openstack/kolla-ansible
Code Issues Proposed changes

Reduce container metrics cardinality

Browse Source 
Adds support for passing extra runtime options to cAdvisor.
By default new options disable exporting rarely useful metrics
and labels by cAdvisor. This helps reducing the load on Prometheus
and cAdvisor itself.

Change-Id: Id0144e8fa518e3236cb94ba2e3961fb455d36443
This commit is contained in:
Piotr Parczewski 2021-03-30 16:45:47 +02:00
parent 75a9ef4ad5
commit c6259158e3
4 changed files with 26 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ansible/roles/prometheus/defaults/main.yml
View File

@ -257,3 +257,5 @@ prometheus_openstack_exporter_disabled_dns: "{{ '--disable-service.dns' if not e
prometheus_openstack_exporter_disabled_object: "{{ '--disable-service.object-store' if not enable_swift | bool else '' }}"
prometheus_openstack_exporter_disabled_lb: "{{ '--disable-service.load-balancer --disable-metric=neutron-loadbalancers --disable-metric=neutron-loadbalancers_not_active' if not enable_octavia | bool else '' }}"
prometheus_openstack_exporter_disabled_items: "{{ [prometheus_openstack_exporter_disabled_volume, prometheus_openstack_exporter_disabled_dns, prometheus_openstack_exporter_disabled_object, prometheus_openstack_exporter_disabled_lb|trim]|join(' ')|trim }}"
prometheus_cadvisor_cmdline_extras: "--docker_only --store_container_labels=false --disable_metrics=percpu,referenced_memory,cpu_topology,resctrl,udp,advtcp,sched,hugetlb,memory_numa,tcp,process"

2
ansible/roles/prometheus/templates/prometheus-cadvisor.json.j2
View File

@ -1,5 +1,5 @@
{
"command": "/opt/cadvisor --port={{ prometheus_cadvisor_port }} --log_dir=/var/log/kolla/prometheus",
"command": "/opt/cadvisor --port={{ prometheus_cadvisor_port }} --log_dir=/var/log/kolla/prometheus {{ prometheus_cadvisor_cmdline_extras }}",
"config_files": [],
"permissions": [
{

4
etc/kolla/globals.yml
View File

@ -696,6 +696,10 @@
# List of extra parameters passed to prometheus. You can add as many to the list.
#prometheus_cmdline_extras:
# List of extra parameters passed to cAdvisor. By default system cgroups
# and container labels are not exposed to reduce time series cardinality.
#prometheus_cadvisor_cmdline_extras: "--docker_only --store_container_labels=false --disable_metrics=percpu,referenced_memory,cpu_topology,resctrl,udp,advtcp,sched,hugetlb,memory_numa,tcp,process"
# Example of setting endpoints for prometheus ceph mgr exporter.
# You should add all ceph mgr's in your external ceph deployment.
#prometheus_ceph_mgr_exporter_endpoints:

19
releasenotes/notes/reduce-cadvisor-cardinality-1213854b9fe0c828.yaml Normal file
View File

@ -0,0 +1,19 @@
---
features:
- |
Adds support for passing extra runtime options to cAdvisor via
``prometheus_cadvisor_cmdline_extras`` new variable. By default
system cgroups' metrics are disabled, plus container labels
don't get exposed to Prometheus. Expensive metrics that usually
should not be exported are also enforced to be disabled - consult
`<https://github.com/google/cadvisor/blob/master/docs/runtime_options.md#metrics>`_
for a list. These defaults create savings in resources usage by both
cAdvisor and Prometheus.
upgrade:
- |
cAdvisor has now reduced number of Prometheus metrics and labels
exported by default. This means that corresponding timeseries will
no longer be created. If existing setup relies on these, eg. for
visualisation or alerting, they could be explicitly enabled prior to
upgrading with the ``prometheus_cadvisor_cmdline_extras`` new variable.
Reference for the possible options: `<https://github.com/google/cadvisor/blob/master/docs/runtime_options.md#metrics>`_.