diff options
Diffstat (limited to 'roles')
| -rwxr-xr-x | roles/openshift_facts/library/openshift_facts.py | 24 | ||||
| -rw-r--r-- | roles/os_zabbix/vars/template_openshift_master.yml | 25 |
2 files changed, 40 insertions, 9 deletions
diff --git a/roles/openshift_facts/library/openshift_facts.py b/roles/openshift_facts/library/openshift_facts.py index 85c8abdf0..d11af307b 100755 --- a/roles/openshift_facts/library/openshift_facts.py +++ b/roles/openshift_facts/library/openshift_facts.py @@ -1070,6 +1070,28 @@ def set_container_facts_if_unset(facts): return facts +def set_installed_variant_rpm_facts(facts): + """ Set RPM facts of installed variant + Args: + facts (dict): existing facts + Returns: + dict: the facts dict updated with installed_variant_rpms + """ + installed_rpms = [] + for base_rpm in ['openshift', 'atomic-openshift', 'origin']: + optional_rpms = ['master', 'node', 'clients', 'sdn-ovs'] + variant_rpms = [base_rpm] + \ + ['{0}-{1}'.format(base_rpm, r) for r in optional_rpms] + \ + ['tuned-profiles-%s-node' % base_rpm] + for rpm in variant_rpms: + exit_code, _, _ = module.run_command(['rpm', '-q', rpm]) + if exit_code == 0: + installed_rpms.append(rpm) + + facts['common']['installed_variant_rpms'] = installed_rpms + return facts + + class OpenShiftFactsInternalError(Exception): """Origin Facts Error""" @@ -1159,6 +1181,8 @@ class OpenShiftFacts(object): facts = set_aggregate_facts(facts) facts = set_etcd_facts_if_unset(facts) facts = set_container_facts_if_unset(facts) + if not facts['common']['is_containerized']: + facts = set_installed_variant_rpm_facts(facts) return dict(openshift=facts) def get_defaults(self, roles): diff --git a/roles/os_zabbix/vars/template_openshift_master.yml b/roles/os_zabbix/vars/template_openshift_master.yml index 9d20eb012..5aae2496a 100644 --- a/roles/os_zabbix/vars/template_openshift_master.yml +++ b/roles/os_zabbix/vars/template_openshift_master.yml @@ -2,13 +2,13 @@ g_template_openshift_master: name: Template Openshift Master zitems: - - name: create_app + - name: openshift.master.app.create applications: - Openshift Master - key: create_app + key: openshift.master.app.create - - key: openshift.master.registry.healthz - description: "Shows the health status of the cluster's docker registry" + - key: openshift.master.registry.healthy_pct + description: "Shows the percentage of healthy registries in the cluster" type: int applications: - Openshift Master @@ -288,14 +288,14 @@ g_template_openshift_master: # Put triggers that depend on other triggers here (deps must be created first) - name: 'Application creation has failed on {HOST.NAME}' - expression: '{Template Openshift Master:create_app.last(#1)}=1 and {Template Openshift Master:create_app.last(#2)}=1' + expression: '{Template Openshift Master:openshift.master.app.create.last(#1)}=1 and {Template Openshift Master:openshift.master.app.create.last(#2)}=1' url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_create_app.asciidoc' dependencies: - 'Openshift Master process not running on {HOST.NAME}' priority: avg - name: 'Application creation has failed multiple times in the last hour on {HOST.NAME}' - expression: '{Template Openshift Master:create_app.sum(1h)}>3' + expression: '{Template Openshift Master:openshift.master.app.create.sum(1h)}>3' url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_create_app.asciidoc' dependencies: - 'Openshift Master process not running on {HOST.NAME}' @@ -333,9 +333,16 @@ g_template_openshift_master: - 'Openshift Master process not running on {HOST.NAME}' priority: avg - - name: 'Docker Registry check failed on {HOST.NAME}' - expression: '{Template Openshift Master:openshift.master.registry.healthz.max(#2)}<1' - url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_master.asciidoc' + - name: 'One or more Docker Registries is unhealthy according to {HOST.NAME}' + expression: '{Template Openshift Master:openshift.master.registry.healthy_pct.last(#2)}<100 and {Template Openshift Master:openshift.master.registry.healthy_pct.max(#2)}>50' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_registry.asciidoc' + dependencies: + - 'Openshift Master process not running on {HOST.NAME}' + priority: avg + + - name: 'Multiple Docker Registries are unhealthy according to {HOST.NAME}' + expression: '{Template Openshift Master:openshift.master.registry.healthy_pct.last(#2)}<51' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_registry.asciidoc' dependencies: - 'Openshift Master process not running on {HOST.NAME}' priority: high |