, PackageLoader, select_autoescape from kubernetes import client, config from kubernetes.stream import stream +from natural.date import delta from urllib3.exceptions import MaxRetryError, NewConnectionError -from jinja2 import Environment, PackageLoader, select_autoescape +from xtesting.core import testcase -from ..base import BaseStep -from .resources import Pod, Container, Service, Job -from .resources import Deployment, StatefulSet, DaemonSet, Pvc, ReplicaSet -from .resources import ConfigMap, Secret, Ingress +from onapsdk.configuration import settings from onaptests.utils.exceptions import StatusCheckException +from ..base import BaseStep +from .resources import (ConfigMap, Container, DaemonSet, Deployment, Ingress, + Job, Pod, Pvc, ReplicaSet, Secret, Service, + StatefulSet) + NAMESPACE = settings.K8S_ONAP_NAMESPACE -FULL_LOGS_CONTAINERS = [ - 'dcae-bootstrap', 'dcae-cloudify-manager', 'aai-resources', - 'aai-traversal', 'aai-modelloader', 'sdnc', 'so', 'so-bpmn-infra', - 'so-openstack-adapter', 'so-sdc-controller', 'mariadb-galera', 'sdc-be', - 'sdc-fe' -] - -# patterns to be excluded from the check -WAIVER_LIST = ['integration'] - -SPECIFIC_LOGS_CONTAINERS = { - 'sdc-be': ['/var/log/onap/sdc/sdc-be/error.log'], - 'sdc-onboarding-be': ['/var/log/onap/sdc/sdc-onboarding-be/error.log'], - 'aaf-cm': [ - '/opt/app/osaaf/logs/cm/cm-service.log', - '/opt/app/osaaf/logs/cm/cm-init.log' - ], - 'aaf-fs': [ - '/opt/app/osaaf/logs/fs/fs-service.log', - '/opt/app/osaaf/logs/fs/fs-init.log' - ], - 'aaf-locate': [ - '/opt/app/osaaf/logs/locate/locate-service.log', - '/opt/app/osaaf/logs/locate/locate-init.log' - ], - 'aaf-service': [ - '/opt/app/osaaf/logs/service/authz-service.log', - '/opt/app/osaaf/logs/service/authz-init.log' - ], - 'sdc-be': [ - '/var/log/onap/sdc/sdc-be/debug.log', - '/var/log/onap/sdc/sdc-be/error.log' - ], - 'sdc-fe': [ - '/var/log/onap/sdc/sdc-fe/debug.log', - '/var/log/onap/sdc/sdc-fe/error.log' - ], - 'vid': [ - '/var/log/onap/vid/audit.log', - '/var/log/onap/vid/application.log', - '/var/log/onap/vid/debug.log', - '/var/log/onap/vid/error.log' - ], -} - -DOCKER_REPOSITORIES = [ - 'nexus3.onap.org:10001', 'docker.elastic.co', 'docker.io', 'library', - 'registry.gitlab.com', 'registry.hub.docker.com', 'k8s.gcr.io', 'gcr.io' -] -DOCKER_REPOSITORIES_NICKNAMES = { - 'nexus3.onap.org:10001': 'onap', - 'docker.elastic.co': 'elastic', - 'docker.io': 'dockerHub (docker.io)', - 'registry.hub.docker.com': 'dockerHub (registry)', - 'registry.gitlab.com': 'gitlab', - 'library': 'dockerHub (library)', - 'default': 'dockerHub', - 'k8s.gcr.io': 'google (k8s.gcr)', - 'gcr.io': 'google (gcr)' -} - -GENERIC_NAMES = { - 'postgreSQL': ['crunchydata/crunchy-postgres', 'postgres'], - 'mariadb': ['adfinissygroup/k8s-mariadb-galera-centos', 'mariadb'], - 'elasticsearch': [ - 'bitnami/elasticsearch', 'elasticsearch/elasticsearch', - 'onap/clamp-dashboard-elasticsearch' - ], - 'nginx': ['bitnami/nginx', 'nginx'], - 'cassandra': [ - 'cassandra', 'onap/music/cassandra_3_11', 'onap/music/cassandra_music', - 'onap/aaf/aaf_cass' - ], - 'zookeeper': ['google_samples/k8szk', 'onap/dmaap/zookeeper', 'zookeeper'], - 'redis': [ - 'onap/vfc/db', - 'onap/org.onap.dcaegen2.deployments.redis-cluster-container' - ], - 'consul': ['consul', 'oomk8s/consul'], - 'rabbitmq': ['ansible/awx_rabbitmq', 'rabbitmq'] -} - -MAX_LOG_BYTES = 512000 - - -class CheckNamespaceStatusStep(BaseStep): - """Check status of all k8s resources in the selected namespace.""" + +class CheckK8sResourcesStep(BaseStep): __logger = logging.getLogger(__name__) - def __init__(self, cleanup: bool = False,**kwargs): - """Init CheckNamespaceStatusStep.""" - super().__init__(cleanup=cleanup) + def __init__(self, resource_type: str, **kwargs): + """Init CheckK8sResourcesStep.""" + super().__init__(cleanup=False) + self.core = client.CoreV1Api() + self.batch = client.BatchV1Api() + self.app = client.AppsV1Api() + self.networking = client.NetworkingV1Api() if settings.STATUS_RESULTS_DIRECTORY: self.res_dir = f"{settings.STATUS_RESULTS_DIRECTORY}" else: self.res_dir = f"{testcase.TestCase.dir_results}/kubernetes-status" - if settings.IN_CLUSTER: - config.load_incluster_config() - else: - config.load_kube_config(config_file=settings.K8S_CONFIG) - - self.core = client.CoreV1Api() - self.batch = client.BatchV1Api() - self.app = client.AppsV1Api() - self.networking = client.NetworkingV1Api() + self.failing = False + self.resource_type = resource_type + self.k8s_resources = [] + self.all_resources = [] + self.failing_resources = [] + self.jinja_env = Environment(autoescape=select_autoescape(['html']), + loader=PackageLoader('onaptests.templates','status')) - self.__logger.debug("namespace status init started") - self.pods = [] - self.services = [] - self.jobs = [] - self.deployments = [] - self.replicasets =[] - self.statefulsets = [] - self.daemonsets = [] - self.pvcs = [] - self.configmaps = [] - self.secrets = [] - self.ingresses = [] - self.details = {} + @property + def component(self) -> str: + """Component name.""" + return "ALL" @property def description(self) -> str: """Step description.""" - return "Check status of all k8s resources in the selected namespace." + return f"Check status of all k8s {self.resource_type}s in the {NAMESPACE} namespace." - @property - def component(self) -> str: - """Component name.""" - return "ALL" + def _init_resources(self): + self.__logger.debug(f"Loading all k8s {self.resource_type}s in the {NAMESPACE} namespace") + + def _parse_resources(self): + """Parse the resources.""" + return [] + + def execute(self): + super().execute() + os.makedirs(self.res_dir, exist_ok=True) + try: + self._init_resources() + if len(self.k8s_resources) > 0: + self.__logger.info("%4s %ss in the namespace", len(self.k8s_resources), self.resource_type) + self._parse_resources() + self.__logger.info("%4s %ss parsed, %s failing", len(self.all_resources), self.resource_type, + len(self.failing_resources)) + except (ConnectionRefusedError, MaxRetryError, NewConnectionError): + self.__logger.error("Test of k8s %ss failed.", self.resource_type) + self.__logger.error("Cannot connect to Kubernetes.") + +class CheckBasicK8sResourcesStep(CheckK8sResourcesStep): + + __logger = logging.getLogger(__name__) + + def __init__(self, resource_type: str, k8s_res_class, cleanup: bool = False,**kwargs): + """Init CheckBasicK8sResourcesStep.""" + super().__init__(resource_type) + self.k8s_res_class = k8s_res_class + + def _parse_resources(self): + """Parse simple k8s resources.""" + super()._parse_resources() + for k8s in self.k8s_resources: + resource = self.k8s_res_class(k8s=k8s) + self.all_resources.append(resource) @BaseStep.store_state def execute(self): - """Check status of all k8s resources in the selected namespace. + super().execute() - Use settings values: - - K8S_ONAP_NAMESPACE. +class CheckK8sConfigMapsStep(CheckBasicK8sResourcesStep): + + __logger = logging.getLogger(__name__) + + def __init__(self, cleanup: bool = False,**kwargs): + """Init CheckK8sConfigMapsStep.""" + super().__init__("configmap", ConfigMap) + + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.core.list_namespaced_config_map(NAMESPACE).items + +class CheckK8sSecretsStep(CheckBasicK8sResourcesStep): + + __logger = logging.getLogger(__name__) + + def __init__(self, cleanup: bool = False,**kwargs): + """Init CheckK8sSecretsStep.""" + super().__init__("secret", Secret) + + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.core.list_namespaced_secret(NAMESPACE).items + +class CheckK8sIngressesStep(CheckBasicK8sResourcesStep): + + __logger = logging.getLogger(__name__) + + def __init__(self, cleanup: bool = False,**kwargs): + """Init CheckK8sIngressesStep.""" + super().__init__("ingress", Ingress) + + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.networking.list_namespaced_ingress(NAMESPACE).items + +class CheckK8sPvcsStep(CheckK8sResourcesStep): + + __logger = logging.getLogger(__name__) + + def __init__(self, cleanup: bool = False,**kwargs): + """Init CheckK8sPvcsStep.""" + super().__init__("pvc") + + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.core.list_namespaced_persistent_volume_claim(NAMESPACE).items + def _parse_resources(self): + """Parse the jobs. + Return a list of Pods that were created to perform jobs. """ + super()._parse_resources() + jobs_pods = [] + for k8s in self.k8s_resources: + pvc = Pvc(k8s=k8s) + field_selector = f"involvedObject.name={pvc.name},involvedObject.kind=PersistentVolumeClaim" + pvc.events = self.core.list_namespaced_event( + NAMESPACE, + field_selector=field_selector).items + + if k8s.status.phase != "Bound": + self.__logger.warning("a PVC is in error: {}".format(pvc.name)) + self.failing_resources.append(pvc) + self.failing = True + self.all_resources.append(pvc) + + @BaseStep.store_state + def execute(self): super().execute() - os.makedirs(self.res_dir, exist_ok=True) - self.__logger.debug("start test") - try: - self.k8s_pods = self.core.list_namespaced_pod(NAMESPACE).items - self.__logger.info("%4s Pods in the namespace", len(self.k8s_pods)) - - self.k8s_jobs = self.batch.list_namespaced_job(NAMESPACE).items - self.__logger.info("%4s Jobs in the namespace", len(self.k8s_jobs)) - - self.k8s_deployments = self.app.list_namespaced_deployment( - NAMESPACE).items - self.__logger.info("%4s Deployments in the namespace", - len(self.k8s_deployments)) - - self.k8s_replicasets = self.app.list_namespaced_replica_set( - NAMESPACE).items - self.__logger.info("%4s Replicasets in the namespace", - len(self.k8s_replicasets)) - - self.k8s_statefulsets = self.app.list_namespaced_stateful_set( - NAMESPACE).items - self.__logger.info("%4s StatefulSets in the namespace", - len(self.k8s_statefulsets)) - - self.k8s_daemonsets = self.app.list_namespaced_daemon_set( - NAMESPACE).items - self.__logger.info("%4s DaemonSets in the namespace", - len(self.k8s_daemonsets)) - - self.k8s_services = self.core.list_namespaced_service( - NAMESPACE).items - self.__logger.info("%4s Services in the namespace", - len(self.k8s_services)) - - self.k8s_pvcs = self.core.list_namespaced_persistent_volume_claim( - NAMESPACE).items - self.__logger.info("%4s PVCs in the namespace", len(self.pvcs)) - - self.k8s_configmaps = self.core.list_namespaced_config_map( - NAMESPACE).items - self.__logger.info("%4s ConfigMaps in the namespace", - len(self.configmaps)) - - self.k8s_secrets = self.core.list_namespaced_secret( - NAMESPACE).items - self.__logger.info("%4s Secrets in the namespace", - len(self.secrets)) - - self.k8s_ingresses = self.networking.list_namespaced_ingress( - NAMESPACE).items - self.__logger.info("%4s Ingresses in the namespace", - len(self.ingresses)) - except (ConnectionRefusedError, MaxRetryError, NewConnectionError): - self.__logger.error("namespace status test failed.") - self.__logger.error("cannot connect to Kubernetes.") - return testcase.TestCase.EX_TESTCASE_FAILED - - self.failing_statefulsets = [] - self.failing_jobs = [] - self.failing_deployments = [] - self.failing_replicasets = [] - self.failing_daemonsets = [] - self.failing_pvcs = [] - self.failing = False - self.jinja_env = Environment(autoescape=select_autoescape(['html']), - loader=PackageLoader('onaptests.templates','status')) - self.parse_services() - jobs_pods = self.parse_jobs() - self.parse_pods(excluded_pods=jobs_pods) - self.parse_deployments() - self.parse_replicasets() - self.parse_statefulsets() - self.parse_daemonsets() - self.parse_pvcs() - self.parse_configmaps() - self.parse_secrets() - self.parse_ingresses() - self.parse_versions() - self.jinja_env.get_template('index.html.j2').stream( - ns=self, - delta=delta).dump('{}/index.html'.format(self.res_dir)) - self.jinja_env.get_template('raw_output.txt.j2').stream( - ns=self, namespace=NAMESPACE).dump('{}/onap-k8s.log'.format( - self.res_dir)) +class CheckK8sResourcesUsingPodsStep(CheckK8sResourcesStep): - if len(self.jobs) > 0: - self.details['jobs'] = { - 'number': len(self.jobs), - 'number_failing': len(self.failing_jobs), - 'failing': self.map_by_name(self.failing_jobs) - } - if len(self.deployments) > 0: - self.details['deployments'] = { - 'number': len(self.deployments), - 'number_failing': len(self.failing_deployments), - 'failing': self.map_by_name(self.failing_deployments) - } - if len(self.replicasets) > 0: - self.details['replicasets'] = { - 'number': len(self.replicasets), - 'number_failing': len(self.failing_replicasets), - 'failing': self.map_by_name(self.failing_replicasets) - } - if len(self.statefulsets) > 0: - self.details['statefulsets'] = { - 'number': len(self.statefulsets), - 'number_failing': len(self.failing_statefulsets), - 'failing': self.map_by_name(self.failing_statefulsets) - } - if len(self.daemonsets) > 0: - self.details['daemonsets'] = { - 'number': len(self.daemonsets), - 'number_failing': len(self.failing_daemonsets), - 'failing': self.map_by_name(self.failing_daemonsets) - } - if len(self.pvcs) > 0: - self.details['pvcs'] = { - 'number': len(self.pvcs), - 'number_failing': len(self.failing_pvcs), - 'failing': self.map_by_name(self.failing_pvcs) - } - if self.failing: - self.__logger.error("namespace status test failed.") - self.__logger.error("number of errored Jobs: %s", - len(self.failing_jobs)) - self.__logger.error("number of errored Deployments: %s", - len(self.failing_deployments)) - self.__logger.error("number of errored Replicasets: %s", - len(self.failing_replicasets)) - self.__logger.error("number of errored StatefulSets: %s", - len(self.failing_statefulsets)) - self.__logger.error("number of errored DaemonSets: %s", - len(self.failing_daemonsets)) - self.__logger.error("number of errored PVCs: %s", - len(self.failing_pvcs)) - raise StatusCheckException + __logger = logging.getLogger(__name__) + + def __init__(self, resource_type: str, pods_source, cleanup: bool = False,**kwargs): + """Init CheckK8sResourcesUsingPodsStep.""" + super().__init__(resource_type) + self.pods_source = pods_source + + def _get_used_pods(self): + pods = [] + if self.pods_source is not None: + pods = self.pods_source.all_resources + return pods + + def _find_child_pods(self, selector): + pods_used = self._get_used_pods() + pods_list = [] + failed_pods = 0 + if selector: + raw_selector = '' + for key, value in selector.items(): + raw_selector += key + '=' + value + ',' + raw_selector = raw_selector[:-1] + pods = self.core.list_namespaced_pod( + NAMESPACE, label_selector=raw_selector).items + for pod in pods: + for known_pod in pods_used: + if known_pod.name == pod.metadata.name: + pods_list.append(known_pod) + if not known_pod.ready(): + failed_pods += 1 + return (pods_list, failed_pods) + + @BaseStep.store_state + def execute(self): + super().execute() + +class CheckK8sJobsStep(CheckK8sResourcesUsingPodsStep): + + __logger = logging.getLogger(__name__) + + def __init__(self, cleanup: bool = False,**kwargs): + """Init CheckK8sJobsStep.""" + super().__init__("job", None) + + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.batch.list_namespaced_job(NAMESPACE).items + + def _parse_resources(self): + """Parse the jobs. + Return a list of Pods that were created to perform jobs. + """ + super()._parse_resources() + jobs_pods = [] + for k8s in self.k8s_resources: + job = Job(k8s=k8s) + job_pods = [] + + if k8s.spec.selector and k8s.spec.selector.match_labels: + (job.pods, job.failed_pods) = self._find_child_pods( + k8s.spec.selector.match_labels) + job_pods += job.pods + field_selector = "involvedObject.name={}".format(job.name) + field_selector += ",involvedObject.kind=Job" + job.events = self.core.list_namespaced_event( + NAMESPACE, + field_selector=field_selector).items + + self.jinja_env.get_template('job.html.j2').stream(job=job).dump( + '{}/job-{}.html'.format(self.res_dir, job.name)) + + # timemout job + if not k8s.status.completion_time: + self.__logger.warning("a Job is in error: {}".format(job.name)) + if any( + waiver_elt not in job.name for waiver_elt in settings.WAIVER_LIST): + self.failing_resources.append(job) + self.failing = True + # completed job + if any(waiver_elt not in job.name for waiver_elt in settings.WAIVER_LIST): + self.all_resources.append(job) + jobs_pods += job_pods + +class CheckK8sPodsStep(CheckK8sResourcesUsingPodsStep): + + __logger = logging.getLogger(__name__) + + def __init__(self, pods, cleanup: bool = False,**kwargs): + """Init CheckK8sPodsStep.""" + super().__init__("pod", pods) - def parse_pods(self, excluded_pods=None): - """Parse the pods status.""" - self.__logger.info("%4s pods to parse", len(self.k8s_pods)) - for k8s in self.k8s_pods: + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.core.list_namespaced_pod(NAMESPACE).items + + def _parse_resources(self): + """Parse the pods.""" + super()._parse_resources() + excluded_pods = self._get_used_pods() + pod_versions = [] + containers = {} + for k8s in self.k8s_resources: pod = Pod(k8s=k8s) + # check version firstly + if settings.CHECK_POD_VERSIONS: + pod_component = k8s.metadata.name + if 'app' in k8s.metadata.labels: + pod_component = k8s.metadata.labels['app'] + else: + if 'app.kubernetes.io/name' in k8s.metadata.labels: + pod_component = k8s.metadata.labels[ + 'app.kubernetes.io/name'] + else: + self.__logger.error("pod %s has no 'app' or 'app.kubernetes.io/name' in metadata: %s", pod_component, k8s.metadata.labels) + + ## looks for docker version + for container in k8s.spec.containers: + pod_version = {} + pod_container_version = container.image.rsplit(":", 1) + pod_container_image = pod_container_version[0] + pod_container_tag = "latest" + if len(pod_container_version) > 1: + pod_container_tag = pod_container_version[1] + + pod_version.update({ + 'container': container.name, + 'component': pod_component, + 'image': pod_container_image, + 'version': pod_container_tag + }) + pod_versions.append(pod_version) + + search_rule = "^(?P<source>[^/]*)/*(?P<container>[^:]*):*(?P<version>.*)$" + search = re.search(search_rule, container.image) + name = "{}/{}".format(search.group('source'), + search.group('container')) + version = search.group('version') + if name[-1] == '/': + name = name[0:-1] + source = "default" + if search.group('source') in settings.DOCKER_REPOSITORIES: + source = search.group('source') + name = search.group('container') + container_search_rule = "^library/(?P<real_container>[^:]*)$" + container_search = re.search(container_search_rule, name) + if container_search: + name = container_search.group('real_container') + for common_component in settings.GENERIC_NAMES.keys(): + if name in settings.GENERIC_NAMES[common_component]: + version = "{}:{}".format(name, version) + name = common_component + break + + repository = settings.DOCKER_REPOSITORIES_NICKNAMES[source] + if name in containers: + if version in containers[name]['versions']: + if not (pod_component in containers[name]['versions'] + [version]['components']): + containers[name]['versions'][version][ + 'components'].append(pod_component) + containers[name]['number_components'] += 1 + if not (repository in containers[name]['versions'] + [version]['repositories']): + containers[name]['versions'][version][ + 'repositories'].append(repository) + else: + containers[name]['versions'][version] = { + 'repositories': [repository], + 'components': [pod_component] + } + containers[name]['number_components'] += 1 + else: + containers[name] = { + 'versions': { + version: { + 'repositories': [repository], + 'components': [pod_component] + } + }, + 'number_components': 1 + } + # pod version check end if excluded_pods and pod in excluded_pods: continue if k8s.status.init_container_statuses: for k8s_container in k8s.status.init_container_statuses: - pod.runned_init_containers += self.parse_container( + pod.runned_init_containers += self._parse_container( pod, k8s_container, init=True) if k8s.status.container_statuses: for k8s_container in k8s.status.container_statuses: - pod.running_containers += self.parse_container( + pod.running_containers += self._parse_container( pod, k8s_container) pod.events = self.core.list_namespaced_event( NAMESPACE, field_selector="involvedObject.name={}".format(pod.name)).items self.jinja_env.get_template('pod.html.j2').stream(pod=pod).dump( '{}/pod-{}.html'.format(self.res_dir, pod.name)) - if any(waiver_elt in pod.name for waiver_elt in WAIVER_LIST): + if any(waiver_elt in pod.name for waiver_elt in settings.WAIVER_LIST): self.__logger.warn("Waiver pattern found in pod, exclude %s", pod.name) else: - self.pods.append(pod) - - def parse_container(self, pod, k8s_container, init=False): + self.all_resources.append(pod) + + if settings.CHECK_POD_VERSIONS: + self.jinja_env.get_template('version.html.j2').stream( + pod_versions=pod_versions).dump('{}/versions.html'.format( + self.res_dir)) + self.jinja_env.get_template('container_versions.html.j2').stream( + containers=containers).dump('{}/container_versions.html'.format( + self.res_dir)) + # create a json file for version tracking + with open(self.res_dir + "/onap_versions.json", "w") as write_file: + json.dump(pod_versions, write_file) + + def _parse_container(self, pod, k8s_container, init=False): """Get the logs of a container.""" logs = "" old_logs = "" @@ -358,7 +419,7 @@ class CheckNamespaceStatusStep(BaseStep): pod.name, NAMESPACE, container=container.name, - limit_bytes=MAX_LOG_BYTES, + limit_bytes=settings.MAX_LOG_BYTES, ) except UnicodeDecodeError: logs= "{0} has an unicode decode error...".format(pod.name) @@ -382,7 +443,7 @@ class CheckNamespaceStatusStep(BaseStep): container.name), 'w') as log_result: log_result.write(old_logs) - if (container.name in FULL_LOGS_CONTAINERS): + if (container.name in settings.FULL_LOGS_CONTAINERS): logs = self.core.read_namespaced_pod_log( pod.name, NAMESPACE, container=container.name) with open( @@ -390,8 +451,8 @@ class CheckNamespaceStatusStep(BaseStep): pod.name, container.name), 'w') as log_result: log_result.write(logs) - if (container.name in SPECIFIC_LOGS_CONTAINERS): - for log_file in SPECIFIC_LOGS_CONTAINERS[container.name]: + if (container.name in settings.SPECIFIC_LOGS_CONTAINERS): + for log_file in settings.SPECIFIC_LOGS_CONTAINERS[container.name]: exec_command = ['/bin/sh', '-c', "cat {}".format(log_file)] log_files[log_file] = stream( self.core.connect_get_namespaced_pod_exec, @@ -420,7 +481,7 @@ class CheckNamespaceStatusStep(BaseStep): old_logs=old_logs, log_files=log_files).dump('{}/pod-{}-{}-logs.html'.format( self.res_dir, pod.name, container.name)) - if any(waiver_elt in container.name for waiver_elt in WAIVER_LIST): + if any(waiver_elt in container.name for waiver_elt in settings.WAIVER_LIST): self.__logger.warn( "Waiver pattern found in container, exclude %s", container.name) else: @@ -429,10 +490,22 @@ class CheckNamespaceStatusStep(BaseStep): return 1 return 0 - def parse_services(self): +class CheckK8sServicesStep(CheckK8sResourcesUsingPodsStep): + + __logger = logging.getLogger(__name__) + + def __init__(self, pods, cleanup: bool = False,**kwargs): + """Init CheckK8sServicesStep.""" + super().__init__("service", pods) + + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.core.list_namespaced_service(NAMESPACE).items + + def _parse_resources(self): """Parse the services.""" - self.__logger.info("%4s services to parse", len(self.k8s_services)) - for k8s in self.k8s_services: + super()._parse_resources() + for k8s in self.k8s_resources: service = Service(k8s=k8s) (service.pods, @@ -441,53 +514,28 @@ class CheckNamespaceStatusStep(BaseStep): self.jinja_env.get_template('service.html.j2').stream( service=service).dump('{}/service-{}.html'.format( self.res_dir, service.name)) - self.services.append(service) + self.all_resources.append(service) - def parse_jobs(self): - """Parse the jobs. - Return a list of Pods that were created to perform jobs. - """ - self.__logger.info("%4s jobs to parse", len(self.k8s_jobs)) - jobs_pods = [] - for i in range(len(self.k8s_jobs)): - k8s = self.k8s_jobs[i] - job = Job(k8s=k8s) - job_pods = [] +class CheckK8sDeploymentsStep(CheckK8sResourcesUsingPodsStep): - if k8s.spec.selector and k8s.spec.selector.match_labels: - (job.pods, job.failed_pods) = self._find_child_pods( - k8s.spec.selector.match_labels) - job_pods += job.pods - field_selector = "involvedObject.name={}".format(job.name) - field_selector += ",involvedObject.kind=Job" - job.events = self.core.list_namespaced_event( - NAMESPACE, - field_selector=field_selector).items + __logger = logging.getLogger(__name__) - self.jinja_env.get_template('job.html.j2').stream(job=job).dump( - '{}/job-{}.html'.format(self.res_dir, job.name)) + def __init__(self, pods, cleanup: bool = False,**kwargs): + """Init CheckK8sDeploymentsStep.""" + super().__init__("deployment", pods) - # timemout job - if not k8s.status.completion_time: - self.__logger.warning("a Job is in error: {}".format(job.name)) - if any( - waiver_elt not in job.name for waiver_elt in WAIVER_LIST): - self.failing_jobs.append(job) - self.failing = True - # completed job - if any(waiver_elt not in job.name for waiver_elt in WAIVER_LIST): - self.jobs.append(job) - jobs_pods += job_pods - return jobs_pods + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.app.list_namespaced_deployment(NAMESPACE).items - def parse_deployments(self): + def _parse_resources(self): """Parse the deployments.""" - self.__logger.info("%4s deployments to parse", - len(self.k8s_deployments)) - for i in range(len(self.k8s_deployments)): - k8s = self.k8s_deployments[i] + super()._parse_resources() + for k8s in self.k8s_resources: deployment = Deployment(k8s=k8s) + if settings.IGNORE_EMPTY_REPLICAS and k8s.spec.replicas == 0: + continue if k8s.spec.selector and k8s.spec.selector.match_labels: (deployment.pods, deployment.failed_pods) = self._find_child_pods( @@ -504,19 +552,32 @@ class CheckNamespaceStatusStep(BaseStep): if k8s.status.unavailable_replicas: self.__logger.warning("a Deployment is in error: {}".format(deployment.name)) - self.failing_deployments.append(deployment) + self.failing_resources.append(deployment) self.failing = True - self.deployments.append(deployment) + self.all_resources.append(deployment) + +class CheckK8sResplicaSetsStep(CheckK8sResourcesUsingPodsStep): - def parse_replicasets(self): + __logger = logging.getLogger(__name__) + + def __init__(self, pods, cleanup: bool = False,**kwargs): + """Init CheckK8sResplicaSetsStep.""" + super().__init__("replicaset", pods) + + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.app.list_namespaced_replica_set(NAMESPACE).items + + def _parse_resources(self): """Parse the replicasets.""" - self.__logger.info("%4s replicasets to parse", - len(self.k8s_replicasets)) - for i in range(len(self.k8s_replicasets)): - k8s = self.k8s_replicasets[i] + super()._parse_resources() + for k8s in self.k8s_resources: replicaset = ReplicaSet(k8s=k8s) + if settings.IGNORE_EMPTY_REPLICAS and k8s.spec.replicas == 0: + continue + if k8s.spec.selector and k8s.spec.selector.match_labels: (replicaset.pods, replicaset.failed_pods) = self._find_child_pods( @@ -534,19 +595,32 @@ class CheckNamespaceStatusStep(BaseStep): if (not k8s.status.ready_replicas or (k8s.status.ready_replicas < k8s.status.replicas)): self.__logger.warning("a ReplicaSet is in error: {}".format(replicaset.name)) - self.failing_replicasets.append(replicaset) + self.failing_resources.append(replicaset) self.failing = True - self.replicasets.append(replicaset) + self.all_resources.append(replicaset) + +class CheckK8sStatefulSetsStep(CheckK8sResourcesUsingPodsStep): + + __logger = logging.getLogger(__name__) + + def __init__(self, pods, cleanup: bool = False,**kwargs): + """Init CheckK8sStatefulSetsStep.""" + super().__init__("statefulset", pods) - def parse_statefulsets(self): + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.app.list_namespaced_stateful_set(NAMESPACE).items + + def _parse_resources(self): """Parse the statefulsets.""" - self.__logger.info("%4s statefulsets to parse", - len(self.k8s_statefulsets)) - for i in range(len(self.k8s_statefulsets)): - k8s = self.k8s_statefulsets[i] + super()._parse_resources() + for k8s in self.k8s_resources: statefulset = StatefulSet(k8s=k8s) + if settings.IGNORE_EMPTY_REPLICAS and k8s.spec.replicas == 0: + continue + if k8s.spec.selector and k8s.spec.selector.match_labels: (statefulset.pods, statefulset.failed_pods) = self._find_child_pods( @@ -564,18 +638,32 @@ class CheckNamespaceStatusStep(BaseStep): if ((not k8s.status.ready_replicas) or (k8s.status.ready_replicas < k8s.status.replicas)): self.__logger.warning("a StatefulSet is in error: {}".format(statefulset.name)) - self.failing_statefulsets.append(statefulset) + self.failing_resources.append(statefulset) self.failing = True - self.statefulsets.append(statefulset) + self.all_resources.append(statefulset) + +class CheckK8sDaemonSetsStep(CheckK8sResourcesUsingPodsStep): - def parse_daemonsets(self): + __logger = logging.getLogger(__name__) + + def __init__(self, pods, cleanup: bool = False,**kwargs): + """Init CheckK8sDaemonSetsStep.""" + super().__init__("daemonset", pods) + + def _init_resources(self): + super()._init_resources() + self.k8s_resources = self.app.list_namespaced_daemon_set(NAMESPACE).items + + def _parse_resources(self): """Parse the daemonsets.""" - self.__logger.info("%4s daemonsets to parse", len(self.k8s_daemonsets)) - for i in range(len(self.k8s_daemonsets)): - k8s = self.k8s_daemonsets[i] + super()._parse_resources() + for k8s in self.k8s_resources: daemonset = DaemonSet(k8s=k8s) + if settings.IGNORE_EMPTY_REPLICAS and k8s.spec.replicas == 0: + continue + if k8s.spec.selector and k8s.spec.selector.match_labels: (daemonset.pods, daemonset.failed_pods) = self._find_child_pods( @@ -592,160 +680,110 @@ class CheckNamespaceStatusStep(BaseStep): if (k8s.status.number_ready < k8s.status.desired_number_scheduled): self.__logger.warning("a DaemonSet is in error: {}".format(daemonset.name)) - self.failing_daemonsets.append(daemonset) + self.failing_resources.append(daemonset) self.failing = True - self.daemonsets.append(daemonset) + self.all_resources.append(daemonset) - def parse_pvcs(self): - """Parse the persistent volume claims.""" - self.__logger.info("%4s pvcs to parse", len(self.k8s_pvcs)) - for k8s in self.k8s_pvcs: - pvc = Pvc(k8s=k8s) - field_selector = f"involvedObject.name={pvc.name},involvedObject.kind=PersistentVolumeClaim" - pvc.events = self.core.list_namespaced_event( - NAMESPACE, - field_selector=field_selector).items +class CheckNamespaceStatusStep(CheckK8sResourcesStep): + """Check status of all k8s resources in the selected namespace.""" - if k8s.status.phase != "Bound": - self.__logger.warning("a PVC is in error: {}".format(pvc.name)) - self.failing_pvcs.append(pvc) - self.failing = True + __logger = logging.getLogger(__name__) - self.pvcs.append(pvc) - - def parse_configmaps(self): - """Parse the config maps.""" - self.__logger.info("%4s config maps to parse", - len(self.k8s_configmaps)) - for k8s in self.k8s_configmaps: - configmap = ConfigMap(k8s=k8s) - self.configmaps.append(configmap) - - def parse_secrets(self): - """Parse the secrets.""" - self.__logger.info("%4s secrets to parse", len(self.k8s_secrets)) - for k8s in self.k8s_secrets: - secret = Secret(k8s=k8s) - self.secrets.append(secret) - - def parse_ingresses(self): - """Parse the ingresses.""" - self.__logger.info("%4s ingresses to parse", len(self.k8s_ingresses)) - for k8s in self.k8s_secrets: - ingress = Ingress(k8s=k8s) - self.ingresses.append(ingress) - - def parse_versions(self): - """Parse the versions of the pods.""" - self.__logger.info("%4s pods to parse", len(self.k8s_pods)) - pod_versions = [] - containers = {} - for pod in self.k8s_pods: - pod_component = pod.metadata.name - if 'app' in pod.metadata.labels: - pod_component = pod.metadata.labels['app'] - else: - if 'app.kubernetes.io/name' in pod.metadata.labels: - pod_component = pod.metadata.labels[ - 'app.kubernetes.io/name'] - else: - self.__logger.error("pod %s has no 'app' or 'app.kubernetes.io/name' in metadata: %s", pod_component, pod.metadata.labels) - - # looks for docker version - for container in pod.spec.containers: - pod_version = {} - pod_container_version = container.image.rsplit(":", 1) - pod_container_image = pod_container_version[0] - pod_container_tag = "latest" - if len(pod_container_version) > 1: - pod_container_tag = pod_container_version[1] - - pod_version.update({ - 'container': container.name, - 'component': pod_component, - 'image': pod_container_image, - 'version': pod_container_tag - }) - pod_versions.append(pod_version) - - search_rule = "^(?P<source>[^/]*)/*(?P<container>[^:]*):*(?P<version>.*)$" - search = re.search(search_rule, container.image) - name = "{}/{}".format(search.group('source'), - search.group('container')) - version = search.group('version') - if name[-1] == '/': - name = name[0:-1] - source = "default" - if search.group('source') in DOCKER_REPOSITORIES: - source = search.group('source') - name = search.group('container') - container_search_rule = "^library/(?P<real_container>[^:]*)$" - container_search = re.search(container_search_rule, name) - if container_search: - name = container_search.group('real_container') - for common_component in GENERIC_NAMES.keys(): - if name in GENERIC_NAMES[common_component]: - version = "{}:{}".format(name, version) - name = common_component - break - - repository = DOCKER_REPOSITORIES_NICKNAMES[source] - if name in containers: - if version in containers[name]['versions']: - if not (pod_component in containers[name]['versions'] - [version]['components']): - containers[name]['versions'][version][ - 'components'].append(pod_component) - containers[name]['number_components'] += 1 - if not (repository in containers[name]['versions'] - [version]['repositories']): - containers[name]['versions'][version][ - 'repositories'].append(repository) - else: - containers[name]['versions'][version] = { - 'repositories': [repository], - 'components': [pod_component] - } - containers[name]['number_components'] += 1 - else: - containers[name] = { - 'versions': { - version: { - 'repositories': [repository], - 'components': [pod_component] - } - }, - 'number_components': 1 - } + def __init__(self, cleanup: bool = False, **kwargs): + """Init CheckNamespaceStatusStep.""" + super().__init__("") + self.__logger.debug("%s namespace status test init started", NAMESPACE) + if settings.IN_CLUSTER: + config.load_incluster_config() + else: + config.load_kube_config(config_file=settings.K8S_CONFIG) - self.jinja_env.get_template('version.html.j2').stream( - pod_versions=pod_versions).dump('{}/versions.html'.format( - self.res_dir)) - self.jinja_env.get_template('container_versions.html.j2').stream( - containers=containers).dump('{}/container_versions.html'.format( + self.job_list_step = CheckK8sJobsStep() + self.pod_list_step = CheckK8sPodsStep(self.job_list_step) + self.service_list_step = CheckK8sServicesStep(self.pod_list_step) + self.deployment_list_step = CheckK8sDeploymentsStep(self.pod_list_step) + self.replicaset_list_step = CheckK8sResplicaSetsStep(self.pod_list_step) + self.statefulset_list_step = CheckK8sStatefulSetsStep(self.pod_list_step) + self.daemonset_list_step = CheckK8sDaemonSetsStep(self.pod_list_step) + self.configmap_list_step = CheckK8sConfigMapsStep() + self.secret_list_step = CheckK8sSecretsStep() + self.ingress_list_step = CheckK8sIngressesStep() + self.pvc_list_step = CheckK8sPvcsStep() + self.add_step(self.job_list_step) + self.add_step(self.pod_list_step) + self.add_step(self.service_list_step) + self.add_step(self.deployment_list_step) + self.add_step(self.replicaset_list_step) + self.add_step(self.statefulset_list_step) + self.add_step(self.daemonset_list_step) + self.add_step(self.configmap_list_step) + self.add_step(self.secret_list_step) + self.add_step(self.ingress_list_step) + self.add_step(self.pvc_list_step) + + @property + def description(self) -> str: + """Step description.""" + return "Check status of all k8s resources in the selected namespace." + + @property + def component(self) -> str: + """Component name.""" + return "ALL" + + @BaseStep.store_state + def execute(self): + """Check status of all k8s resources in the selected namespace. + + Use settings values: + - K8S_ONAP_NAMESPACE + - STATUS_RESULTS_DIRECTORY + - STORE_ARTIFACTS + - CHECK_POD_VERSIONS + """ + super().execute() + + self.pods = self.pod_list_step.all_resources + self.services = self.service_list_step.all_resources + self.jobs = self.job_list_step.all_resources + self.deployments = self.deployment_list_step.all_resources + self.replicasets = self.replicaset_list_step.all_resources + self.statefulsets = self.statefulset_list_step.all_resources + self.daemonsets = self.daemonset_list_step.all_resources + self.pvcs = self.pvc_list_step.all_resources + self.configmaps = self.configmap_list_step.all_resources + self.secrets = self.secret_list_step.all_resources + self.ingresses = self.ingress_list_step.all_resources + + self.failing_statefulsets = self.statefulset_list_step.failing_resources + self.failing_jobs = self.job_list_step.failing_resources + self.failing_deployments = self.deployment_list_step.failing_resources + self.failing_replicasets = self.replicaset_list_step.failing_resources + self.failing_daemonsets = self.daemonset_list_step.failing_resources + self.failing_pvcs = self.pvc_list_step.failing_resources + + self.jinja_env.get_template('index.html.j2').stream( + ns=self, + delta=delta).dump('{}/index.html'.format(self.res_dir)) + self.jinja_env.get_template('raw_output.txt.j2').stream( + ns=self, namespace=NAMESPACE).dump('{}/onap-k8s.log'.format( self.res_dir)) - # create a json file for version tracking - with open(self.res_dir + "/onap_versions.json", "w") as write_file: - json.dump(pod_versions, write_file) - def _find_child_pods(self, selector): - pods_list = [] - failed_pods = 0 - if selector: - raw_selector = '' - for key, value in selector.items(): - raw_selector += key + '=' + value + ',' - raw_selector = raw_selector[:-1] - pods = self.core.list_namespaced_pod( - NAMESPACE, label_selector=raw_selector).items - for pod in pods: - for known_pod in self.pods: - if known_pod.name == pod.metadata.name: - pods_list.append(known_pod) - if not known_pod.ready(): - failed_pods += 1 - return (pods_list, failed_pods) + details = {} + for step in self._steps: + if step.failing: + self.failing = True + self.__logger.info("%s failing: %s", step.resource_type, len(step.failing_resources)) + details[step.resource_type] = { + 'number_all': len(step.all_resources), + 'number_failing': len(step.failing_resources), + 'failing': self.map_by_name(step.failing_resources) + } + with (Path(self.res_dir).joinpath(settings.STATUS_DETAILS_JSON)).open('w') as file: + json.dump(details, file, indent=4) + if self.failing: + raise StatusCheckException def map_by_name(self, resources): return list(map(lambda resource: resource.name, resources)) |