Open jtpio opened 4 years ago
jtpio
commented
4 years ago For now the code in the API handler:
Duplicates some of the logic of the Prometheus handler, which relies on a PSUtilMetricsLoader:
It would be nice if both could share the same approach when it comes to retrieving metrics from psutil.
jtpio
commented
4 years ago It would be interesting to have a look at the glances API: https://github.com/nicolargo/glances/wiki/The-Glances-RESTFULL-JSON-API
Glances relies on psutil to provide aggregated metrics from the host machine.
There might be some ideas that could be applied to nbresuse if we start expanding the API surface.
yuvipanda
commented
4 years ago https://github.com/yuvipanda/nbresuse/pull/68 provides /api/metrics/v1, without removing /metrics. We can probably tweak it more in a /api/metrics/v2
Gsbreddy
commented
4 years ago We can add the per kernel metrics https://github.com/yuvipanda/nbresuse/pull/41 since this has been a long ask for lots of people. I myself am using this in my org but was overriding /metrics to enable this capability. Now that the soft deprecate went it, I think we can integrate https://github.com/yuvipanda/nbresuse/pull/41 onto /api/metrics/v2. Also CPU per kernel as well is a big addition. What do you guys think?
@yuvipanda @jtpio
https://github.com/yuvipanda/nbresuse/pull/45 restored the
/metricsendpoint, so that JupyterLab (and other frontends) can display the metrics:The reason to move away from
/metricsis to not shadow the default Prometheus endpoint (see https://github.com/yuvipanda/nbresuse/pull/22 for more infos).However this is a breaking change. We should carefully plan the rollout to avoid confusing users with an older version of JupyterLab. For the classic notebook this is not really an issue as the notebook extension is bundled with the package on PyPI.
cc @Gsbreddy who has already started working on this in side PRs.