On Fri, Sep 18, 2015 at 12:40 PM, Willy Tarreau <w...@1wt.eu> wrote: > On Fri, Sep 18, 2015 at 10:51:13AM +0200, Baptiste wrote: >> Please find a patch in attachment which fix the issue so you can run >> temporarly the latest code in production. >> That said, I need to discuss with Willy to ensure this is the best way >> to do it, since this patch changes the design we did. >> Anyway, I've updated my server-state code to match the new behavior, >> soe everyone should be satisfied. > > I totally agree with your patch Baptiste and I've merged it. It > does what we initially designed, I was the one who got confused > by it during the first review (in part due to too much time > between design and review). Sorry for that. Keep in mind you'll > have to be careful in your reload patch to properly consider > FMAINT and CMAINT separately. > > Thanks, > Willy >
Ok, thanks Willy. I'm using it carefully. I set up a matrix of all combination from old config, state at run time in old process and state in new config and came to the conclusion that: - in case of a configuration change, we apply the setting from the new configuration, regardless of old running state (I'll check if the documentation also relates this behavior) - if no configuration change, we apply old running state only if old running state is different from new configuration state Actually, the same apply to the server's iweight parameter to decide about DRAIN state and computation of eweight. Looking forward to send you the patches regarding server states in a few minutes. Baptiste