Lack of support for partial reads is one reason to keep collections fairly small; another is that there is a built-in size limit for collections that Cassandra can store. In particular, any given collection can contain no more than 64 KB of data. Nothing will prevent you from inserting more than 64 KB of data into a collection, but when you try to read the collection back, the result will be truncated at 64 KB, resulting in data loss.
Of course, a collection that large would be quite unwieldy to work with, so you should stick with collections that are far smaller than that threshold.
In general, data that can be expected to grow in an unbounded fashion is inappropriate for collection columns; stick with datasets that ...
