Performance Issues related with high request call rates to T _host_get_clock_service

[sky-nunomarcos]

I have been troubleshooting a node implementation that is currently using your lib to expose prometheus metrics and using clinicJS tool notices that almost 80% of total cpu usage time is spend in T _host_get_clock_service call. Already updated lib version to 14.0.1, but issues continues to be present.

Clinic Flamechart is this one:

Looking at your source code it can be related with Date.now(); requests performed in lib/timeWindowQuantiles.js or lib/metrics/processStartTime.js.

Do you have any suggestion on how to troubleshoot these issues and try to mitigate them?

Many Thanks for your help

[zbjornson]

Thanks for reporting this. I'm hoping to look into this this month, but please don't let that stop anyone else from taking a look if they have time before I do.

[timofriedlberlin]

Any update on this? Due to the high resource usage of this library we cannot use it in production.

[jdmarshall]

Other than the already mentioned issues and improvements coming in v0.16, my other concern is that somehow you're getting multiple copies of prom-client loaded and what you're actually seeing here is all of them gathering their defaultMetrics, rather than the one instance attached to the scraping logic.