Java – how Cassandra pages behave during concurrent inserts

I use the Java client and pagingstate to page the large result set of cassanda 2.2, as follows: https://datastax.github.io/java-driver/2.2.0-rc2/features/paging/

This works well, but I can't find any information about Cassandra's behavior. When browsing the results, insert new records (or existing updates) Does the result contain such new or changed records, or is the result set immutable?

The use case is a stateless web service in which the client can query a large result set

Edit: Generally speaking, the problem of resultset paging is the same (Cassandra performs automatic delay extraction here)

Edit2: as far as I know, Cassandra does not support acid but aid transactions, so I will expect a kind of isolation when conducting result sets

Solution

There is no such isolation because it is too expensive to implement The whole result set will not be kept in memory, and the row to be returned in the next page is unknown when the current result set will be sent to the client

An interesting result of this is that it breaks the batch update guarantee. It is stated in the document as follows:

There is an open issue about this

There is also some performance impact, because a lot of work done to get page n must be done again to get page n 1 (such as opening and reading index files and data files) Scylla is a substitute for Cassandra I participated in and is trying to develop fixing this