Hi, i have the need to process quite a lot of data in form of an event stream. Each event is connected to an user and for most of my operations have to operate on the entire set of events for a given users. So every time a new event arrives I have to load all events from some data storage, do my processing, store the event in the storage and send the results somewhere. The problem is that reading the old events from the storage is quite expensive, in particular if done over a network. Therefore it would make sense to use the machine the bolt is running on for storage as well. For install a locally installed database system like leveldb. This way I can exploit locality and reduce network bandwidth.
I have seen that Trident has some support for state, but I have the impression it's just an abstraction to a remote storage. If I would implement my own local storage, would Storm keep the bolt on the same machine during run-time, or could it somehow be stopped and redeployed on another machine? Cheers, Klaus
