Hi,

Recently I started to use carbon plugin (uWSGI 1.4.4) to graph some
metrics from uWSGI. I have experienced a few problems on the way to
make it work the way I want it..

1. Hostnames with dots
In our environment (CentOS/RedHat) hostnames are FQDN as recommended by
RedHat/CentOS(*) This does not play nicely with carbon metric names as
dot is a namespace separator.

2. Different namespace
Currently carbon plugin does not allow me to put my metrics in
arbitrary location (besides I found default metric path a bit too long
for my needs).

As a workaround to above I patched uwsgi - I hijacked "carbon-id"
parameter which now I use to specify base path of all metrics eg.

carbon-id = sys.hostname_subdomain_com.uwsgi
will generate
sys.hostname_subdomain_com.uwsgi.requests
sys.hostname_subdomain_com.uwsgi.worker1.requests
...

Maybe extra option(?) to allow to override base path for metrics ?

3. Missing metrics
When I look at stats server output I see quite some metrics there which
are not present in carbon plugin and I find them useful eg.

respawn_count
exceptions

Probably other people might find some other metrics important for them. 

4. Busy / idle workers
Currently we only have busyness which is not exactly what busy/idle
is eg. yesterday one of our instances locked up, so all workers were
busy and uwsgi couldn't dispatch any requests to workers (no request
was served by that time period). During that time busyness was 0 (most
likely due to the fact that no request was served during that period).

5. Empty metrics
Some of the metrics are generated even though their value is always 0
(eg. rss/vsz) as those only get updated in certain circumstances
(enabled resource limiting base don memory footprint or memory usage
reporting). In such circumstances I don't think it makes sense to
generate them at all.

6. Calculation for cheaped / active workers
I see that in some cases we calculate some total metrics (regading
whole instance) for all (even cheaped) workers and some only for active
workers (process is running).
I would say that eg. harakiri metric should be counted over all
rather than active only workers as it's a counter.  

Any plans on addressing the above ? Let me know if I can be of any help
here..
Regards,

Marcin

*)
http://www.centos.org/docs/5/html/5.2/Deployment_Guide/s2-sysconfig-network.html
https://access.redhat.com/knowledge/docs/en-US/Red_Hat_Enterprise_Linux/5/html/Deployment_Guide/ch-sysconfig.html#s2-sysconfig-network
_______________________________________________
uWSGI mailing list
[email protected]
http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi

Reply via email to