Re-prepare statements after schema changes

[levkk]

Automatically re-prepare and retry transactions that hit this error:

cached plan must not change result type.

The implementation should do something like this:

1. Catch the error
2. Re-prepare statement in request
3. Retry the request

To make this safe, make sure the request has only one Bind & Execute messages - any more and the request is not safe to retry because it can run the same query twice.

[dowzhong]

Hey, i'm guessing that beyond having only 1 bind and execute, we would also have to make sure that the statement is currently the only one in-flight right?

for example, if we had 2 statements:

---- statement 1 ----
parse
bind
execute <- cached plan must not change result type
sync
---- statement 2 ----
bind
execute
sync

retrying statement 1 would cause it's response to come back out of order to the client (since we're not buffering statement 2, it's sent immediately after).

i could be misunderstanding this though, what do you think

[levkk]

This is correct. However, our client handling code guarantees that there is only one request / statement in-flight at any given point, so the implementation here doesn't have to handle that edge case.