On Mon, Apr 27, 2015 at 11:01PM, Ognen Duzlevski wrote: > Nikita, (thanks for the addition to an already nice answer by Cos! See my > question below; > > On Mon, Apr 27, 2015 at 6:46 PM, Nikita Ivanov <[email protected]> wrote: > > > Just to add to excellent Cos's response: > > - The code base for Apache Ignite has been in production usage since 2007. > > It's the only in-memory system that I'm aware of that can boast over 2000 > > nodes in a single mission critical installation working in a fully > > transactional topology. > > > > Perhaps you can then answer my question from a different thread: what do > people do when they want to add additional proprietary classes to be > cached to a live ignite topology? Surely they don't stop all 2,000 nodes, > copy a new jar around and restart all 2,000 nodes? :-)
There's so called 'Peer class loading' as per http://apacheignite.readme.io/v1.0/docs/zero-deployment but looks like you're looking for somewhat different usecase. I wonder if you should be able to distribute the classes to the nodes and then do the rolling restart. Just thinking out load - not 100% sure. Perhaps some of the old-timer can chime in Cos
