Vladimir Smirnov

https://github.com/go-graphite/carbonapi/blob/fab6ee9ed2a8bd5e899f87af148f07a68d3a37db/expr/functions/asPercent/function.go#L85-L93 That would be the place where error would be triggered. Unfortunately error doesn't tell what exactly is the problem here, but I suspect that amount of replies in timeShift...

What's your go-carbon.conf file? Also, if you'll have a look at the systemd example service here: https://github.com/go-graphite/go-carbon/blob/master/deploy/go-carbon.service#L7-L12 You might notice that at least with default config, type should be "forking"...

@gkramer problem with no index at all approach is that you rely on OS to be able to give you list of files fast. However, depending on your filesystem and...

I think key to understanding of what is the bottleneck in your case is gathering metrics and some debug information from your server. For example if you suspect that it's...

@bom-d-van I think important part here is that it's `context canceled` - which means either timeout set for too small value or that upstream gave up immediately (AFAIR go-carbon in...

@bom-d-van depends on what you do, you can pass deadline from upstream and reuse it for specific request, for example. So it can be somewhat implied. As well as in...

@bom-d-van if they are stuck - it can be that runtime will be 0.0 from go-carbon's point of view, but it will be `timeout` setting from the carbonapi's standpoint. However...

I guess as the title have "WIP" in the name it is considered to be work-in-progress (that means I won't review it in-depth). One note, however. Current PR expect backend...

Alternatively it might be worth do a find request and post-process the response, if of course find message have enough data.

> structure an acceptable or comparable way to do this? I guess in that case it is ok to use the same structure as for find. > By default the...