[Bug]: Service checks with lots of graphs are slowing down browser performance

[K0nne]

Please try to fill out as much of the information below as you can. Thank you!

• Yes, I've searched similar issues on GitHub and didn't find any.

Which version contains the bug?

0.2.2

Describe the bug

Hello!

We have started to test the module. It works fine for checks with only a few metrics.
When we opened the icingadb check, the browser started to lag.

Upon expanding the graphs the browser became almost unusable. It feels like even in collapsed state all graphs are being rendered in the background.

How to recreate the bug?

1. enable the modules "perfdatagraphs" and "perfdatagraphs-influxdbv2"
2. open the check "icingadb"
3. expand the graphs
4. scroll a bit

[martialblog]

Yeah, we know https://github.com/NETWAYS/icingaweb2-module-perfdatagraphs?tab=readme-ov-file#known-issues

At the moment I'm evaluating how to best handle rendering when a service has a lot of timeseries. I have ideas, but nothing concrete yet.

[carraroj]

I know we had the same problem with the Icinga check, but it works for me in version 0.2.2. The RAM increases slightly when scrolling, but then stabilizes again. I have now also tested the icingadb check, and it behaves in exactly the same way and I use the browser Mozilla Firefox.

[martialblog]

Currently the only viable idea I have is to overhaul how graphs are displayed. Changing it from all available metrics to a single metric and a dropdown list to select others.

If anyone has ideas, let me know.

[RincewindsHat]

hm, if not configured differently, pick one/two/three series and display them + "Expand" button for the rest?

[martialblog]

hm, if not configured differently, pick one/two/three series and display them + "Expand" button for the rest?

That is basically what we do now. Maybe there are some lazy loading/rendering tricks we can use to load/unload a charts when there are not needed.

[martialblog]

I tried something here, the plots that are not on-screen are hidden so the browser won't render them.

https://github.com/NETWAYS/icingaweb2-module-perfdatagraphs/tree/lazyloading

if someone could test that to see if it helps with the performance that would be great. @K0nne @carraroj

[K0nne]

Hello!
We have tested a bit. As long as it is collapsed, the usability has improved. When everything is shown, the cpu usage goes up. On a notebook it's ~4-10% and in a vm ~40-80%.

[martialblog]

Thanks for the feedback.

When everything is shown, the cpu usage goes up.

Some usage is to be expected since we're rendering things, and with the Icinga Web autorefresh we have to re-render things unfortunately.

Alternative solutions have their disadvantages as well.

Dropdown menu to render only one metric

• Single metric, single plot. Should have little performance impact on the page, but is obviously limited when you want to compare/investigate multiple metrics

Render only explicitly configured metrics

• Only metrics that are configured explicitly are rendered. Will keep the page from loading too many plots, but you would have to know which metrics you want to see.

[martialblog]

I think what we can do is render only three (magic number) charts by default unless include/exclude is set, then these are rendered.

When there are more charts then a "Show All" link is added, which will load a new page will all charts for the selected host/service. This page can then also be used in a dashboard, which is an additional benefit.