Sergey, I see your point. We can refactor continuous routines to use discovery custom messages (same as for dynamic caches) and make all servers know about routines started so far (again, same as for dynamic caches). This way, whenever new client or server joins, other servers in topology can supply all the info without querying clients. This will make the topology more stable and almost fully independent from client connection stability and speed.
Let me know if you have questions. --Yakov 2015-04-29 12:41 GMT+03:00 Sergey Evdokimov <sevdoki...@gridgain.com>: > We have a problem in GridContinuousProcessor, when new node is joined to > topology it collects registered routines in collectDiscoveryData() on each > node. But data collected in collectDiscoveryData() on client will no be > send to another nodes, because client node is not in the ring. So routines > created on client node will not be registered on new node, see > TcpClientDiscoverySelfTest#testDataExchangeFromClient() test. > > Have any idea how to fix it? >
