Merge remote-tracking branch 'upstream/stacklight' into merge-branch
Change-Id: I02375439dce9dca8098c64768d8483cb9e47820d
This commit is contained in:
Simon Pasquier
commit
96a8ed3c84
|
|
@ -11,3 +11,5 @@ parameters:
|
|||
enabled: true
|
||||
config:
|
||||
enabled: true
|
||||
grafana:
|
||||
enabled: true
|
||||
|
|
|
|||
|
|
@ -0,0 +1,11 @@
|
|||
Import "hypervisor_stats"
|
||||
|
||||
<Module "hypervisor_stats">
|
||||
KeystoneUrl "{{ plugin.url }}"
|
||||
Tenant "{{ plugin.project }}"
|
||||
Username "{{ plugin.user }}"
|
||||
Password "{{ plugin.password }}"
|
||||
Timeout "5"
|
||||
CpuAllocationRatio "{{ plugin.cpu_allocation_ratio }}"
|
||||
</Module>
|
||||
|
||||
|
|
@ -1,10 +1,9 @@
|
|||
LoadPlugin libvirt
|
||||
<LoadPlugin libvirt>
|
||||
Globals false
|
||||
</LoadPlugin>
|
||||
|
||||
<Plugin "libvirt">
|
||||
Connection "qemu:///system"
|
||||
RefreshInterval 60
|
||||
# Domain "nova"
|
||||
# BlockDevice "name:device"
|
||||
# InterfaceDevice "name:interface"
|
||||
# IgnoreSelected false
|
||||
HostnameFormat "uuid"
|
||||
</Plugin>
|
||||
|
|
|
|||
|
|
@ -0,0 +1,5 @@
|
|||
Import "collectd_libvirt_check"
|
||||
|
||||
<Module "collectd_libvirt_check">
|
||||
Uri "qemu:///system"
|
||||
</Module>
|
||||
|
|
@ -1,17 +0,0 @@
|
|||
<LoadPlugin "python">
|
||||
Globals true
|
||||
</LoadPlugin>
|
||||
|
||||
<Plugin "python">
|
||||
ModulePath "/usr/lib/collectd-plugins/openstack"
|
||||
|
||||
Import "openstack_nova"
|
||||
|
||||
<Module "openstack_nova">
|
||||
Username "admin"
|
||||
Password "123456"
|
||||
TenantName "openstack"
|
||||
AuthURL "https://api.example.com:5000/v2.0"
|
||||
Verbose "False"
|
||||
</Module>
|
||||
</Plugin>
|
||||
|
|
@ -0,0 +1,10 @@
|
|||
Import "openstack_nova"
|
||||
|
||||
<Module "openstack_nova">
|
||||
KeystoneUrl "{{ plugin.url }}"
|
||||
Tenant "{{ plugin.project }}"
|
||||
Username "{{ plugin.user }}"
|
||||
Password "{{ plugin.password }}"
|
||||
Timeout "20"
|
||||
MaxRetries "2"
|
||||
</Module>
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
|
|
@ -1,14 +1,36 @@
|
|||
{%- if pillar.nova.compute is defined %}
|
||||
plugin:
|
||||
local_plugin:
|
||||
nova_compute_libvirt:
|
||||
plugin: interface
|
||||
interval: 60
|
||||
plugin: libvirt
|
||||
template: nova/files/collectd_libvirt.conf
|
||||
nova_libvirt_check:
|
||||
plugin: python
|
||||
template: nova/files/collectd_libvirt_check.conf
|
||||
{%- endif %}
|
||||
{%- if pillar.nova.controller is defined %}
|
||||
plugin: {}
|
||||
# nova_controller_nova:
|
||||
# plugin: nova
|
||||
# interval: 60
|
||||
# template: nova/files/collectd_nova.conf
|
||||
{%- endif %}
|
||||
{% from "nova/map.jinja" import controller with context %}
|
||||
{%- if controller.get('enabled', False) %}
|
||||
local_plugin:
|
||||
collectd_check_local_endpoint:
|
||||
endpoint:
|
||||
nova-api:
|
||||
expected_code: 200
|
||||
url: "http://{{ controller.bind.private_address|replace('0.0.0.0', '127.0.0.1') }}:8774/"
|
||||
remote_plugin:
|
||||
nova_resources:
|
||||
plugin: python
|
||||
template: nova/files/collectd_openstack_nova.conf
|
||||
url: "http://{{ controller.identity.host }}:{{ controller.identity.port }}/v{% if controller.identity.get('api_version', 2)|int == 2 %}2.0{% else %}3{% endif %}"
|
||||
project: {{ controller.identity.tenant }}
|
||||
user: {{ controller.identity.user }}
|
||||
password: {{ controller.identity.password }}
|
||||
nova_hypervisor_stats:
|
||||
plugin: python
|
||||
template: nova/files/collectd_hypervisor_stats.conf
|
||||
url: "http://{{ controller.identity.host }}:{{ controller.identity.port }}/v{% if controller.identity.get('api_version', 2)|int == 2 %}2.0{% else %}3{% endif %}"
|
||||
project: {{ controller.identity.tenant }}
|
||||
user: {{ controller.identity.user }}
|
||||
password: {{ controller.identity.password }}
|
||||
cpu_allocation_ratio: {{ controller.cpu_allocation_ratio }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
|
|
|
|||
|
|
@ -0,0 +1,7 @@
|
|||
dashboard:
|
||||
nova:
|
||||
format: json
|
||||
template: nova/files/grafana_dashboards/nova_influxdb.json
|
||||
hypervisor:
|
||||
format: json
|
||||
template: nova/files/grafana_dashboards/hypervisor_influxdb.json
|
||||
|
|
@ -1,54 +1,300 @@
|
|||
input:
|
||||
{%- if pillar.nova.compute is defined %}
|
||||
nova_compute:
|
||||
engine: logstreamer
|
||||
log_directory: /var/log/nova
|
||||
file_match: nova-compute\.log\.?(?P<Index>\d+)?(.gz)?
|
||||
priority: ["^Index"]
|
||||
decoder: multidecoder_openstack
|
||||
libvirtd:
|
||||
engine: logstreamer
|
||||
log_directory: /var/log/libvirt
|
||||
file_match: libvirtd\.log\.?(?P<Index>\d+)?(.gz)?
|
||||
priority: ['^Index']
|
||||
decoder: multidecoder_libvirtd
|
||||
{%- endif %}
|
||||
{%- if pillar.nova.controller is defined %}
|
||||
nova_compute:
|
||||
engine: logstreamer
|
||||
log_directory: /var/log/nova
|
||||
file_match: nova-api\.log\.?(?P<Index>\d+)?(.gz)?
|
||||
priority: ["^Index"]
|
||||
decoder: multidecoder_openstack
|
||||
{%- endif %}
|
||||
decoder:
|
||||
openstack:
|
||||
engine: multidecoder
|
||||
subs: [ 'Payloadregex_OpenStackLog','Payloadregex_OpenStackLogTrace' ]
|
||||
cascade_strategy: first-wins
|
||||
log_sub_errors: "false"
|
||||
OpenStackLog:
|
||||
engine: payloadregex
|
||||
match_regex: ^(?P<timestamp>\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}:\d{2}.\d{3})\s(?P<pid>\d+)\s(?P<severity>\w+)\s(?P<programname>\D+)\s\[(?P<id>.+)\]\s(?P<payload>.*)
|
||||
timestamp_layout: "2015-10-06 11:34:37.243"
|
||||
OpenStackLogTrace:
|
||||
engine: payloadregex
|
||||
match_regex: ^(?P<timestamp>\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}:\d{2}.\d{3})\s(?P<pid>\d+)\s(?P<severity>TRACE)\s(?P<programname>.\S+)\s(?P<payload>.*)
|
||||
timestamp_layout: "2015-10-06 11:34:37.243"
|
||||
openstacksandbox:
|
||||
engine: sandbox
|
||||
file_name: "/etc/heka/plugins.d/decoders/openstack.lua"
|
||||
module_directory: "/etc/heka/plugins.d/common;/usr/share/heka/lua_modules"
|
||||
libvirtd:
|
||||
engine: multidecoder
|
||||
subs: [ 'Payloadregex_libvirtcommon', 'Payloadregex_libvirtaudit' ]
|
||||
cascade_strategy: first-wins
|
||||
log_sub_errors: "false"
|
||||
libvirtcommon:
|
||||
engine: payloadregex
|
||||
match_regex: ^(?P<timestamp>\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}:\d{2}.\d{3}\+\d{4}):\s+(?P<pid>\d+):\s+(?P<severity>\w+)\s+:\s+(?P<qemutrace>\w+:\d+)\s+:\s+(?P<payload>.*)
|
||||
timestamp_layout: "2015-09-01 15:56:14.675+0000"
|
||||
libvirtaudit:
|
||||
engine: payloadregex
|
||||
match_regex: ^(?P<timestamp>\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}:\d{2}.\d{3}\+\d{4}):\s+(?P<pid>\d+):\s+(?P<severity>\w+)\s+:\s+(?P<message>.+):\s(?P<payload>.+)
|
||||
timestamp_layout: "2015-09-01 15:56:14.675+0000"
|
||||
log_collector:
|
||||
decoder:
|
||||
nova:
|
||||
engine: sandbox
|
||||
module_file: /usr/share/lma_collector/decoders/openstack_log.lua
|
||||
module_dir: /usr/share/lma_collector/common;/usr/share/heka/lua_modules
|
||||
adjust_timezone: true
|
||||
{%- if pillar.nova.compute is defined %}
|
||||
libvirt:
|
||||
engine: sandbox
|
||||
module_file: /usr/share/lma_collector/decoders/libvirt_log.lua
|
||||
module_dir: /usr/share/lma_collector/common;/usr/share/heka/lua_modules
|
||||
{%- endif %}
|
||||
splitter:
|
||||
nova:
|
||||
engine: token
|
||||
delimiter: '\n'
|
||||
input:
|
||||
nova_log:
|
||||
engine: logstreamer
|
||||
log_directory: "/var/log"
|
||||
file_match: 'nova/(?P<Service>.+)\.log\.?(?P<Seq>\d*)$'
|
||||
differentiator: ['nova', '_', 'Service']
|
||||
priority: ["^Seq"]
|
||||
decoder: "nova_decoder"
|
||||
splitter: "nova_splitter"
|
||||
{%- if pillar.nova.compute is defined %}
|
||||
libvirt_log:
|
||||
engine: logstreamer
|
||||
log_directory: "/var/log"
|
||||
file_match: 'libvirt/libvirtd.log'
|
||||
differentiator: ['libvirt']
|
||||
decoder: "libvirt_decoder"
|
||||
splitter: "TokenSplitter"
|
||||
{%- endif %}
|
||||
metric_collector:
|
||||
trigger:
|
||||
nova_logs_error:
|
||||
description: 'Too many errors have been detected in Nova logs'
|
||||
severity: warning
|
||||
no_data_policy: okay
|
||||
rules:
|
||||
- metric: log_messages
|
||||
field:
|
||||
service: nova
|
||||
level: error
|
||||
relational_operator: '>'
|
||||
threshold: 0.1
|
||||
window: 70
|
||||
periods: 0
|
||||
function: max
|
||||
{%- if pillar.nova.controller is defined %}
|
||||
nova_api_local_endpoint:
|
||||
description: 'Nova API is locally down'
|
||||
severity: down
|
||||
rules:
|
||||
- metric: openstack_check_local_api
|
||||
field:
|
||||
service: nova-api
|
||||
relational_operator: '=='
|
||||
threshold: 0
|
||||
window: 60
|
||||
periods: 0
|
||||
function: last
|
||||
{%- endif %}
|
||||
alarm:
|
||||
{%- if pillar.nova.compute is defined %}
|
||||
nova_logs_compute:
|
||||
alerting: enabled
|
||||
triggers:
|
||||
- nova_logs_error
|
||||
dimension:
|
||||
service: nova-logs-compute
|
||||
{%- endif %}
|
||||
{%- if pillar.nova.controller is defined %}
|
||||
nova_logs:
|
||||
alerting: enabled
|
||||
triggers:
|
||||
- nova_logs_error
|
||||
dimension:
|
||||
service: nova-logs
|
||||
nova_api_endpoint:
|
||||
alerting: enabled
|
||||
triggers:
|
||||
- nova_api_local_endpoint
|
||||
dimension:
|
||||
service: nova-api-endpoint
|
||||
{%- endif %}
|
||||
remote_collector:
|
||||
trigger:
|
||||
{%- if pillar.nova.controller is defined %}
|
||||
nova_api_check_failed:
|
||||
description: 'Endpoint check for nova-api is failed'
|
||||
severity: down
|
||||
rules:
|
||||
- metric: openstack_check_api
|
||||
field:
|
||||
service: nova-api
|
||||
relational_operator: '=='
|
||||
threshold: 0
|
||||
window: 60
|
||||
periods: 0
|
||||
function: last
|
||||
{%- for nova_service in ('cert', 'consoleauth', 'compute', 'conductor', 'scheduler') %}
|
||||
nova_{{ nova_service }}_one_down:
|
||||
description: 'At least one Nova {{ nova_service }} is down'
|
||||
severity: warning
|
||||
rules:
|
||||
- metric: openstack_nova_services
|
||||
field:
|
||||
service: {{ nova_service }}
|
||||
state: down
|
||||
relational_operator: '>'
|
||||
threshold: 0
|
||||
window: 60
|
||||
periods: 0
|
||||
function: last
|
||||
nova_{{ nova_service }}_majority_down:
|
||||
description: 'Majority of Nova {{ nova_service }}s are down'
|
||||
severity: critical
|
||||
rules:
|
||||
- metric: openstack_nova_services_percent
|
||||
field:
|
||||
service: {{ nova_service }}
|
||||
state: up
|
||||
relational_operator: '<='
|
||||
threshold: 50
|
||||
window: 60
|
||||
periods: 0
|
||||
function: last
|
||||
nova_{{ nova_service }}_all_down:
|
||||
description: 'All Nova {{ nova_service }}s are down'
|
||||
severity: down
|
||||
rules:
|
||||
- metric: openstack_nova_services
|
||||
field:
|
||||
service: {{ nova_service }}
|
||||
state: up
|
||||
relational_operator: '=='
|
||||
threshold: 0
|
||||
window: 60
|
||||
periods: 0
|
||||
function: last
|
||||
{%- endfor %}
|
||||
nova_total_free_vcpu_warning:
|
||||
description: 'There is no VCPU available for new instances'
|
||||
severity: warning
|
||||
rules:
|
||||
- metric: openstack_nova_total_free_vcpus
|
||||
relational_operator: '=='
|
||||
threshold: 10
|
||||
window: 60
|
||||
periods: 0
|
||||
function: max
|
||||
nova_total_free_memory_warning:
|
||||
description: 'There is no memory available for new instances'
|
||||
severity: warning
|
||||
rules:
|
||||
- metric: openstack_nova_total_free_ram
|
||||
relational_operator: '=='
|
||||
threshold: 0
|
||||
window: 60
|
||||
periods: 0
|
||||
function: max
|
||||
{%- endif %}
|
||||
alarm:
|
||||
{%- if pillar.nova.controller is defined %}
|
||||
nova_api_check:
|
||||
alerting: true
|
||||
triggers:
|
||||
- nova_api_check_failed
|
||||
dimension:
|
||||
service: nova-api-check
|
||||
{%- for nova_service in ('cert', 'consoleauth', 'compute', 'conductor', 'scheduler') %}
|
||||
nova_{{ nova_service }}:
|
||||
alerting: true
|
||||
triggers:
|
||||
- nova_{{ nova_service }}_all_down
|
||||
- nova_{{ nova_service }}_majority_down
|
||||
- nova_{{ nova_service }}_one_down
|
||||
dimension:
|
||||
service: nova-{{ nova_service }}
|
||||
{%- endfor %}
|
||||
nova_free_vcpu:
|
||||
alerting: enabled
|
||||
triggers:
|
||||
- nova_total_free_vcpu_warning
|
||||
dimension:
|
||||
service: nova-free-vcpu
|
||||
nova_free_memory:
|
||||
alerting: enabled
|
||||
triggers:
|
||||
- nova_total_free_memory_warning
|
||||
dimension:
|
||||
service: nova-free-memory
|
||||
{%- endif %}
|
||||
aggregator:
|
||||
alarm_cluster:
|
||||
nova_logs_compute:
|
||||
policy: highest_severity
|
||||
group_by: hostname
|
||||
alerting: enabled
|
||||
match:
|
||||
service: nova-logs-compute
|
||||
members:
|
||||
- nova_logs_compute
|
||||
dimension:
|
||||
service: nova-compute
|
||||
nagios_host: 01-service-clusters
|
||||
nova_logs:
|
||||
policy: highest_severity
|
||||
group_by: hostname
|
||||
alerting: enabled
|
||||
match:
|
||||
service: nova-logs
|
||||
members:
|
||||
- nova_logs
|
||||
dimension:
|
||||
service: nova-control
|
||||
nagios_host: 01-service-clusters
|
||||
nova_api_endpoint:
|
||||
policy: availability_of_members
|
||||
group_by: hostname
|
||||
alerting: enabled
|
||||
match:
|
||||
service: nova-api-endpoint
|
||||
members:
|
||||
- nova_api_endpoint
|
||||
dimension:
|
||||
service: nova-control
|
||||
nagios_host: 01-service-clusters
|
||||
nova_api_check:
|
||||
policy: highest_severity
|
||||
alerting: enabled
|
||||
match:
|
||||
service: nova-api-check
|
||||
members:
|
||||
- nova_api_check
|
||||
dimension:
|
||||
service: nova-control
|
||||
nagios_host: 01-service-clusters
|
||||
{%- for nova_service in ('cert', 'consoleauth', 'compute', 'conductor', 'scheduler') %}
|
||||
nova_{{ nova_service }}:
|
||||
policy: highest_severity
|
||||
alerting: enabled
|
||||
match:
|
||||
service: nova-{{ nova_service }}
|
||||
members:
|
||||
- nova_{{ nova_service }}
|
||||
dimension:
|
||||
service: nova-control
|
||||
nagios_host: 01-service-clusters
|
||||
{%- endfor %}
|
||||
nova_free_vcpu:
|
||||
policy: highest_severity
|
||||
alerting: enabled
|
||||
match:
|
||||
service: nova-free-vcpu
|
||||
members:
|
||||
- nova_free_vcpu
|
||||
dimension:
|
||||
service: nova-compute
|
||||
nagios_host: 01-service-clusters
|
||||
nova_free_memory:
|
||||
policy: highest_severity
|
||||
alerting: enabled
|
||||
match:
|
||||
service: nova-free-memory
|
||||
members:
|
||||
- nova_free_memory
|
||||
dimension:
|
||||
service: nova-compute
|
||||
nagios_host: 01-service-clusters
|
||||
nova_control:
|
||||
policy: highest_severity
|
||||
alerting: enabled_with_notification
|
||||
match:
|
||||
service: nova-control
|
||||
members:
|
||||
- nova_logs
|
||||
- nova_api_endpoint
|
||||
- nova_api_check
|
||||
{%- for nova_service in ('cert', 'consoleauth', 'compute', 'conductor', 'scheduler') %}
|
||||
- nova_{{ nova_service }}
|
||||
{%- endfor %}
|
||||
dimension:
|
||||
cluster_name: nova-control
|
||||
nagios_host: 00-top-clusters
|
||||
nova_compute:
|
||||
policy: highest_severity
|
||||
alerting: enabled_with_notification
|
||||
match:
|
||||
service: nova-compute
|
||||
members:
|
||||
- nova_logs_compute
|
||||
- nova_free_vcpu
|
||||
- nova_free_memory
|
||||
dimension:
|
||||
cluster_name: nova-compute
|
||||
nagios_host: 00-top-clusters
|
||||
|
|
|
|||
Loading…
Reference in New Issue