Armada exposes metric data, for consumption by Prometheus.
Metric data can be exported via:
- API: Prometheus exporter in the /metrics endpoint. The Armada chart includes the appropriate Prometheus scrape configurations for this endpoint.
- CLI: –metrics-output=<path> of apply command. The node exporter text file collector can then be used to export the produced text files to Prometheus.
Metric names are as follows:
armada_ + <action> + _ + <metric>
The below tree of <action>s are measured. Supported prometheus labels are noted. Labels are inherited by sub-actions except as noted.
- description: apply a manifest
- labels: manifest
- description: fully handle a chart (see below sub-actions)
- action (install|upgrade|noop) (not included in sub-actions)
- description: delete a chart (e.g. due to FAILED status)
- labels: chart
- failure_total: total failed attempts
- attempt_total: total attempts
- attempt_inprogress: total attempts in progress
- duration_seconds: duration of each attempt
The chart_handle and chart_test actions additionally include the following metrics:
- timeout_duration_seconds: configured chart timeout duration in seconds
- timeout_usage_ratio: = duration_seconds / timeout_duration_seconds
These can help identify charts whose timeouts may need to be changed to avoid potential failures or to acheive faster failures.
The chart_handle action additionally includes the following metric:
- concurrency_count: count of charts being handled concurrently
This can help identify opportunities for greater chart concurrency.