slowness when rendering a graph with a lot of targets

[Dieterbe]

if you didn't filter down a lot, it can happen that graphs contain hundreds of metrics.

• due to the grouping, there's basically no limit to how many targets appear on one graph. (upto limit, if all targets go on one graph)
• 200 targets on 1 graph (not aggregated) takes about 30s to render for me. it could be way worse, esp. with aggregation.

• the same set of targets being rendered on one graph, vs on multiple graphs, is a big difference. I think the problem is, when graphite gets a big request (with multiple targets), it handles them all in sequence instead of in parrallel (need to confirm this). timeserieswidget used to have code that splits up a request in multiple subrequest, i could reintroduce that, or fix graphite to use parrallellisation (unlikely), than in GE/timeserieswidget we just need to do something to make the graph's height match the legend.

  or in GE limit the amount of targets per graph. (i.e. max targets per graph 50 or so) you can see the difference usually with something like catchall statsd vs catchall statsd group by n1,n2,n3,n4

PRO of splitting up to multiple graphs:

• you can see the entire course of a metric without scrolling
• so many metrics on one graph can be overkill anyway (though not always, and the interactive popups and future highlighting support can alleviate this)

CON of splitting up to multiple graphs:

• can be hard to relate targets on different graphs
• doesn't really help for other tools.

[zehome]

I think graphs with more than 10 targets are very often useless and un-exploitable. (Because you can't differenciate colors visually, then the legend takes the most of your screen space).

Table export would be more usefull with that kind of graph I think.

I would recommand the user to aggregate in that kind of cases.

[Dieterbe]

Table export would be more usefull with that kind of graph I think.

you mean, show a html table with all the data in it?

the graph can be made a little clearer, an upcoming timeserieswidget/flot feature is that you hoover over legend elements and it highlights the corresponding trendline.

[Dieterbe]

another interesting concept could be, if the legend would grow larger than 1 window height, show some kind of systemic overview of the variable tags that would otherwise show up for all targets in the legend. i.e. like a facet display saying which tag keys occur in all targets, and across how many targets, and maybe for each how many different values, and maybe show a few of the most common ones, or something.

we could also show the most deviant ones and hide the rest.

[nikicat]

+1