Regression in `information_schema.key_column_usage`?

CrateDB version

crate/crate:nightly (still works on 5.0.1)

CrateDB setup information

docker pull crate/crate:nightly
docker run --rm -it --publish=4200:4200 crate/crate:nightly -Cdiscovery.type=single-node -Ccluster.routing.allocation.disk.threshold_enabled=false

vs.

docker run --rm -it --publish=4200:4200 crate/crate:5.0.1 -Cdiscovery.type=single-node -Ccluster.routing.allocation.disk.threshold_enabled=false

Steps to Reproduce

crash <<SQL
  CREATE TABLE testdrive ("id" TEXT PRIMARY KEY);
  SELECT column_name FROM information_schema.key_column_usage WHERE table_name = 'testdrive' AND table_catalog = 'doc';
SQL

Expected Result

cr> CREATE TABLE testdrive ("id" TEXT PRIMARY KEY);
CREATE OK, 1 row affected  (0.217 sec)
cr> SELECT column_name FROM information_schema.key_column_usage WHERE table_name = 'testdrive' AND table_catalog = 'doc';
+-------------+
| column_name |
+-------------+
| id          |
+-------------+
SELECT 1 row in set (0.006 sec)

Actual Result

cr> CREATE TABLE testdrive ("id" TEXT PRIMARY KEY);
CREATE OK, 1 row affected  (0.235 sec)
cr> SELECT column_name FROM information_schema.key_column_usage WHERE table_name = 'testdrive' AND table_catalog = 'doc';
+-------------+
| column_name |
+-------------+
+-------------+
SELECT 0 rows in set (0.080 sec)

[amotl]

Hi again,

this issue has been converged into a patch to the Python driver, see https://github.com/crate/crate-python/pull/459. As far as I can see, no other relevant/popular driver for CrateDB needs to be updated.

With kind regards, Andreas.

[seut]

But unfortunately key_column_usage.table_schema did return a wrong, hardcoded value on CrateDB 2.3.x, which was fixed at CrateDB 3.0. So we have to add a switch at the SQLAlchemy dialect for this versions unfortunately.