summaryrefslogtreecommitdiffstats
path: root/roles
diff options
context:
space:
mode:
Diffstat (limited to 'roles')
-rw-r--r--roles/openshift_master/tasks/main.yml6
-rw-r--r--roles/openshift_node/tasks/main.yml6
-rw-r--r--roles/os_zabbix/vars/template_docker.yml10
-rw-r--r--roles/os_zabbix/vars/template_openshift_master.yml8
-rw-r--r--roles/os_zabbix/vars/template_zagg_server.yml16
-rw-r--r--roles/oso_host_monitoring/templates/oso-rhel7-zagg-client.service.j22
6 files changed, 38 insertions, 10 deletions
diff --git a/roles/openshift_master/tasks/main.yml b/roles/openshift_master/tasks/main.yml
index 8995863ec..43647cc49 100644
--- a/roles/openshift_master/tasks/main.yml
+++ b/roles/openshift_master/tasks/main.yml
@@ -228,7 +228,7 @@
register: start_result
- set_fact:
- master_service_status_changed = start_result | changed
+ master_service_status_changed: start_result | changed
when: not openshift_master_ha | bool
- name: Start and enable master api
@@ -237,7 +237,7 @@
register: start_result
- set_fact:
- master_api_service_status_changed = start_result | changed
+ master_api_service_status_changed: start_result | changed
when: openshift_master_ha | bool and openshift.master.cluster_method == 'native'
- name: Start and enable master controller
@@ -246,7 +246,7 @@
register: start_result
- set_fact:
- master_controllers_service_status_changed = start_result | changed
+ master_controllers_service_status_changed: start_result | changed
when: openshift_master_ha | bool and openshift.master.cluster_method == 'native'
- name: Install cluster packages
diff --git a/roles/openshift_node/tasks/main.yml b/roles/openshift_node/tasks/main.yml
index eef7bec9a..38bffc2e5 100644
--- a/roles/openshift_node/tasks/main.yml
+++ b/roles/openshift_node/tasks/main.yml
@@ -85,11 +85,11 @@
docker_additional_registries: "{{ lookup('oo_option', 'docker_additional_registries')
| oo_split() | union(['registry.access.redhat.com'])
| difference(['']) }}"
- when: openshift.common.deployment_type == 'enterprise'
+ when: openshift.common.deployment_type in ['enterprise', 'openshift-enterprise', 'atomic-enterprise']
- set_fact:
docker_additional_registries: "{{ lookup('oo_option', 'docker_additional_registries')
| oo_split() | difference(['']) }}"
- when: openshift.common.deployment_type != 'enterprise'
+ when: openshift.common.deployment_type not in ['enterprise', 'openshift-enterprise', 'atomic-enterprise']
- name: Add personal registries
lineinfile:
@@ -131,4 +131,4 @@
register: start_result
- set_fact:
- node_service_status_changed = start_result | changed
+ node_service_status_changed: start_result | changed
diff --git a/roles/os_zabbix/vars/template_docker.yml b/roles/os_zabbix/vars/template_docker.yml
index bfabf50c5..91a2c400e 100644
--- a/roles/os_zabbix/vars/template_docker.yml
+++ b/roles/os_zabbix/vars/template_docker.yml
@@ -12,6 +12,11 @@ g_template_docker:
- Docker Daemon
value_type: int
+ - key: docker.container.dns.resolution
+ applications:
+ - Docker Daemon
+ value_type: int
+
- key: docker.storage.is_loopback
applications:
- Docker Storage
@@ -62,6 +67,11 @@ g_template_docker:
url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_docker_ping.asciidoc'
priority: high
+ - name: 'docker.container.dns.resolution failed on {HOST.NAME}'
+ expression: '{Template Docker:docker.container.dns.resolution.max(#3)}>0'
+ url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_docker_dns.asciidoc'
+ priority: high
+
- name: 'Docker storage is using LOOPBACK on {HOST.NAME}'
expression: '{Template Docker:docker.storage.is_loopback.last()}<>0'
url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_docker_loopback.asciidoc'
diff --git a/roles/os_zabbix/vars/template_openshift_master.yml b/roles/os_zabbix/vars/template_openshift_master.yml
index 514d6fd24..a0ba8d104 100644
--- a/roles/os_zabbix/vars/template_openshift_master.yml
+++ b/roles/os_zabbix/vars/template_openshift_master.yml
@@ -269,6 +269,14 @@ g_template_openshift_master:
- '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'
+ url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_create_app.asciidoc'
+ dependencies:
+ - 'Openshift Master process not running on {HOST.NAME}'
+ description: The application create loop has failed 4 or more times in the last hour
+ priority: avg
+
- name: 'Openshift Master API health check is failing on {HOST.NAME}'
expression: '{Template Openshift Master:openshift.master.api.healthz.max(#3)}<1'
url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_master.asciidoc'
diff --git a/roles/os_zabbix/vars/template_zagg_server.yml b/roles/os_zabbix/vars/template_zagg_server.yml
index 0e8e53bb7..db5665993 100644
--- a/roles/os_zabbix/vars/template_zagg_server.yml
+++ b/roles/os_zabbix/vars/template_zagg_server.yml
@@ -7,7 +7,12 @@ g_template_zagg_server:
- Zagg Server
value_type: int
- - key: zagg.server.processor.errors
+ - key: zagg.server.metrics.errors
+ applications:
+ - Zagg Server
+ value_type: int
+
+ - key: zagg.server.heartbeat.errors
applications:
- Zagg Server
value_type: int
@@ -18,8 +23,13 @@ g_template_zagg_server:
value_type: int
ztriggers:
- - name: 'Error sending metrics on {HOST.NAME}'
- expression: '{Template Zagg Server:zagg.server.processor.errors.min(#3)}>0'
+ - name: 'Error processing metrics on {HOST.NAME}'
+ expression: '{Template Zagg Server:zagg.server.metrics.errors.min(#3)}>0'
+ url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/zagg_server.asciidoc'
+ priority: average
+
+ - name: 'Error processing heartbeats on {HOST.NAME}'
+ expression: '{Template Zagg Server:zagg.server.heartbeat.errors.min(#3)}>0'
url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/zagg_server.asciidoc'
priority: average
diff --git a/roles/oso_host_monitoring/templates/oso-rhel7-zagg-client.service.j2 b/roles/oso_host_monitoring/templates/oso-rhel7-zagg-client.service.j2
index 978e40b88..bcc8a5e03 100644
--- a/roles/oso_host_monitoring/templates/oso-rhel7-zagg-client.service.j2
+++ b/roles/oso_host_monitoring/templates/oso-rhel7-zagg-client.service.j2
@@ -42,7 +42,7 @@ ExecStart=/usr/bin/docker run --name {{ osohm_zagg_client }}
-v /etc/localtime:/etc/localtime \
-v /run/pcp:/run/pcp \
-v /var/run/docker.sock:/var/run/docker.sock \
- -v /var/run/openvswitch/db.sock:/var/run/openvswitch/db.sock \
+ -v /var/run/openvswitch:/var/run/openvswitch \
{% if hostvars[inventory_hostname]['ec2_tag_host-type'] == 'master' %}
-v /etc/openshift/master/admin.kubeconfig:/etc/openshift/master/admin.kubeconfig \
-v /etc/openshift/master/master.etcd-client.crt:/etc/openshift/master/master.etcd-client.crt \