spanconfigsplitterccl: TestDataDriven is brittle (perhaps due to weakness in spanconfigtestcluster? unsure) #106818
Assignees
Labels
A-multitenancy
Related to multi-tenancy
A-testing
Testing tools and infrastructure
C-bug
Code not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.
db-cy-23
T-sql-foundations
SQL Foundations Team (formerly SQL Schema + SQL Sessions)
Comments
knz
added
C-bug
|
Ditto in spanconfigsqltranslatorccl. |
knz
added a commit
to knz/cockroach
that referenced
this issue
Jul 14, 2023
There's a mix of tests that control their tenants directly, and tests that should really work with virtualization enabled but don't. Followup issues: cockroachdb#106821 and cockroachdb#106818. Release note: None
knz
added a commit
to knz/cockroach
that referenced
this issue
Jul 14, 2023
There's a mix of tests that control their tenants directly, and tests that should really work with virtualization enabled but don't. Followup issues: cockroachdb#106821 and cockroachdb#106818. Release note: None
knz
added
A-kv
craig bot
pushed a commit
that referenced
this issue
Jul 14, 2023
106738: logic: skip_on_retry works when errors are expected r=Xiang-Gu a=Xiang-Gu Previously, we have `skip_on_retry` directive for logic test which, when set, it skips the rest of test if a statement fails with TransactionRetryError. However, it won't skip if the statement is expected to fail with certain error message. This PR ensures that whenever we have a TransactionRetryError and `skip_on_retry` is set, we always skip the rest of the test, even if the stmt is expected to fail. fixes #104464 Release note: None 106759: streamingccl: unskip TestStreamDeleteRange r=msbutler a=stevendanna This test had previously timed out. The timeout we saw was the result of a couple of issues. When waiting for all delete ranges, our loop exit condition was very strict. We would only stop looking for rows if the number of delete ranges was exactly 3. If, however, we got 4 delete ranges, with 2 coming in a single batch, we would never hit this condition. How would that happen though? One possibility are rangefeed duplicates. Another, and what appears to have been happening in this test, is that the representation of the range deletes observed by the rangefeed consumer is slightly different depending on whether the range delete is delivered as part of a catch-up scan or as part of the rangefeeds steady state. I believe this is because the range deletes overlap but are issued at different time points. When we get them as part of the steady state, we get a trimmed version of the original event. When we get them as part of the catch-up scan, we get them broke up at the point of overlap. Fixes #93568 Epic: none Release note: None 106814: testutils: add helper to target transactions for retries r=lidorcarmel a=stevendanna This helper makes it a little quicker to write a test that tests whether a particular transaction is retry safe. Informs #106417 Epic: none Release note: none 106822: spanconfigccl: remove uses of `TODOTestTenantDisabled` r=stevendanna a=knz Informs #76378 . Epic: CRDB-18499 There's a mix of tests that control their tenants directly, and tests that should really work with virtualization enabled but don't. Followup issues: #106821 and #106818. Release note: None 106832: server: bark loudly if the test tenant cannot be created r=herkolategan a=knz Informs #76378 Informs #103772. Epic: CRDB-18499 For context, the automatic test tenant machinery is currently dependent on a CCL enterprise license check. (This is fundamentally not necessary - see #103772 - but sadly this is the way it is for now) Prior to this patch, if the user or a test selected the creation of a test tenant, but the test code forgot to import the required CCL go package, the framework would announce that "a test tenant was created" but it was actually silently failing to do so. This led to confusing investigations where a test tenant was expected, a test was appearing to succeed, but with a release build the same condition would fail. This commit enhances the situation by ensuring we have clear logging output when the test tenant cannot be created due to the missing CCL import. Release note: None Co-authored-by: Xiang Gu <[email protected]> Co-authored-by: Steven Danna <[email protected]> Co-authored-by: Raphael 'kena' Poss <[email protected]>
msbutler
pushed a commit
to msbutler/cockroach
that referenced
this issue
Jul 17, 2023
There's a mix of tests that control their tenants directly, and tests that should really work with virtualization enabled but don't. Followup issues: cockroachdb#106821 and cockroachdb#106818. Release note: None
andrewbaptist
added
T-sql-foundations
|
@andrewbaptist I noticed you moved this to the SQL Foundations team. Could you provide a few more details? I don't think our team has worked on this test before, but if the error is pointing towards something in the SQL Foundations area we can take a look. |
rafiss
added
C-test-failure
rafiss
removed
the
C-test-failure
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Informs #76378
Describe the problem
When
TestDataDrivenis run with a random test tenant server assigned, the test crashes with a nil pointer exception.It's unclear what is happening.
To Reproduce
Replace the use of
TestDoesNotWorkWithSecondaryTenantsButWeDontKnowWhyYetbyTestTenantAlwaysEnabledand run the test.Expected behavior
From a test quality perspective, if the test code lands in an unsupported configuration it should fail with
t.Fatalnot with a crash.Jira issue: CRDB-29728
Epic: CRDB-18499
The text was updated successfully, but these errors were encountered: