[prometheus-to-sd] Support for SUMMARY metric types about k8s-stackdriver HOT 10 OPEN

Any update on this?? From a maintainer of the repo?

from k8s-stackdriver.

1. Quantile metric data is disregarded, mostly because it doesn't make a whole lot of sense to aggregate over the quantile metric data. Aggregating across this data will give you averages. You will lose quantile information.

There are lots of use cases where engineers want to monitor quintile measurements without aggregating them. Ex: what is the p75 latency measurement from our master node. Use cases like this are why summary metrics can make sense in prometheus.

Omitting this feature from prom-to-sd has repeatedly made it difficult for my team to use stackdriver as a drop in replacement for prom, given the wealth of third party prometheus extensions that output summary metrics (ex: kafka prom exporter, jenkins prom exporter plugin).

from k8s-stackdriver.

cc @GoogleCloudPlatform/logmon

from k8s-stackdriver.

Is this not feasible or do you need help to implement this?

from k8s-stackdriver.

What do you mean?

from k8s-stackdriver.

I also want support for SUMMARY and I am hoping that a maintainer is listening to this issue.

from k8s-stackdriver.

Running into this now myself. Curious if this project is still being maintained.

from k8s-stackdriver.

@logicalhan Want to paste summary (e.g. from of #221) and close issue?

from k8s-stackdriver.

@erictune, sure thing.

#221 adds partial summary metric importing ability for prom-to-sd. There are some important caveats, however:

1. Quantile metric data is disregarded, mostly because it doesn't make a whole lot of sense to aggregate over the quantile metric data. Aggregating across this data will give you averages. You will lose quantile information.
2. If you want to import summary metric information, you must first ensure that the metric you are summing does not have negative values (this is despite the fact that prometheus documentation expressly allows for this possibility). Unfortunately, this must be the case due to data type impedance mismatching between prometheus and stackdriver. So if you want to use this feature, you should first ensure your metric is measuring positive values. With things like latency measurements, this shouldn't be a problem.
3. If you are whitelisting metrics, then you must whitelist the summary metric name + '_sum' as well as the summary metric name + "_count". This is due to the fact that this metric type maps to multiple stackdriver metrics.

/close

from k8s-stackdriver.

Hey Everyone, I've gotten the same error "Error while processing metric jvm_gc_collection_seconds: Metric type SUMMARY of family jvm_gc_collection_seconds not supported".
Is there any update about it?

from k8s-stackdriver.