JAVA-3055 CqlPrepareAsyncProcessor must handle cancellations of the r…

[tomas-bilka]

JAVA-3055 CqlPrepareAsyncProcessor must handle cancellations of the returned Future

[hhughes]

Hi @tomas-bilka thank you for the PR.

Have you signed the Contributor License Agreement for contributions to DataStax open source projects? If not you can find it at https://cla.datastax.com/. Thanks!

[tomas-bilka]

I just signed the CLA.

[lucboutier]

I made another PR for 3055 some time ago. Then had a review from a colleague for a better and simpler implementation which I believe is quite nice.

See https://github.com/apache/cassandra-java-driver/pull/1757

[mcoolive]

Any progress?

[aerain]

I think this approach doesn't seem to be able to prevent the underlying problem.

There is edge case When two duplicate prepare requests A and B are received (assuming A arrived slightly earlier), the cancellation of one request propagates to the other.

1. Request A
2. Cache loaded with request A. Asynchronous prepare call to Cassandra (referred to as 'a')
3. Request B
4. B waits for a response from the cache based on request A
5. cancel() in request A
6. B receives the cancel() from request A and propagates a CancellationException
7. Response from 'a' completes. Upon receiving the cancellation of A, cache invalidation proceeds at this time

In my opinion, lucboutier's PR looks better to me.