On Fri, Apr 17, 2015 at 9:46 PM, Dennis Jacobfeuerborn < denni...@conversis.de> wrote:
> Yes, you add a peer section: > > peers haproxy-peers > peer haproxy1 127.0.0.1:1024 > > (haproxy1 must the the hostname of the local system) > > and then you add a reference to that peer section in you stick table > definition for example: > > stick-table type ip size 100k expire 20m peers haproxy-peers > > On a reload the new process will then sync the stick table from the old > process. > > If you want to do SSL offloading (and thus need nbproc>1) the best > workaround I found is to setup two haproxy instances (systemd makes that > pretty easy) one for the SSL offloading (nbproc>1) which then forwards > the traffic using abstract namespace socket to the other instance > (nbproc=1) that does the actual load-balancing. > > That way you can reload the balancing instance keeping the stick tables > but also utilize multiple cpu cores for SSL offloading. > > Regards, > Dennis On recent update it is given that now nbproc can be used along with peers. Another improvement consists in a relax of the restriction between peers > and nbproc. Now it is possible to use peers provided that the whole > section is only used by tables belonging to the same process. This makes > it easier to run SSL offloading in multiple processes now. Ah, and it's > also possible to disable a peers section without having to comment all > places where it's referenced! Can anyone provide a sample/example config for same? Jai
