Scylla rejects deletions with empty-tuple restrictions

[dekimir]

This is an unintended consequence of 1fdaeca.

Scylla:

cqlsh:ks> CREATE TABLE test (k1 int, k2 int, v int, PRIMARY KEY (k1, k2));
cqlsh:ks> DELETE FROM test WHERE k1 IN ();
InvalidRequest: Error from server: code=2200 [Invalid query] message="Invalid partition key in a modification statement"

Cassandra:

cqlsh:ks> CREATE TABLE test (k1 int, k2 int, v int, PRIMARY KEY (k1, k2));
cqlsh:ks> DELETE FROM test WHERE k1 IN ();

[bhalevy]

@scylladb/scylla-maint we hit this issue in https://jenkins.scylladb.com/job/scylla-4.4/job/dtest-release/121/testReport/cql_additional_tests/TestCQL/empty_in_test/

Since branch-4.4 now includes 146f7b5

Please backport

[psarna]

will do

[psarna]

update: very non-trivial backport, it will take me a while, since it's no longer possible to ask the author to try and do so

[bhalevy]

@psarna Then maybe we should revert 146f7b5

[psarna]

we could, but are empty tuple restrictions ever seen outside of our test environment? Perhaps it's better to keep the patch that rejects updates with null key values at the price of not allowing empty tuple restrictions. Unless some 4.4 users actually rely on empty tuple restrictions, in which case it's better to revert

[nyh]

@psarna so you're suggesting that for 4.4, we need to disable the test, and leave the code as it is now?

[bhalevy]

On Tue, 2021-12-07 at 18:29 +0700, @aleksbykov wrote:

It is new behavior according this fix:
146f7b5

On Tue, Dec 7, 2021 at 6:10 PM Alex Bykov alex.bykov@scylladb.com wrote:

Hi all,
All rolling-upgrade tests for 4.4.8 are failed due to next error:
Cassandra.InvalidRequest: Error from server: code=2200 [Invalid query] message="Invalid partition key in a modification statement"
for next queries:

"DELETE FROM empty_in_test1 WHERE k1 IN ()",
"UPDATE empty_in_test1 SET v = 3 WHERE k1 IN () AND k2 = 2",
"DELETE FROM empty_in_test2 WHERE k1 IN ()",
"UPDATE empty_in_test2 SET v = 3 WHERE k1 IN () AND k2 = 2",

This error happened after the first node was upgraded to 4.4.8.
Was any changes validating cql queries and it is expected?
or it could be regression?

--
Thanks, Aleksandr Bykov

[psarna]

@nyh yes, that's one of the ways to deal with the problem, but it depends on at least two important factors:

1. Are any of our users actually hit by this problem? If there are any, we should consider either a backport or a revert
2. How long until 4.4 reaches end-of-life? Doing a complicated backport for the sake of fixing a release for 2 weeks is probably not worth it

[avikivity]

It's not an important bug.

How did it start happening? Was a new test added?

[aleksbykov]

Now, rolling upgrade stay the same. Just for v4.4.8 upgrade starts failing once first node was upgrade and these query failed with error instead of return empty results, Also i think the upgrade tests from 4.4 -> 4.5 could fail before upgrade because same query executed before upgrade

[avikivity]

I reverted it from 4.4, we can revisit it but this gets us back to status quo ante.

[bhalevy]

@psarna how about 4.5?
It has 44c784c
but not 6afdc60

[bhalevy]

how about 4.5?
It has 44c784c
but not 6afdc60

@psarna ^^ (wrong user tagged previously, sorry)

[bhalevy]

And this regression since https://jenkins.scylladb.com/view/scylla-4.5/job/scylla-4.5/job/dtest-release/111/testReport/cql_additional_tests/TestCQL/empty_in_test/
at 44c784c

@avikivity so we should probably revert 44c784c from 4.5

[avikivity]

Reverted from 4.5.

[fgelcer]

Reverted from 4.5.

EDIT:
this job ran before the 44c784c was reverted from branch 4.5

now all rolling upgrade jobs from 4.4 and 4.5 to 4.5.3 are failing with this error:

DELETE FROM empty_in_test1 WHERE k1 IN () < t:2021-12-22 04:26:51,359 f:fill_db_data.py l:3220 c:sdcm.fill_db_data    p:ERROR > DELETE FROM empty_in_test1 WHERE k1 IN ()
Traceback (most recent call last):
  File "/jenkins/slave/workspace/scylla-4.5/rolling-upgrade/rolling-upgrade-centos7-test@2/scylla-cluster-tests/sdcm/fill_db_data.py", line 3217, in _run_db_queries
    res = session.execute(item['queries'][i])
  File "/jenkins/slave/workspace/scylla-4.5/rolling-upgrade/rolling-upgrade-centos7-test@2/scylla-cluster-tests/sdcm/utils/common.py", line 1263, in execute_verbose
    return execute_orig(*args, **kwargs)
  File "cassandra/cluster.py", line 2611, in cassandra.cluster.Session.execute
  File "cassandra/cluster.py", line 4829, in cassandra.cluster.ResponseFuture.result
cassandra.InvalidRequest: Error from server: code=2200 [Invalid query] message="Invalid partition key in a modification statement"

it means that, once these changes were reverted for 4.5 (and 4.4) the tests in SCT also had to be changed to not fail with this empty tuple...

is it still supported for 4.6?

db cluster logs
sct logs

[avikivity]

The bug should be fixed in 4.6 (and not in 4.4 and 4.5, after the reverts).

[avikivity]

Already backported to all vulnerable branches, removing "Backport candidate" label.