[IMPALA-1262] user initiated query cancellation should return operation_state = CANCELED_STATE

• Type: Bug
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: Impala 1.4
• Component/s: Backend
• Labels: query-lifecycle
• Target Version: Product Backlog
• Assignee: Bikramjeet Vig
• Reporter: Daniel Hecht
• Created: 16/Sep/14 23:56
• Updated: 15/Apr/19 17:10

Description

I've been told that if the user cancels the query, our intention is to not set query_state to EXCEPTION. (Is that the desired behavior?) Adding a test case to test_cancellation.py as follows, shows that this is not the current behavior (at least not in all circumstances).

--- a/tests/query_test/test_cancellation.py
+++ b/tests/query_test/test_cancellation.py
@@ -102,6 +102,7 @@ class TestCancellation(ImpalaTestSuite):
       sleep(vector.get_value('cancel_delay'))
       assert self.client.get_state(handle) != self.client.QUERY_STATES['EXCEPTION']
       cancel_result = self.client.cancel(handle)
+      assert self.client.get_state(handle) != self.client.QUERY_STATES['EXCEPTION']
       assert cancel_result.status_code == 0,\
           'Unexpected status code from cancel request: %s' % cancel_result
       thread.join()

One example of how this can happen is if the cancel happens while the wait thread is still in WaitInternal, then WaitInternal will return Status::CANCELLED, and UpdateQueryStatus indiscriminately sets query_state_ to EXCEPTION if !ok(). But, I don't think that side of the execution can distinguish between user initiated cancel and error initiated cancel. Probably, QueryExecState::Cancel needs to remember which kind of cancellation it was so that the cancelled side can do the right thing.

We don't have a general workaround.

Issue Links

• is duplicated by: IMPALA-2371 Query state/status of cancelled query is reported as FINISHED/OK
• is related to: IMPALA-8411 A race during query cancellation by a client can make the client see it as still running
• is related to: IMPALA-4714 Idle session expired query goes in to exception state - And this is confusing
• is related to: IMPALA-7561 Expired queries should end up in a consistent state (EXCEPTION, CANCELLED, or something else)
• relates to: IMPALA-6614 ClientRequestState should use HS2 TOperationState states rather than Beeswax QueryState