From the amazon develop guide:
Shards are ephemeral: They are created and deleted automatically, as needed. Any shard can also split into multiple new shards; this also occurs automatically. (Note that it is also possible for a parent shard to have just one child shard.) A shard might split in response to high levels of write activity on its parent table, so that applications can process records from multiple shards in parallel.
In my application, I don't really care which shard data comes from, I just want to process all changes. When you set up your stream processors, though, you have to subscribe to a shard or set of shards. How can you be sure you're getting all the changes if shards can be removed/split up/added at any time?
Should a client poll the list of shards for changes?
