[
https://issues.apache.org/jira/browse/FLINK-40568?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Purushottam Sinha updated FLINK-40568:
--------------------------------------
Description:
h3. Problem
{{ProcessTableFunctionSemanticTests}} is flaky on master for the
{{process-updating-multi-input}} program. It runs the two-input updating PTF
join {{UpdatingJoinFunction}} and asserts the sink's raw changelog with
{{{}containsExactlyInAnyOrder{}}}, but a two-input operator has no
deterministic cross-input arrival order, so the emitted changelog varies. The
test pins one interleaving, so an alternate order fails the assertion.
h3. Evidence
* master run
[https://github.com/apache/flink/actions/runs/33989974423/job/101371696744] —
{{runSteps(TableTestProgram)[64]}} {{{}[process-updating-multi-input]{}}},
{{{}Tests run: 11458, Failures: 1{}}}.
* Failing assertion (job is awaited, so this is the complete output, not a
truncation):
{noformat}
Expecting actual:
["+I[Bob, score 2 in city London]", "+U[Bob, score 2 in city Berlin]"]
to contain exactly in any order:
["+I[Bob, score 5 in city London]", "+I[Alice, score 2 in city Zurich]",
"+U[Bob, score 3 in city London]", "+U[Bob, score 3 in city Berlin]",
"-D[Bob, null]", "+I[Bob, score 2 in city Berlin]", "-D[Alice, null]"]
{noformat}
* The failing run's head SHA is current master ({{{}c51e71536ed{}}}); the same
commit passed 51/51 local runs (JDK 17), so this is a flake, not a regression
at this commit.
h3. Notes
* The final upsert-materialized result converges across interleavings; only
the raw changelog differs. A fix could assert the materialized result instead
of the raw changelog for this program, or otherwise make the expectation
interleave-independent.
was:
h3. Problem
{{ProcessTableFunctionSemanticTests}} is flaky on master for the
{{process-updating-multi-input}} program. It runs the two-input updating PTF
join {{UpdatingJoinFunction}} and asserts the sink's raw changelog with
{{containsExactlyInAnyOrder}}, but a two-input operator has no deterministic
cross-input arrival order, so the emitted changelog varies. The test pins one
interleaving, so an alternate order fails the assertion.
h3. Evidence
* master run [https://github.com/apache/flink/actions/runs/33989974423]
(module: table) — {{runSteps(TableTestProgram)[64]}}
{{[process-updating-multi-input]}}, {{Tests run: 11458, Failures: 1}}.
* Failing assertion (job is awaited, so this is the complete output, not a
truncation):
{noformat}
Expecting actual:
["+I[Bob, score 2 in city London]", "+U[Bob, score 2 in city Berlin]"]
to contain exactly in any order:
["+I[Bob, score 5 in city London]", "+I[Alice, score 2 in city Zurich]",
"+U[Bob, score 3 in city London]", "+U[Bob, score 3 in city Berlin]",
"-D[Bob, null]", "+I[Bob, score 2 in city Berlin]", "-D[Alice, null]"]
{noformat}
* The failing run's head SHA is current master ({{c51e71536ed}}); the same
commit passed 51/51 local runs (JDK 17), so this is a flake, not a regression
at this commit.
h3. Notes
* The final upsert-materialized result converges across interleavings; only the
raw changelog differs. A fix could assert the materialized result instead of
the raw changelog for this program, or otherwise make the expectation
interleave-independent.
> ProcessTableFunctionSemanticTests[process-updating-multi-input] is flaky on
> master
> ----------------------------------------------------------------------------------
>
> Key: FLINK-40568
> URL: https://issues.apache.org/jira/browse/FLINK-40568
> Project: Flink
> Issue Type: Bug
> Components: Table SQL / Planner
> Affects Versions: 2.4.0
> Reporter: Purushottam Sinha
> Priority: Major
> Labels: pull-request-available, test-stability
>
> h3. Problem
> {{ProcessTableFunctionSemanticTests}} is flaky on master for the
> {{process-updating-multi-input}} program. It runs the two-input updating PTF
> join {{UpdatingJoinFunction}} and asserts the sink's raw changelog with
> {{{}containsExactlyInAnyOrder{}}}, but a two-input operator has no
> deterministic cross-input arrival order, so the emitted changelog varies. The
> test pins one interleaving, so an alternate order fails the assertion.
> h3. Evidence
> * master run
> [https://github.com/apache/flink/actions/runs/33989974423/job/101371696744] —
> {{runSteps(TableTestProgram)[64]}} {{{}[process-updating-multi-input]{}}},
> {{{}Tests run: 11458, Failures: 1{}}}.
> * Failing assertion (job is awaited, so this is the complete output, not a
> truncation):
> {noformat}
> Expecting actual:
> ["+I[Bob, score 2 in city London]", "+U[Bob, score 2 in city Berlin]"]
> to contain exactly in any order:
> ["+I[Bob, score 5 in city London]", "+I[Alice, score 2 in city Zurich]",
> "+U[Bob, score 3 in city London]", "+U[Bob, score 3 in city Berlin]",
> "-D[Bob, null]", "+I[Bob, score 2 in city Berlin]", "-D[Alice, null]"]
> {noformat}
> * The failing run's head SHA is current master ({{{}c51e71536ed{}}}); the
> same commit passed 51/51 local runs (JDK 17), so this is a flake, not a
> regression at this commit.
> h3. Notes
> * The final upsert-materialized result converges across interleavings; only
> the raw changelog differs. A fix could assert the materialized result instead
> of the raw changelog for this program, or otherwise make the expectation
> interleave-independent.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
