On Feb 5, 2013, at 11:15 AM, Łukasz Mierzwa <[email protected]> wrote:

> It's calculated as:
> 
> $(number of seconds all requests took in last minute) / $(carbon interval)
> 
> The "number of seconds all requests took" is taken from spent_time variable 
> of uwsgi_server struct and this variable is updated when each request is 
> completed (request end time - request start time). It's not going to be 100% 
> accurate but it good enough. Also it's possible that this does not work 
> properly with gevent.
> But response times are related not only to backend load but also to DB or 
> other remote services so there is nothing fundamentally unusual about those 
> graphs.

Thanks. It makes sense now.

> Yous should have per worker metrics, There is --carbon-no-workers switch but 
> AFAIR it is disabled by default.

I do have per-worker metrics, I just can't find one that tells me how many 
concurrent requests are being served at a given time. Did I miss it? I only 
found one metric named "requests" but this seems just a running counter.

André

_______________________________________________
uWSGI mailing list
[email protected]
http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi

Reply via email to