Wrong handling of multiple consecutive empty pages in Java Driver 4.x

[avelanarius]

When Scylla returns multiple consecutive empty pages, Java Driver 4.x doesn't handle it correctly. Reproducer:

CqlSession cqlSession = CqlSession.builder()
        .addContactPoint(InetSocketAddress.createUnresolved("127.0.0.2", 9042))
        .withLocalDatacenter("datacenter1").build();

cqlSession.execute("CREATE KEYSPACE IF NOT EXISTS ks WITH REPLICATION = {'class': 'SimpleStrategy', 'replication_factor': 1}");

cqlSession.execute("CREATE TABLE IF NOT EXISTS ks.t (pk int, ck int, v int, PRIMARY KEY(pk, ck))");

for (int i = 0; i < 50; i++) {
    System.out.println("Insert " + i);
    cqlSession.execute("INSERT INTO ks.t(pk, ck, v) VALUES (?, ?, ?)", i, i, 15);
    cqlSession.execute("INSERT INTO ks.t(pk, ck, v) VALUES (?, ?, ?)", i, i, 32);
}

cqlSession.execute("INSERT INTO ks.t(pk, ck, v) VALUES (?, ?, ?)", 8, 8, 14);
cqlSession.execute("INSERT INTO ks.t(pk, ck, v) VALUES (?, ?, ?)", 11, 11, 14);
cqlSession.execute("INSERT INTO ks.t(pk, ck, v) VALUES (?, ?, ?)", 14, 14, 14);

SimpleStatement st = SimpleStatement.newInstance("SELECT * FROM ks.t WHERE v = 14 ALLOW FILTERING");
st = st.setPageSize(1);

cqlSession.execute(st).all().forEach(r -> {
    System.out.println(r);
});
cqlSession.close();

The code is supposed to print 3 rows, but it doesn't print anything.

[avelanarius]

Maybe this is a wrong repo to write this, but as for other drivers:

• Rust Driver does not have this problem.
• Java Driver 3.x does not have this problem.

[vladzcloudius]

First of all I'm not sure the "expected behavior" is correct: you set the page size to 1 and execute the query just once. AFAIR this way you are supposed to get a single page only.
And as such you should be getting a single row at most. Don't you?

If I'm right then the behavior of the 4.x driver is actually right (it apparently brings you a single empty page) while the behavior of other drivers isn't.

Or possibly the semantics of execute() got changed between drivers and Java driver versions:

• It 4.x it brings you a single page.
• In 3.x and in Rust - brings you all pages (which is kinda weird and seems bogus).

[avelanarius]

First of all I'm not sure the "expected behavior" is correct: you set the page size to 1 and execute the query just once. AFAIR this way you are supposed to get a single page only. And as such you should be getting a single row at most. Don't you?

No, the driver is supposed to transparently fetch new pages as you iterate (https://docs.datastax.com/en/developer/java-driver/4.13/manual/core/paging/#quick-overview).

[vladzcloudius]

First of all I'm not sure the "expected behavior" is correct: you set the page size to 1 and execute the query just once. AFAIR this way you are supposed to get a single page only. And as such you should be getting a single row at most. Don't you?

No, the driver is supposed to transparently fetch new pages as you iterate (https://docs.datastax.com/en/developer/java-driver/4.13/manual/core/paging/#quick-overview).

Got it. Then there is some bug indeed.
I remembered that you needed some combo of hasMorePages(...) and execute() in a loop to fetch results. But I see that they hid these in the iterator these days.
Thanks for explaining.

[roydahan]

@avelanarius are you sending a PR with the fix (avelanarius@40da47a)?
(Unless you already did and I missed it.

[sylwiaszunejko]

Update about other drivers:

• Python Driver does not have this problem.
• gocql does not have this problem.

[mykaul]

• gocql does have this problem.

Thanks @sylwiaszunejko - please open an issue on its repo so we can track it properly.

[sylwiaszunejko]

• gocql does have this problem.

Thanks @sylwiaszunejko - please open an issue on its repo so we can track it properly.

Sorry for my mistake, I already edited the comment it does not have this problem

[mykaul]

• gocql does have this problem.

Thanks @sylwiaszunejko - please open an issue on its repo so we can track it properly.

Sorry for my mistake, I already edited the comment it does not have this problem

Even better! So now lets just make sure Rust and Java are safe (and btw, would be good to add tests to them)

[roydahan]

@avelanarius are we sending PR to the upstream driver? to our fork? Both?

[mykaul]

@avelanarius - what's the latest status here?

[avelanarius]

Solved by merged a04fadf