I'm trying to create an alert using promql/prometheus but having trouble generating the proper time series. In K8s, my objective is to display any code/version mismatch for a particular app sitting in multiple clusters. By using count, any app with a value greater than 1 would tell me that there are > 1 version deployed. Currently my query will generate the following:
count(kube_deployment_labels{label_app=~".*"}) by (label_version, label_app)
| label_app | label_version | Value #A |
-------------------------------------------
| app_1 | 0.0.111 | 2 |
| app_1 | 0.0.222 | 1 |
| app_2 | 0.0.111 | 2 |
| app_2 | 0.0.222 | 1 |
| app_3 | 0.0.111 | 3 |
The values on the 4th column represent the # of said code (label_version) deployed in each cluster; for example, app_1 & app_2 has version 0.0.111 deployed in two clusters & 0.0.222 deployed in one cluster, but app_3 has the same version deployed in all three clusters.
My end goal is to only count distinct label_version and have the time series populate in this way:
| label_app | Value #A |
-------------------------
| app_1 | 2 |
| app_2 | 2 |
| app_3 | 1 |
Executing (count(group by(label_version)kube_deployment_labels{label_app=~".*"})) gives me the correct Value but I'd like to list out all the apps associated to the Value as well. I've tried a variety of grouping but was unsuccessful.
Any help would be appreciated! Thanks in advance.
You can apply one more count by over your previous result.
count(
count(
kube_deployment_labels{label_app=~".*"}
) by (label_version, label_app)
) by (label_app)
will return number of label_versions associated with each label_app.
Somewhat similar query can be seen in this demo.