diff options
Diffstat (limited to 'roles/openshift_health_checker/openshift_checks/logging')
7 files changed, 879 insertions, 0 deletions
diff --git a/roles/openshift_health_checker/openshift_checks/logging/__init__.py b/roles/openshift_health_checker/openshift_checks/logging/__init__.py new file mode 100644 index 000000000..e69de29bb --- /dev/null +++ b/roles/openshift_health_checker/openshift_checks/logging/__init__.py diff --git a/roles/openshift_health_checker/openshift_checks/logging/curator.py b/roles/openshift_health_checker/openshift_checks/logging/curator.py new file mode 100644 index 000000000..f82ae64d7 --- /dev/null +++ b/roles/openshift_health_checker/openshift_checks/logging/curator.py @@ -0,0 +1,54 @@ +"""Check for an aggregated logging Curator deployment""" + +from openshift_checks.logging.logging import LoggingCheck + + +class Curator(LoggingCheck): + """Check for an aggregated logging Curator deployment""" + + name = "curator" + tags = ["health", "logging"] + + logging_namespace = None + + def run(self): + self.logging_namespace = self.get_var("openshift_logging_namespace", default="logging") + curator_pods, error = super(Curator, self).get_pods_for_component( + self.logging_namespace, + "curator", + ) + if error: + return {"failed": True, "changed": False, "msg": error} + check_error = self.check_curator(curator_pods) + + if check_error: + msg = ("The following Curator deployment issue was found:" + "\n-------\n" + "{}".format(check_error)) + return {"failed": True, "changed": False, "msg": msg} + + # TODO(lmeyer): run it all again for the ops cluster + return {"failed": False, "changed": False, "msg": 'No problems found with Curator deployment.'} + + def check_curator(self, pods): + """Check to see if curator is up and working. Returns: error string""" + if not pods: + return ( + "There are no Curator pods for the logging stack,\n" + "so nothing will prune Elasticsearch indexes.\n" + "Is Curator correctly deployed?" + ) + + not_running = super(Curator, self).not_running_pods(pods) + if len(not_running) == len(pods): + return ( + "The Curator pod is not currently in a running state,\n" + "so Elasticsearch indexes may increase without bound." + ) + if len(pods) - len(not_running) > 1: + return ( + "There is more than one Curator pod running. This should not normally happen.\n" + "Although this doesn't cause any problems, you may want to investigate." + ) + + return None diff --git a/roles/openshift_health_checker/openshift_checks/logging/elasticsearch.py b/roles/openshift_health_checker/openshift_checks/logging/elasticsearch.py new file mode 100644 index 000000000..1e478c04d --- /dev/null +++ b/roles/openshift_health_checker/openshift_checks/logging/elasticsearch.py @@ -0,0 +1,210 @@ +"""Check for an aggregated logging Elasticsearch deployment""" + +import json +import re + +from openshift_checks.logging.logging import LoggingCheck + + +class Elasticsearch(LoggingCheck): + """Check for an aggregated logging Elasticsearch deployment""" + + name = "elasticsearch" + tags = ["health", "logging"] + + logging_namespace = None + + def run(self): + """Check various things and gather errors. Returns: result as hash""" + + self.logging_namespace = self.get_var("openshift_logging_namespace", default="logging") + es_pods, error = super(Elasticsearch, self).get_pods_for_component( + self.logging_namespace, + "es", + ) + if error: + return {"failed": True, "changed": False, "msg": error} + check_error = self.check_elasticsearch(es_pods) + + if check_error: + msg = ("The following Elasticsearch deployment issue was found:" + "\n-------\n" + "{}".format(check_error)) + return {"failed": True, "changed": False, "msg": msg} + + # TODO(lmeyer): run it all again for the ops cluster + return {"failed": False, "changed": False, "msg": 'No problems found with Elasticsearch deployment.'} + + def _not_running_elasticsearch_pods(self, es_pods): + """Returns: list of pods that are not running, list of errors about non-running pods""" + not_running = super(Elasticsearch, self).not_running_pods(es_pods) + if not_running: + return not_running, [( + 'The following Elasticsearch pods are not running:\n' + '{pods}' + 'These pods will not aggregate logs from their nodes.' + ).format(pods=''.join( + " {} ({})\n".format(pod['metadata']['name'], pod['spec'].get('host', 'None')) + for pod in not_running + ))] + return not_running, [] + + def check_elasticsearch(self, es_pods): + """Various checks for elasticsearch. Returns: error string""" + not_running_pods, error_msgs = self._not_running_elasticsearch_pods(es_pods) + running_pods = [pod for pod in es_pods if pod not in not_running_pods] + pods_by_name = { + pod['metadata']['name']: pod for pod in running_pods + # Filter out pods that are not members of a DC + if pod['metadata'].get('labels', {}).get('deploymentconfig') + } + if not pods_by_name: + return 'No logging Elasticsearch pods were found. Is logging deployed?' + error_msgs += self._check_elasticsearch_masters(pods_by_name) + error_msgs += self._check_elasticsearch_node_list(pods_by_name) + error_msgs += self._check_es_cluster_health(pods_by_name) + error_msgs += self._check_elasticsearch_diskspace(pods_by_name) + return '\n'.join(error_msgs) + + @staticmethod + def _build_es_curl_cmd(pod_name, url): + base = "exec {name} -- curl -s --cert {base}cert --key {base}key --cacert {base}ca -XGET '{url}'" + return base.format(base="/etc/elasticsearch/secret/admin-", name=pod_name, url=url) + + def _check_elasticsearch_masters(self, pods_by_name): + """Check that Elasticsearch masters are sane. Returns: list of error strings""" + es_master_names = set() + error_msgs = [] + for pod_name in pods_by_name.keys(): + # Compare what each ES node reports as master and compare for split brain + get_master_cmd = self._build_es_curl_cmd(pod_name, "https://localhost:9200/_cat/master") + master_name_str = self._exec_oc(get_master_cmd, []) + master_names = (master_name_str or '').split(' ') + if len(master_names) > 1: + es_master_names.add(master_names[1]) + else: + error_msgs.append( + 'No master? Elasticsearch {pod} returned bad string when asked master name:\n' + ' {response}'.format(pod=pod_name, response=master_name_str) + ) + + if not es_master_names: + error_msgs.append('No logging Elasticsearch masters were found. Is logging deployed?') + return '\n'.join(error_msgs) + + if len(es_master_names) > 1: + error_msgs.append( + 'Found multiple Elasticsearch masters according to the pods:\n' + '{master_list}\n' + 'This implies that the masters have "split brain" and are not correctly\n' + 'replicating data for the logging cluster. Log loss is likely to occur.' + .format(master_list='\n'.join(' ' + master for master in es_master_names)) + ) + + return error_msgs + + def _check_elasticsearch_node_list(self, pods_by_name): + """Check that reported ES masters are accounted for by pods. Returns: list of error strings""" + + if not pods_by_name: + return ['No logging Elasticsearch masters were found. Is logging deployed?'] + + # get ES cluster nodes + node_cmd = self._build_es_curl_cmd(list(pods_by_name.keys())[0], 'https://localhost:9200/_nodes') + cluster_node_data = self._exec_oc(node_cmd, []) + try: + cluster_nodes = json.loads(cluster_node_data)['nodes'] + except (ValueError, KeyError): + return [ + 'Failed to query Elasticsearch for the list of ES nodes. The output was:\n' + + cluster_node_data + ] + + # Try to match all ES-reported node hosts to known pods. + error_msgs = [] + for node in cluster_nodes.values(): + # Note that with 1.4/3.4 the pod IP may be used as the master name + if not any(node['host'] in (pod_name, pod['status'].get('podIP')) + for pod_name, pod in pods_by_name.items()): + error_msgs.append( + 'The Elasticsearch cluster reports a member node "{node}"\n' + 'that does not correspond to any known ES pod.'.format(node=node['host']) + ) + + return error_msgs + + def _check_es_cluster_health(self, pods_by_name): + """Exec into the elasticsearch pods and check the cluster health. Returns: list of errors""" + error_msgs = [] + for pod_name in pods_by_name.keys(): + cluster_health_cmd = self._build_es_curl_cmd(pod_name, 'https://localhost:9200/_cluster/health?pretty=true') + cluster_health_data = self._exec_oc(cluster_health_cmd, []) + try: + health_res = json.loads(cluster_health_data) + if not health_res or not health_res.get('status'): + raise ValueError() + except ValueError: + error_msgs.append( + 'Could not retrieve cluster health status from logging ES pod "{pod}".\n' + 'Response was:\n{output}'.format(pod=pod_name, output=cluster_health_data) + ) + continue + + if health_res['status'] not in ['green', 'yellow']: + error_msgs.append( + 'Elasticsearch cluster health status is RED according to pod "{}"'.format(pod_name) + ) + + return error_msgs + + def _check_elasticsearch_diskspace(self, pods_by_name): + """ + Exec into an ES pod and query the diskspace on the persistent volume. + Returns: list of errors + """ + error_msgs = [] + for pod_name in pods_by_name.keys(): + df_cmd = 'exec {} -- df --output=ipcent,pcent /elasticsearch/persistent'.format(pod_name) + disk_output = self._exec_oc(df_cmd, []) + lines = disk_output.splitlines() + # expecting one header looking like 'IUse% Use%' and one body line + body_re = r'\s*(\d+)%?\s+(\d+)%?\s*$' + if len(lines) != 2 or len(lines[0].split()) != 2 or not re.match(body_re, lines[1]): + error_msgs.append( + 'Could not retrieve storage usage from logging ES pod "{pod}".\n' + 'Response to `df` command was:\n{output}'.format(pod=pod_name, output=disk_output) + ) + continue + inode_pct, disk_pct = re.match(body_re, lines[1]).groups() + + inode_pct_thresh = self.get_var('openshift_check_efk_es_inode_pct', default='90') + if int(inode_pct) >= int(inode_pct_thresh): + error_msgs.append( + 'Inode percent usage on the storage volume for logging ES pod "{pod}"\n' + ' is {pct}, greater than threshold {limit}.\n' + ' Note: threshold can be specified in inventory with {param}'.format( + pod=pod_name, + pct=str(inode_pct), + limit=str(inode_pct_thresh), + param='openshift_check_efk_es_inode_pct', + )) + disk_pct_thresh = self.get_var('openshift_check_efk_es_storage_pct', default='80') + if int(disk_pct) >= int(disk_pct_thresh): + error_msgs.append( + 'Disk percent usage on the storage volume for logging ES pod "{pod}"\n' + ' is {pct}, greater than threshold {limit}.\n' + ' Note: threshold can be specified in inventory with {param}'.format( + pod=pod_name, + pct=str(disk_pct), + limit=str(disk_pct_thresh), + param='openshift_check_efk_es_storage_pct', + )) + + return error_msgs + + def _exec_oc(self, cmd_str, extra_args): + return super(Elasticsearch, self).exec_oc( + self.logging_namespace, + cmd_str, + extra_args, + ) diff --git a/roles/openshift_health_checker/openshift_checks/logging/fluentd.py b/roles/openshift_health_checker/openshift_checks/logging/fluentd.py new file mode 100644 index 000000000..063e707a9 --- /dev/null +++ b/roles/openshift_health_checker/openshift_checks/logging/fluentd.py @@ -0,0 +1,167 @@ +"""Check for an aggregated logging Fluentd deployment""" + +import json + +from openshift_checks.logging.logging import LoggingCheck + + +class Fluentd(LoggingCheck): + """Check for an aggregated logging Fluentd deployment""" + + name = "fluentd" + tags = ["health", "logging"] + + logging_namespace = None + + def run(self): + """Check various things and gather errors. Returns: result as hash""" + + self.logging_namespace = self.get_var("openshift_logging_namespace", default="logging") + fluentd_pods, error = super(Fluentd, self).get_pods_for_component( + self.logging_namespace, + "fluentd", + ) + if error: + return {"failed": True, "changed": False, "msg": error} + check_error = self.check_fluentd(fluentd_pods) + + if check_error: + msg = ("The following Fluentd deployment issue was found:" + "\n-------\n" + "{}".format(check_error)) + return {"failed": True, "changed": False, "msg": msg} + + # TODO(lmeyer): run it all again for the ops cluster + return {"failed": False, "changed": False, "msg": 'No problems found with Fluentd deployment.'} + + @staticmethod + def _filter_fluentd_labeled_nodes(nodes_by_name, node_selector): + """Filter to all nodes with fluentd label. Returns dict(name: node), error string""" + label, value = node_selector.split('=', 1) + fluentd_nodes = { + name: node for name, node in nodes_by_name.items() + if node['metadata']['labels'].get(label) == value + } + if not fluentd_nodes: + return None, ( + 'There are no nodes with the fluentd label {label}.\n' + 'This means no logs will be aggregated from the nodes.' + ).format(label=node_selector) + return fluentd_nodes, None + + def _check_node_labeling(self, nodes_by_name, fluentd_nodes, node_selector): + """Note if nodes are not labeled as expected. Returns: error string""" + intended_nodes = self.get_var('openshift_logging_fluentd_hosts', default=['--all']) + if not intended_nodes or '--all' in intended_nodes: + intended_nodes = nodes_by_name.keys() + nodes_missing_labels = set(intended_nodes) - set(fluentd_nodes.keys()) + if nodes_missing_labels: + return ( + 'The following nodes are supposed to be labeled with {label} but are not:\n' + ' {nodes}\n' + 'Fluentd will not aggregate logs from these nodes.' + ).format(label=node_selector, nodes=', '.join(nodes_missing_labels)) + return None + + @staticmethod + def _check_nodes_have_fluentd(pods, fluentd_nodes): + """Make sure fluentd is on all the labeled nodes. Returns: error string""" + unmatched_nodes = fluentd_nodes.copy() + node_names_by_label = { + node['metadata']['labels']['kubernetes.io/hostname']: name + for name, node in fluentd_nodes.items() + } + node_names_by_internal_ip = { + address['address']: name + for name, node in fluentd_nodes.items() + for address in node['status']['addresses'] + if address['type'] == "InternalIP" + } + for pod in pods: + for name in [ + pod['spec']['nodeName'], + node_names_by_internal_ip.get(pod['spec']['nodeName']), + node_names_by_label.get(pod.get('spec', {}).get('host')), + ]: + unmatched_nodes.pop(name, None) + if unmatched_nodes: + return ( + 'The following nodes are supposed to have a Fluentd pod but do not:\n' + '{nodes}' + 'These nodes will not have their logs aggregated.' + ).format(nodes=''.join( + " {}\n".format(name) + for name in unmatched_nodes.keys() + )) + return None + + def _check_fluentd_pods_running(self, pods): + """Make sure all fluentd pods are running. Returns: error string""" + not_running = super(Fluentd, self).not_running_pods(pods) + if not_running: + return ( + 'The following Fluentd pods are supposed to be running but are not:\n' + '{pods}' + 'These pods will not aggregate logs from their nodes.' + ).format(pods=''.join( + " {} ({})\n".format(pod['metadata']['name'], pod['spec'].get('host', 'None')) + for pod in not_running + )) + return None + + def check_fluentd(self, pods): + """Verify fluentd is running everywhere. Returns: error string""" + + node_selector = self.get_var( + 'openshift_logging_fluentd_nodeselector', + default='logging-infra-fluentd=true' + ) + + nodes_by_name, error = self.get_nodes_by_name() + + if error: + return error + fluentd_nodes, error = self._filter_fluentd_labeled_nodes(nodes_by_name, node_selector) + if error: + return error + + error_msgs = [] + error = self._check_node_labeling(nodes_by_name, fluentd_nodes, node_selector) + if error: + error_msgs.append(error) + error = self._check_nodes_have_fluentd(pods, fluentd_nodes) + if error: + error_msgs.append(error) + error = self._check_fluentd_pods_running(pods) + if error: + error_msgs.append(error) + + # Make sure there are no extra fluentd pods + if len(pods) > len(fluentd_nodes): + error_msgs.append( + 'There are more Fluentd pods running than nodes labeled.\n' + 'This may not cause problems with logging but it likely indicates something wrong.' + ) + + return '\n'.join(error_msgs) + + def get_nodes_by_name(self): + """Retrieve all the node definitions. Returns: dict(name: node), error string""" + nodes_json = self._exec_oc("get nodes -o json", []) + try: + nodes = json.loads(nodes_json) + except ValueError: # no valid json - should not happen + return None, "Could not obtain a list of nodes to validate fluentd. Output from oc get:\n" + nodes_json + if not nodes or not nodes.get('items'): # also should not happen + return None, "No nodes appear to be defined according to the API." + return { + node['metadata']['name']: node + for node in nodes['items'] + }, None + + def _exec_oc(self, cmd_str, extra_args): + return super(Fluentd, self).exec_oc( + self.logging_namespace, + cmd_str, + extra_args, + ) diff --git a/roles/openshift_health_checker/openshift_checks/logging/kibana.py b/roles/openshift_health_checker/openshift_checks/logging/kibana.py new file mode 100644 index 000000000..60f94e106 --- /dev/null +++ b/roles/openshift_health_checker/openshift_checks/logging/kibana.py @@ -0,0 +1,226 @@ +""" +Module for performing checks on a Kibana logging deployment +""" + +import json +import ssl + +try: + from urllib2 import HTTPError, URLError + import urllib2 +except ImportError: + from urllib.error import HTTPError, URLError + import urllib.request as urllib2 + +from openshift_checks.logging.logging import LoggingCheck + + +class Kibana(LoggingCheck): + """Module that checks an integrated logging Kibana deployment""" + + name = "kibana" + tags = ["health", "logging"] + + logging_namespace = None + + def run(self): + """Check various things and gather errors. Returns: result as hash""" + + self.logging_namespace = self.get_var("openshift_logging_namespace", default="logging") + kibana_pods, error = super(Kibana, self).get_pods_for_component( + self.logging_namespace, + "kibana", + ) + if error: + return {"failed": True, "changed": False, "msg": error} + check_error = self.check_kibana(kibana_pods) + + if not check_error: + check_error = self._check_kibana_route() + + if check_error: + msg = ("The following Kibana deployment issue was found:" + "\n-------\n" + "{}".format(check_error)) + return {"failed": True, "changed": False, "msg": msg} + + # TODO(lmeyer): run it all again for the ops cluster + return {"failed": False, "changed": False, "msg": 'No problems found with Kibana deployment.'} + + def _verify_url_internal(self, url): + """ + Try to reach a URL from the host. + Returns: success (bool), reason (for failure) + """ + args = dict( + url=url, + follow_redirects='none', + validate_certs='no', # likely to be signed with internal CA + # TODO(lmeyer): give users option to validate certs + status_code=302, + ) + result = self.execute_module('uri', args) + if result.get('failed'): + return result['msg'] + return None + + @staticmethod + def _verify_url_external(url): + """ + Try to reach a URL from ansible control host. + Returns: success (bool), reason (for failure) + """ + # This actually checks from the ansible control host, which may or may not + # really be "external" to the cluster. + + # Disable SSL cert validation to work around internally signed certs + ctx = ssl.create_default_context() + ctx.check_hostname = False # or setting CERT_NONE is refused + ctx.verify_mode = ssl.CERT_NONE + + # Verify that the url is returning a valid response + try: + # We only care if the url connects and responds + return_code = urllib2.urlopen(url, context=ctx).getcode() + except HTTPError as httperr: + return httperr.reason + except URLError as urlerr: + return str(urlerr) + + # there appears to be no way to prevent urlopen from following redirects + if return_code != 200: + return 'Expected success (200) but got return code {}'.format(int(return_code)) + + return None + + def check_kibana(self, pods): + """Check to see if Kibana is up and working. Returns: error string.""" + + if not pods: + return "There are no Kibana pods deployed, so no access to the logging UI." + + not_running = self.not_running_pods(pods) + if len(not_running) == len(pods): + return "No Kibana pod is in a running state, so there is no access to the logging UI." + elif not_running: + return ( + "The following Kibana pods are not currently in a running state:\n" + "{pods}" + "However at least one is, so service may not be impacted." + ).format(pods="".join(" " + pod['metadata']['name'] + "\n" for pod in not_running)) + + return None + + def _get_kibana_url(self): + """ + Get kibana route or report error. + Returns: url (or empty), reason for failure + """ + + # Get logging url + get_route = self._exec_oc("get route logging-kibana -o json", []) + if not get_route: + return None, 'no_route_exists' + + route = json.loads(get_route) + + # check that the route has been accepted by a router + ingress = route["status"]["ingress"] + # ingress can be null if there is no router, or empty if not routed + if not ingress or not ingress[0]: + return None, 'route_not_accepted' + + host = route.get("spec", {}).get("host") + if not host: + return None, 'route_missing_host' + + return 'https://{}/'.format(host), None + + def _check_kibana_route(self): + """ + Check to see if kibana route is up and working. + Returns: error string + """ + known_errors = dict( + no_route_exists=( + 'No route is defined for Kibana in the logging namespace,\n' + 'so the logging stack is not accessible. Is logging deployed?\n' + 'Did something remove the logging-kibana route?' + ), + route_not_accepted=( + 'The logging-kibana route is not being routed by any router.\n' + 'Is the router deployed and working?' + ), + route_missing_host=( + 'The logging-kibana route has no hostname defined,\n' + 'which should never happen. Did something alter its definition?' + ), + ) + + kibana_url, error = self._get_kibana_url() + if not kibana_url: + return known_errors.get(error, error) + + # first, check that kibana is reachable from the master. + error = self._verify_url_internal(kibana_url) + if error: + if 'urlopen error [Errno 111] Connection refused' in error: + error = ( + 'Failed to connect from this master to Kibana URL {url}\n' + 'Is kibana running, and is at least one router routing to it?' + ).format(url=kibana_url) + elif 'urlopen error [Errno -2] Name or service not known' in error: + error = ( + 'Failed to connect from this master to Kibana URL {url}\n' + 'because the hostname does not resolve.\n' + 'Is DNS configured for the Kibana hostname?' + ).format(url=kibana_url) + elif 'Status code was not' in error: + error = ( + 'A request from this master to the Kibana URL {url}\n' + 'did not return the correct status code (302).\n' + 'This could mean that Kibana is malfunctioning, the hostname is\n' + 'resolving incorrectly, or other network issues. The output was:\n' + ' {error}' + ).format(url=kibana_url, error=error) + return 'Error validating the logging Kibana route:\n' + error + + # in production we would like the kibana route to work from outside the + # cluster too; but that may not be the case, so allow disabling just this part. + if not self.get_var("openshift_check_efk_kibana_external", default=True): + return None + error = self._verify_url_external(kibana_url) + if error: + if 'urlopen error [Errno 111] Connection refused' in error: + error = ( + 'Failed to connect from the Ansible control host to Kibana URL {url}\n' + 'Is the router for the Kibana hostname exposed externally?' + ).format(url=kibana_url) + elif 'urlopen error [Errno -2] Name or service not known' in error: + error = ( + 'Failed to resolve the Kibana hostname in {url}\n' + 'from the Ansible control host.\n' + 'Is DNS configured to resolve this Kibana hostname externally?' + ).format(url=kibana_url) + elif 'Expected success (200)' in error: + error = ( + 'A request to Kibana at {url}\n' + 'returned the wrong error code:\n' + ' {error}\n' + 'This could mean that Kibana is malfunctioning, the hostname is\n' + 'resolving incorrectly, or other network issues.' + ).format(url=kibana_url, error=error) + error = ( + 'Error validating the logging Kibana route:\n{error}\n' + 'To disable external Kibana route validation, set in your inventory:\n' + ' openshift_check_efk_kibana_external=False' + ).format(error=error) + return error + return None + + def _exec_oc(self, cmd_str, extra_args): + return super(Kibana, self).exec_oc( + self.logging_namespace, + cmd_str, + extra_args, + ) diff --git a/roles/openshift_health_checker/openshift_checks/logging/logging.py b/roles/openshift_health_checker/openshift_checks/logging/logging.py new file mode 100644 index 000000000..a48e1c728 --- /dev/null +++ b/roles/openshift_health_checker/openshift_checks/logging/logging.py @@ -0,0 +1,92 @@ +""" +Util functions for performing checks on an Elasticsearch, Fluentd, and Kibana stack +""" + +import json +import os + +from openshift_checks import OpenShiftCheck, OpenShiftCheckException + + +class LoggingCheck(OpenShiftCheck): + """Base class for OpenShift aggregated logging component checks""" + + name = "logging" + logging_namespace = "logging" + + def is_active(self): + logging_deployed = self.get_var("openshift_hosted_logging_deploy", default=False) + return logging_deployed and super(LoggingCheck, self).is_active() and self.is_first_master() + + def is_first_master(self): + """Determine if running on first master. Returns: bool""" + # Note: It would be nice to use membership in oo_first_master group, however for now it + # seems best to avoid requiring that setup and just check this is the first master. + hostname = self.get_var("ansible_ssh_host") or [None] + masters = self.get_var("groups", "masters", default=None) or [None] + return masters[0] == hostname + + def run(self): + pass + + def get_pods_for_component(self, namespace, logging_component): + """Get all pods for a given component. Returns: list of pods for component, error string""" + pod_output = self.exec_oc( + namespace, + "get pods -l component={} -o json".format(logging_component), + [], + ) + try: + pods = json.loads(pod_output) + if not pods or not pods.get('items'): + raise ValueError() + except ValueError: + # successful run but non-parsing data generally means there were no pods in the namespace + return None, 'No pods were found for the "{}" logging component.'.format(logging_component) + + return pods['items'], None + + @staticmethod + def not_running_pods(pods): + """Returns: list of pods not in a ready and running state""" + return [ + pod for pod in pods + if not pod.get("status", {}).get("containerStatuses") or any( + container['ready'] is False + for container in pod['status']['containerStatuses'] + ) or not any( + condition['type'] == 'Ready' and condition['status'] == 'True' + for condition in pod['status'].get('conditions', []) + ) + ] + + def exec_oc(self, namespace="logging", cmd_str="", extra_args=None): + """ + Execute an 'oc' command in the remote host. + Returns: output of command and namespace, + or raises OpenShiftCheckException on error + """ + config_base = self.get_var("openshift", "common", "config_base") + args = { + "namespace": namespace, + "config_file": os.path.join(config_base, "master", "admin.kubeconfig"), + "cmd": cmd_str, + "extra_args": list(extra_args) if extra_args else [], + } + + result = self.execute_module("ocutil", args) + if result.get("failed"): + msg = ( + 'Unexpected error using `oc` to validate the logging stack components.\n' + 'Error executing `oc {cmd}`:\n' + '{error}' + ).format(cmd=args['cmd'], error=result['result']) + + if result['result'] == '[Errno 2] No such file or directory': + msg = ( + "This host is supposed to be a master but does not have the `oc` command where expected.\n" + "Has an installation been run on this host yet?" + ) + raise OpenShiftCheckException(msg) + + return result.get("result", "") diff --git a/roles/openshift_health_checker/openshift_checks/logging/logging_index_time.py b/roles/openshift_health_checker/openshift_checks/logging/logging_index_time.py new file mode 100644 index 000000000..b24e88e05 --- /dev/null +++ b/roles/openshift_health_checker/openshift_checks/logging/logging_index_time.py @@ -0,0 +1,130 @@ +""" +Check for ensuring logs from pods can be queried in a reasonable amount of time. +""" + +import json +import time + +from uuid import uuid4 + +from openshift_checks import OpenShiftCheckException +from openshift_checks.logging.logging import LoggingCheck + + +ES_CMD_TIMEOUT_SECONDS = 30 + + +class LoggingIndexTime(LoggingCheck): + """Check that pod logs are aggregated and indexed in ElasticSearch within a reasonable amount of time.""" + name = "logging_index_time" + tags = ["health", "logging"] + + logging_namespace = "logging" + + def run(self): + """Add log entry by making unique request to Kibana. Check for unique entry in the ElasticSearch pod logs.""" + try: + log_index_timeout = int( + self.get_var("openshift_check_logging_index_timeout_seconds", default=ES_CMD_TIMEOUT_SECONDS) + ) + except ValueError: + return { + "failed": True, + "msg": ('Invalid value provided for "openshift_check_logging_index_timeout_seconds". ' + 'Value must be an integer representing an amount in seconds.'), + } + + running_component_pods = dict() + + # get all component pods + self.logging_namespace = self.get_var("openshift_logging_namespace", default=self.logging_namespace) + for component, name in (['kibana', 'Kibana'], ['es', 'Elasticsearch']): + pods, error = self.get_pods_for_component(self.logging_namespace, component) + + if error: + msg = 'Unable to retrieve pods for the {} logging component: {}' + return {"failed": True, "changed": False, "msg": msg.format(name, error)} + + running_pods = self.running_pods(pods) + + if not running_pods: + msg = ('No {} pods in the "Running" state were found.' + 'At least one pod is required in order to perform this check.') + return {"failed": True, "changed": False, "msg": msg.format(name)} + + running_component_pods[component] = running_pods + + uuid = self.curl_kibana_with_uuid(running_component_pods["kibana"][0]) + self.wait_until_cmd_or_err(running_component_pods["es"][0], uuid, log_index_timeout) + return {} + + def wait_until_cmd_or_err(self, es_pod, uuid, timeout_secs): + """Retry an Elasticsearch query every second until query success, or a defined + length of time has passed.""" + deadline = time.time() + timeout_secs + interval = 1 + while not self.query_es_from_es(es_pod, uuid): + if time.time() + interval > deadline: + msg = "expecting match in Elasticsearch for message with uuid {}, but no matches were found after {}s." + raise OpenShiftCheckException(msg.format(uuid, timeout_secs)) + time.sleep(interval) + + def curl_kibana_with_uuid(self, kibana_pod): + """curl Kibana with a unique uuid.""" + uuid = self.generate_uuid() + pod_name = kibana_pod["metadata"]["name"] + exec_cmd = "exec {pod_name} -c kibana -- curl --max-time 30 -s http://localhost:5601/{uuid}" + exec_cmd = exec_cmd.format(pod_name=pod_name, uuid=uuid) + + error_str = self.exec_oc(self.logging_namespace, exec_cmd, []) + + try: + error_code = json.loads(error_str)["statusCode"] + except KeyError: + msg = ('invalid response returned from Kibana request (Missing "statusCode" key):\n' + 'Command: {}\nResponse: {}').format(exec_cmd, error_str) + raise OpenShiftCheckException(msg) + except ValueError: + msg = ('invalid response returned from Kibana request (Non-JSON output):\n' + 'Command: {}\nResponse: {}').format(exec_cmd, error_str) + raise OpenShiftCheckException(msg) + + if error_code != 404: + msg = 'invalid error code returned from Kibana request. Expecting error code "404", but got "{}" instead.' + raise OpenShiftCheckException(msg.format(error_code)) + + return uuid + + def query_es_from_es(self, es_pod, uuid): + """curl the Elasticsearch pod and look for a unique uuid in its logs.""" + pod_name = es_pod["metadata"]["name"] + exec_cmd = ( + "exec {pod_name} -- curl --max-time 30 -s -f " + "--cacert /etc/elasticsearch/secret/admin-ca " + "--cert /etc/elasticsearch/secret/admin-cert " + "--key /etc/elasticsearch/secret/admin-key " + "https://logging-es:9200/project.{namespace}*/_count?q=message:{uuid}" + ) + exec_cmd = exec_cmd.format(pod_name=pod_name, namespace=self.logging_namespace, uuid=uuid) + result = self.exec_oc(self.logging_namespace, exec_cmd, []) + + try: + count = json.loads(result)["count"] + except KeyError: + msg = 'invalid response from Elasticsearch query:\n"{}"\nMissing "count" key:\n{}' + raise OpenShiftCheckException(msg.format(exec_cmd, result)) + except ValueError: + msg = 'invalid response from Elasticsearch query:\n"{}"\nNon-JSON output:\n{}' + raise OpenShiftCheckException(msg.format(exec_cmd, result)) + + return count + + @staticmethod + def running_pods(pods): + """Filter pods that are running.""" + return [pod for pod in pods if pod['status']['phase'] == 'Running'] + + @staticmethod + def generate_uuid(): + """Wrap uuid generator. Allows for testing with expected values.""" + return str(uuid4()) |