Batch execution is incomplete when client doesn't request all results

[lukaseder]

Bug Report

Versions

• Driver: 0.8.4.RELEASE
• SPI: 0.9.0.M1
• Database: 1.4.200
• Java: openjdk version "11.0.9.1" 2020-11-04
• OS: Microsoft Windows [Version 10.0.19042.867]

Current Behavior

The current implementation of r2dbc-h2 does not execute batches completely if clients don't consume (request?) all the results. In my opinion, a batch is an atomic unit of work, irrespective of whether all update counts are consumed, and must be executed atomically by the server.

Steps to reproduce

Run this code

ConnectionFactory connectionFactory = ConnectionFactories.get(
    ConnectionFactoryOptions
        .parse("r2dbc:h2:file://localhost/~/x")
        .mutate()
        .option(ConnectionFactoryOptions.USER, "sa")
        .option(ConnectionFactoryOptions.PASSWORD, "")
        .build()
);

Flux.from(connectionFactory.create())
    .flatMap(c -> c.createStatement("create table t (i int)").execute())
    .collectList()
    .block();

try {
    System.out.println((
        Flux.from(connectionFactory.create())
            .flatMap(c -> c.createBatch().add("insert into t values (1)").add("insert into t values (2)").execute())
            .flatMap(t -> t.getRowsUpdated())
            .blockFirst()
    ));

    System.out.println(
        Flux.from(connectionFactory.create())
            .flatMap(c -> c.createStatement("select * from t").execute())
            .flatMap(it -> it.map((r, m) -> r.get(0)))
            .collectList()
            .block()
    );
}
finally {
    Flux.from(connectionFactory.create())
        .flatMap(c -> c.createStatement("drop table t").execute())
        .collectList()
        .block();
}

The output is:

1
[1]

As can be seen, the second value 2 is never inserted.

Expected behavior/code

The output in r2dbc-postgresql is the expected one:

1
[1, 2]

Only one update count is consumed as requested, but two insertions are made