On Mar 2, 2011, at 2:31 PM, William A. Rowe Jr. wrote: > On 3/2/2011 11:56 AM, Jim Jagielski wrote: >> A current problem with balancers, even without the newer "change >> lbprovider" and/or "add new worker" stuff is that if a worker >> goes away and comes back, it takes some time for the lbmethod >> to get back into sync... basically, even though we know we >> should reset or age the load specifics, we don't. > > Data collection has always had problems over graceful restarts, and > so long as the expiring proxy workers continue to update shm records > of the new generation, none of the methods will remain 'in sync'. >
the idea is that even with graceful restarts, there is no need for any of the workers (balancermembers) to even care about it... The only thing that changes things for the workers is when the list of workers changes or when the balancer lbmethod changes... if none of that happens during a graceful restart, there's no need, imo, to reset anything... > It certainly seems this needs a minimis solution prior to GA, one > which introduces the necessary API's. (I wasn't sure what you mean > by 'post-beta' - if that is after beta .11 and before beta .12, or > if you were meaning after 2.4.0). > After our 1st beta release... 2.3.11...
