On 01/04/2010 12:34 PM, Manik Surtani wrote: Hmmmm, generally yeah but in this case I suspect you don't want actually. Example: Let's say you have 3 caches configured with SSCL and pushWhenCoordinator=true, and they're pointing to the same shared NFS cache loader whose contents are: /e/f [akey, a value] Let's say the coordinator dies and between that and the new coordinator taking over, a put is done (fqn=/e/i [bkey, bvalue]) in the node that is supposed to become coordinator. The new coordinator is not yet active, so /e/i won't be pushed to the cache loader. When the new coordinator finally takes over and tries to get children's of /e, it checks the cache loader and it says that these are: [f], and hence it doesn't push /e/i to the cache loader. The counterpoint here is that if eviction has been set up, and during the coordinator change, /e/f had been evicted, it could maybe mean that either, /e/f is not in memory any more, or /e/f is there but it's empty. This could be OK cos I only push any changes of data that is in memory, so if /e/f is not there, or is empty, I do not push that to the cache loader.