Splitting metrics "export" out of the main controller
#7.373 geöffnet am 16.11.2023
Repository-Metriken
- Stars
- (9.013 Sterne)
- PR-Merge-Metriken
- (Durchschn. Merge 9T 3h) (137 gemergte PRs in 30 T)
Beschreibung
The proposal here is to extract the "metrics" expose part from the main controller, and have a dedicated one for this only. This is a bit similar to what was proposed for CloudEvents with TEP-0137 and inline with https://github.com/tektoncd/community/issues/1076.
Currently, Tekton Pipelines' metrics can be configured by changing config-observability. However, the options that we have are very limited. You can configure to use
metrics.taskrun.level: task, and then you can't aggregate further than that. And if you usemetrics.taskrun.level: taskrun, that's not recommended since can lead to unbounded cardinality.
- Configure what metrics are exposed and possibly define highly specific metrics.
- Reducing "metrics export" impact on the main controller, from performance perspective as well as "code". The main controller's job is to manage Tekton objects, the metrics controller to expose metrics ; the code in those controller remains leaner, and their scalability is not compromised.
There is already a pull-request in tektoncd/experimental on this by @RafaeLeal.
/cc @gabemontero @tektoncd/core-maintainers
Related "metrics" issues: