diff options
| -rw-r--r-- | roles/lib_dyn/library/dyn_record.py | 4 | ||||
| -rw-r--r-- | roles/lib_dyn/tasks/main.yml | 2 | ||||
| -rw-r--r-- | roles/openshift_master/tasks/main.yml | 2 | ||||
| -rw-r--r-- | roles/os_zabbix/vars/template_openshift_master.yml | 19 |
4 files changed, 25 insertions, 2 deletions
diff --git a/roles/lib_dyn/library/dyn_record.py b/roles/lib_dyn/library/dyn_record.py index 5e088a674..f2796ccf2 100644 --- a/roles/lib_dyn/library/dyn_record.py +++ b/roles/lib_dyn/library/dyn_record.py @@ -224,11 +224,13 @@ def main(): # First get a list of existing records for the node values = get_record_values(dyn_node_records) value_key = get_record_key(module.params['record_type']) + param_value = module.params['record_value'] # Check to see if the record is already in place before doing anything. if (dyn_node_records and dyn_node_records[value_key][0].ttl == module.params['record_ttl'] and - module.params['record_value'] in values[value_key]): + (param_value in values[value_key] or + param_value + '.' in values[value_key])): module.exit_json(changed=False) diff --git a/roles/lib_dyn/tasks/main.yml b/roles/lib_dyn/tasks/main.yml index 9b3b1b0b9..965962928 100644 --- a/roles/lib_dyn/tasks/main.yml +++ b/roles/lib_dyn/tasks/main.yml @@ -3,3 +3,5 @@ - name: Make sure python-dyn is installed yum: name=python-dyn state=present + tags: + - lib_dyn diff --git a/roles/openshift_master/tasks/main.yml b/roles/openshift_master/tasks/main.yml index 3f7701519..3613808b6 100644 --- a/roles/openshift_master/tasks/main.yml +++ b/roles/openshift_master/tasks/main.yml @@ -287,7 +287,7 @@ - name: Mask master service command: systemctl mask {{ openshift.common.service_type }}-master - when: openshift_master_ha | bool and openshift.master.cluster_method == 'native' + when: openshift_master_ha | bool and openshift.master.cluster_method == 'native' and not openshift.common.is_atomic | bool - name: Start and enable master api service: name={{ openshift.common.service_type }}-master-api enabled=yes state=started diff --git a/roles/os_zabbix/vars/template_openshift_master.yml b/roles/os_zabbix/vars/template_openshift_master.yml index 5aae2496a..1824d7881 100644 --- a/roles/os_zabbix/vars/template_openshift_master.yml +++ b/roles/os_zabbix/vars/template_openshift_master.yml @@ -201,6 +201,18 @@ g_template_openshift_master: applications: - Openshift Master Metrics + - key: openshift.master.nodesnotready.count + description: "This check shows how many nodes in a cluster are in NotReady state." + type: int + applications: + - Openshift Master + + - key: openshift.master.nodesnotschedulable.count + description: "This check shows how many nodes in a cluster are not schedulable." + type: int + applications: + - Openshift Master + - key: openshift.master.apiserver.latency.summary.pods.quantile.list.5 description: "Value from https://master/metrics. This is the time, in miliseconds, that 50% of the pod operations have taken to completed." type: int @@ -361,6 +373,13 @@ g_template_openshift_master: - 'Openshift Master API health check is failing on {HOST.NAME}' priority: high + - name: 'Hosts not ready according to {HOST.NAME}' + expression: '{Template Openshift Master:openshift.master.nodesnotready.count.last(#2)}>0' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_node.asciidoc' + dependencies: + - 'Openshift Master process not running on {HOST.NAME}' + priority: high + zgraphs: - name: Openshift Master API Server Latency Pods LIST Quantiles width: 900 |