Performance issue

[crysis-ps]

Hello,
since v 1.2+ version I see the performance drop with loading metrics page.
I have the proxmox cluster with multiple nodes and ~500 VMs.
The metrics page has near 7000 lines.

The 1.1.2 version of pve exporter - the metrics page loads for ~400ms
The 1.2 or v2.0.3 version - the metrics page loads for ~7s

[znerol]

There is only one code change between 1.1.2 and 1.2.0: PR #22

The change indeed has potential to slow down scraping on big clusters, since it loops over every guest and fetches the config for each of them which will result in many API calls.

You could remove this line near the end of collector.py:

    registry.register(ClusterNodeConfigCollector(pve))

Then check whether the problem goes away. The only purpose of this collector (as of now) is to extract the onboot flag.

[crysis-ps]

thank you.
yes, if this line is commented out - the metrics page loads for 300-400ms.

[znerol]

Ok, thanks for checking. So the best thing would be to make this configurable.

[znerol]

Published version 2.1.0 to dockerhub and pipy which includes #62. It is now possible to run pve_exporter with the --no-collector.config flag which will deactivate the slow collector.