Search before asking
Description
Writes to a tablet are rejected once its version count passes max_tablet_version_num, and it fails with TOO_MANY_VERSION (E235). The check reads a counter called _approximate_num_rowsets.
The only related metric, tablet_version_num_distribution, reads a different counter that updates less frequently and skips tablets that are currently being written to. This hides potential issues until there is a failure.
There's no Prometheus metric for _approximate_num_rowsets. The only way to read it is via admin HTTP GET to /api/hotspot/tablet?metrics=num_rowsets.
Solution
- Add a metric fed from _approximate_num_rowsets using the same pattern as the existing histogram in doris_metrics.{h,cpp}.
- Report it for all tablets, including ones under active load
Are you willing to submit PR?
Code of Conduct
Search before asking
Description
Writes to a tablet are rejected once its version count passes
max_tablet_version_num, and it fails withTOO_MANY_VERSION(E235). The check reads a counter called _approximate_num_rowsets.The only related metric,
tablet_version_num_distribution, reads a different counter that updates less frequently and skips tablets that are currently being written to. This hides potential issues until there is a failure.There's no Prometheus metric for
_approximate_num_rowsets. The only way to read it is via admin HTTP GET to /api/hotspot/tablet?metrics=num_rowsets.Solution
Are you willing to submit PR?
Code of Conduct