Node panels #136
Node panels #136
Conversation
| @@ -30,10 +30,14 @@ format: deps | |||
|
|
|||
| build: deps $(LIBRARY_PATH) $(outputs) | |||
|
|
|||
| clean: | |||
| clean-all: | |||
There was a problem hiding this comment.
New make task so binaries can still be deleted, but are not deleted every time.
| + options.legend.withSortDesc(true) | ||
| + options.legend.withPlacement('bottom'), | ||
|
|
||
| genericLegendCounter(title, unit, targets, gridPos): |
There was a problem hiding this comment.
New panel type with different legend fields, more relevant for counters and memory.
| @@ -3,8 +3,7 @@ local var = g.dashboard.variable; | |||
|
|
|||
| { | |||
| datasource: | |||
| var.datasource.new('datasource', 'prometheus') | |||
| + var.datasource.withRegex('/^Cluster Prometheus$/'), | |||
There was a problem hiding this comment.
Using a dashboard against a prometheus outside of openshift, I update this variable after importing. Instead, just deleting for good.
There was a problem hiding this comment.
Confirmed on ROSA cluster that the dashboard variable auto-populates to 'Cluster Prometheus'
| panels.timeSeries.genericLegend('ovs-worker CPU Usage', 'percent', queries.OVSCPU.query('$_worker_node'), { x: 0, y: 21, w: 12, h: 8 }), | ||
| panels.timeSeries.genericLegend('ovs-worker Memory Usage', 'bytes', queries.OVSMemory.query('$_worker_node'), { x: 12, y: 21, w: 12, h: 8 }), | ||
| panels.timeSeries.generic('99% Pod Annotation Latency', 's', queries.ovnAnnotationLatency.query(), { x: 0, y: 1, w: 24, h: 12 }), | ||
| panels.timeSeries.generic('99% CNI Request ADD Latency', 's', queries.ovnCNIAdd.query(), { x: 0, y: 13, w: 12, h: 8 }), |
There was a problem hiding this comment.
These y values were causing these three panels to pop out of the row.
Also, the metrics seem far less frequently used than CPU and memory usage, so I also moved them to the bottom so relevant panels stay at the top.
|
lgtm |
Add panels that show cluster view Signed-off-by: Andrew Collins <[email protected]>
Signed-off-by: Andrew Collins <[email protected]> panels and legends updates Signed-off-by: Andrew Collins <[email protected]>
afcollins
force-pushed
the
node-panels
branch
from
3996873 to
ffb3daf
Compare
| query(): | ||
| generateTimeSeriesQuery('ovnkube_master_leader', '{{pod}}'), |
There was a problem hiding this comment.
metric doesn't exist. Replacement is only _leader that is unique, as ovnkube_controller_leader is 0 for all pods.
| }, | ||
|
|
||
| ovnNorthd: { | ||
| query(): | ||
| generateTimeSeriesQuery('ovn_northd_status', '{{pod}}'), | ||
| }, | ||
|
|
||
| ovnNbdbLeader: { | ||
| query(): | ||
| generateTimeSeriesQuery('ovn_db_cluster_server_role{server_role="leader",db_name="OVN_Northbound"}', '{{pod}}'), |
There was a problem hiding this comment.
Removing both as neither metric exists any longer.
| numOnvController: { | ||
| query(): | ||
| generateTimeSeriesQuery('count(ovn_controller_monitor_all) by (namespace)', ''), | ||
| }, | ||
|
|
||
| ovnKubeControlPlaneCPU: { | ||
| query(): | ||
| generateTimeSeriesQuery('irate(container_cpu_usage_seconds_total{pod=~"(ovnkube-master|ovnkube-control-plane).+",namespace="openshift-ovn-kubernetes",container!~"POD|"}[2m])*100','{{container}}-{{pod}}-{{node}}'), | ||
| generateTimeSeriesQuery('sum( irate(container_cpu_usage_seconds_total{pod=~"(ovnkube-master|ovnkube-control-plane).+",namespace="openshift-ovn-kubernetes",container!~"POD|"}[2m])*100 ) by (pod, node)', '{{pod}} - {{node}}'), |
There was a problem hiding this comment.
label formatting to match ocp-performance dashboard
Type of change
Description
Adding a new section to the OCP Performance dashboard that help me get an quick overview of the cluster for any nodes or issues to dive into.
Reordered OVN Dashboard for relevance, but also because the panels pop out of the row.
Also removed old metrics.
Makefile changes to allow generated dashboard cleanup without deleting and redownloading binaries.
variables changes to be more flexible with prometheus that may not be running inside an openshift cluster.
Checklist before requesting a review
Testing
I run
makeand import the generated dashboards in a grafana running locally against a locally running prometheus.