Le 11/09/2023 à 15:12, Cedric Paillet a écrit :
This adds a new 'FULL' status to the Prometheus metric 'srv_state'. It helps
identify servers that have exceeded their maxconn limit and cannot accept new
connections.
Rename server_has_room to !server_is_full to matches what's used at a few places in the
doc in association with servers or backends being "full".
Thanks Cedric,
I have no real opinion on the way the FULL state is detected, I will let Willy
react :)
However, to be merged, the patch should be split, to first rename the function
and then change the metric into the Prometheus exporter. This will simplify
reviews on bugs and ease eventual reverts.
But I'm a bit annoyed to hijack the servers check status this way. First
because, this change the promex metrics and not everyone is interested by this
info and may want to keep the "UP" status. Then I'm also annoyed to have a
difference between the stats applet and the promex one. The first one will still
report a server as "UP" while the second one will be able to report it as "FULL".
I guess it could be a new server metric. A boolean to report it as full or not.
And an aggregate metric can also be added for the backends. But we should be
careful about performances. It could be good to think to merge the processing of
aggregated metrics to not loop on the server list several times. Finally, if a
new Prometheus metric is added for servers, it could be good to see if it
could/should be added on the stats page too.
--
Christopher Faulet
