suyash
commented
6 years ago we use Speed to report about a thousand of histograms, this creates about a thousand of min,max,mean,std_dev instances. I am under impression that it should be possible to create those instances just once and then reuse their IDs.
To create a Histogram, you'll need to specify a name, and the name needs to be unique for each metric since the metric name generation hash isn't randomized and depends on the string, so same strings will hash to the same value. For each metric, the instance names are "min", "max" ... and the ID generation for them uses the same mechanism. (https://github.com/performancecopilot/speed/blob/master/instance.go#L29), so IDs would be shared (because all of them will have the same name).
The problem here is that the PCPHistogram constructor does not accept an InstanceDomain as a constructor. For each histogram, a new InstanceDomain is created, and hence the same instances are written twice, albeit for different instance domains, see the screenshot of an instance block with two histograms generated using mmvdump.
The ideal scenario to share instances would be to share instance domains, across different metrics, which the raw NewPCPInstanceMetric constructor allows to do, but custom metrics don't.
In my opinion, using a single global InstanceDomain object for all histograms would be an ideal fix, since for histograms we are not allowing customization anyways and the instances available are shipped with the library itself.
@natoscott thoughts?
natoscott
lzap
Hey,
we use Speed to report about a thousand of histograms, this creates about a thousand of min,max,mean,std_dev instances. I am under impression that it should be possible to create those instances just once and then reuse their IDs.
If I am wrong just close the RFE ticket, sorry for the noise :-)