[
https://issues.apache.org/jira/browse/IMPALA-14401?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jason Fehr updated IMPALA-14401:
--------------------------------
Description:
Several improvements can be made to the "test_otel_trace.py" custom cluster
test to speed up the tests and deflake them:
1. In test_otel_trace.py, the test_ddl_createtable_fail has a potential to be
flaky. In the test, the second coordinator that drops the database may run that
DDL too quickly before the create database operation executed on the first
coordinator propagates to the second coordinator.
In the test, first retrieve the number of databases from the
"catalog.num-databases" metric:
{code:python}
num_dbs =
self.cluster.get_first_impalad().service.get_metric_value("catalog.num-databases")
{code}
Then, before the drop database DDL, wait for the second coordinator's
"catalog.num-databases" metric to increment by 1:
{code:python}
self.cluster.impalads[1].service.wait_for_metric_value("catalog.num-databases",
num_dbs + 1, 60)
{code}
The other option for this test is to use a debug action to fail the DDL.
2. The "unique_database" fixture is used across multiple tests. This causes 3
extra queries to be run each time this fixture is used (during setup -- drop db
and create db, during teardown -- drop db). Create the database during test
class setup and drop it during test class teardown.
3. There is no test for insert dml failure.
4. Several tests are missing comments explaining their purpose.
5. The code to assert traces would be better in tests/util/otel_trace.py.
Having it in test_otel_trace.py distracts from the test functionality.
6. test_query_exec_fail has an unnecessary sleep in the debug action.
was:
Several improvements can be made to the "test_otel_trace.py" custom cluster
test to speed up the tests and deflake them:
1. In test_otel_trace.py, the test_ddl_createtable_fail has a potential to be
flaky. In the test, the second coordinator that drops the database may run that
DDL too quickly before the create database operation executed on the first
coordinator propagates to the second coordinator.
In the test, first retrieve the number of databases from the
"catalog.num-databases" metric:
{code:python}
num_dbs =
self.cluster.get_first_impalad().service.get_metric_value("catalog.num-databases")
{code}
Then, before the drop database DDL, wait for the second coordinator's
"catalog.num-databases" metric to increment by 1:
{code:python}
self.cluster.impalads[1].service.wait_for_metric_value("catalog.num-databases",
num_dbs + 1, 60)
{code}
The other option for this test is to use a debug action to fail the DDL.
2. The "unique_database" fixture is used across multiple tests. This causes 3
extra queries to be run each time this fixture is used (during setup -- drop db
and create db, during teardown -- drop db). Create the database during test
class setup and drop it during test class teardown.
3. There is no test for insert dml failure.
4. Several tests are missing comments explaining their purpose.
5. The code to assert traces would be better in tests/util/otel_trace.py.
Having it in test_otel_trace.py distracts from the test functionality.
> Deflake/Improve OTel Custom Cluster Tests
> -----------------------------------------
>
> Key: IMPALA-14401
> URL: https://issues.apache.org/jira/browse/IMPALA-14401
> Project: IMPALA
> Issue Type: Improvement
> Affects Versions: Impala 5.0.0
> Reporter: Jason Fehr
> Assignee: Jason Fehr
> Priority: Critical
>
> Several improvements can be made to the "test_otel_trace.py" custom cluster
> test to speed up the tests and deflake them:
> 1. In test_otel_trace.py, the test_ddl_createtable_fail has a potential to be
> flaky. In the test, the second coordinator that drops the database may run
> that DDL too quickly before the create database operation executed on the
> first coordinator propagates to the second coordinator.
> In the test, first retrieve the number of databases from the
> "catalog.num-databases" metric:
> {code:python}
> num_dbs =
> self.cluster.get_first_impalad().service.get_metric_value("catalog.num-databases")
> {code}
> Then, before the drop database DDL, wait for the second coordinator's
> "catalog.num-databases" metric to increment by 1:
> {code:python}
> self.cluster.impalads[1].service.wait_for_metric_value("catalog.num-databases",
> num_dbs + 1, 60)
> {code}
> The other option for this test is to use a debug action to fail the DDL.
> 2. The "unique_database" fixture is used across multiple tests. This causes
> 3 extra queries to be run each time this fixture is used (during setup --
> drop db and create db, during teardown -- drop db). Create the database
> during test class setup and drop it during test class teardown.
> 3. There is no test for insert dml failure.
> 4. Several tests are missing comments explaining their purpose.
> 5. The code to assert traces would be better in tests/util/otel_trace.py.
> Having it in test_otel_trace.py distracts from the test functionality.
> 6. test_query_exec_fail has an unnecessary sleep in the debug action.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
