Josh McKenzie created CASSANDRA-17772:
-----------------------------------------
Summary: Improve documentation around query methods in CQLTester
and GuardrailTester
Key: CASSANDRA-17772
URL: https://issues.apache.org/jira/browse/CASSANDRA-17772
Project: Cassandra
Issue Type: Improvement
Components: Feature/Guardrails
Reporter: Josh McKenzie
Assignee: Josh McKenzie
Anything that relies on CQLTester.executeFormattedQuery (the
assertInvalidThrowMessage methods for instance) will use internal client state
rather than the client state specified for the query, thus nullifying any
guardrail or systems keyspace permission checks as the ClientState.isInternal
flag overrides all such permission checking. Reference:
[link|https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/service/ClientState.java#L385-L388]
See chain of CQLTester -> ClientState.isInternal here if interested:
[CQLTester|https://github.com/apache/cassandra/blob/trunk/test/unit/org/apache/cassandra/cql3/CQLTester.java#L1325]
[QueryProcessor|https://github.com/apache/cassandra/blob/8451acc8d8dcfee20d692d1e70ae11b60d2f004e/src/java/org/apache/cassandra/cql3/QueryProcessor.java#L447]
This is pretty easy to stumble across when testing guardrails as
GuardrailTester adds a variety of assertFails and assertThrows signatures that
_do_ respect the client state and thus apply guardrails
([example|https://github.com/apache/cassandra/blob/trunk/test/unit/org/apache/cassandra/db/guardrails/GuardrailTester.java#L271])
We should add documentation to the method calls in CQLTester and
GuardrailTester to reflect this discrepancy as it can easily trip someone up
writing tests for guardrails.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
