You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "StefanRRichter (via GitHub)" <gi...@apache.org> on 2023/05/15 12:37:51 UTC
[GitHub] [flink] StefanRRichter opened a new pull request, #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
StefanRRichter opened a new pull request, #22584:
URL: https://github.com/apache/flink/pull/22584
[FLINK-31963]
<!--
*Thank you very much for contributing to Apache Flink - we are happy that you want to help us improve Flink. To help the community review your contribution in the best possible way, please go through the checklist below, which will get the contribution into a shape in which it can be best reviewed.*
*Please understand that we do not do this to make contributions to Flink a hassle. In order to uphold a high standard of quality for code contributions, while at the same time managing a large number of contributions, we need contributors to prepare the contributions well, and give reviewers enough contextual information for the review. Please also understand that contributions that do not follow this guide will take longer to review and thus typically be picked up with lower priority by the community.*
## Contribution Checklist
- Make sure that the pull request corresponds to a [JIRA issue](https://issues.apache.org/jira/projects/FLINK/issues). Exceptions are made for typos in JavaDoc or documentation files, which need no JIRA issue.
- Name the pull request in the form "[FLINK-XXXX] [component] Title of the pull request", where *FLINK-XXXX* should be replaced by the actual issue number. Skip *component* if you are unsure about which is the best component.
Typo fixes that have no associated JIRA issue should be named following this pattern: `[hotfix] [docs] Fix typo in event time introduction` or `[hotfix] [javadocs] Expand JavaDoc for PuncuatedWatermarkGenerator`.
- Fill out the template below to describe the changes contributed by the pull request. That will give reviewers the context they need to do the review.
- Make sure that the change passes the automated tests, i.e., `mvn clean verify` passes. You can set up Azure Pipelines CI to do that following [this guide](https://cwiki.apache.org/confluence/display/FLINK/Azure+Pipelines#AzurePipelines-Tutorial:SettingupAzurePipelinesforaforkoftheFlinkrepository).
- Each pull request should address only one issue, not mix up code from multiple issues.
- Each commit in the pull request has a meaningful commit message (including the JIRA id)
- Once all items of the checklist are addressed, remove the above text and this checklist, leaving only the filled out template below.
**(The sections below can be removed for hotfixes of typos)**
-->
## What is the purpose of the change
This commit fixes problems in StateAssignmentOperation for unaligned checkpoints with stateless operators that have upstream operators with output partition state or downstream operators with input channel state. Without this fix, state assignment does not consider such upstream/downstream states for rescaling and skip state reassignment for otherwise stateless operators.
## Brief change log
- Checking for upstream result partitions and downstream input channel states in `StateAssignmentOperation` before skipping reassignment.
- Unit test
- IT test
## Verifying this change
This change added tests and can be verified as follows:
- Run the added IT tests in `UnalignedCheckpointRescaleITCase`
- Run the added unit tests in `StateAssignmentOperationTest`
## Does this pull request potentially affect one of the following parts:
- Dependencies (does it add or upgrade a dependency): (no)
- The public API, i.e., is any changed class annotated with `@Public(Evolving)`: (no)
- The serializers: (no)
- The runtime per-record code paths (performance sensitive): (yes / no / don't know)
- Anything that affects deployment or recovery: JobManager (and its components), Checkpointing, Kubernetes/Yarn, ZooKeeper: (yes: rescaling state)
- The S3 file system connector: (no)
## Documentation
- Does this pull request introduce a new feature? (no)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] rkhachatryan commented on a diff in pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "rkhachatryan (via GitHub)" <gi...@apache.org>.
rkhachatryan commented on code in PR #22584:
URL: https://github.com/apache/flink/pull/22584#discussion_r1193908276
##########
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperationTest.java:
##########
@@ -1029,6 +1192,15 @@ private JobVertex createJobVertex(
return jobVertex;
}
+ private List<TaskStateSnapshot> getRescalingDescriptorsFromVertex(
Review Comment:
`getTaskStateSnapshotFromVertex`?
##########
flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperation.java:
##########
@@ -136,19 +136,24 @@ public void assignStates() {
// repartition state
for (TaskStateAssignment stateAssignment : vertexAssignments.values()) {
- if (stateAssignment.hasNonFinishedState) {
+ if (stateAssignment.hasNonFinishedState
+ // FLINK-31963: We need to run repartitioning for stateless operators that have
+ // upstream output or downstream input states.
+ || stateAssignment.hasUpstreamOutputStates()
+ || stateAssignment.hasDownstreamInputStates()) {
assignAttemptState(stateAssignment);
}
}
// actually assign the state
for (TaskStateAssignment stateAssignment : vertexAssignments.values()) {
- // If upstream has output states, even the empty task state should be assigned for the
- // current task in order to notify this task that the old states will send to it which
- // likely should be filtered.
+ // If upstream has output states or downstream has input states, even the empty task
+ // state should be assigned for the current task in order to notify this task that the
+ // old states will send to it which likely should be filtered.
if (stateAssignment.hasNonFinishedState
|| stateAssignment.isFullyFinished
- || stateAssignment.hasUpstreamOutputStates()) {
+ || stateAssignment.hasUpstreamOutputStates()
+ || stateAssignment.hasDownstreamInputStates()) {
Review Comment:
Should we compute once `hasUpstreamOutputStates` and `hasDownstreamInputStates` similar to `isFullyFinished`?
(that would at least be consistent, and also faster)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] flinkbot commented on pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "flinkbot (via GitHub)" <gi...@apache.org>.
flinkbot commented on PR #22584:
URL: https://github.com/apache/flink/pull/22584#issuecomment-1547794482
<!--
Meta data
{
"version" : 1,
"metaDataEntries" : [ {
"hash" : "ea90d57725eb8d2b1a8849dc1a5d746c419c5671",
"status" : "UNKNOWN",
"url" : "TBD",
"triggerID" : "ea90d57725eb8d2b1a8849dc1a5d746c419c5671",
"triggerType" : "PUSH"
} ]
}-->
## CI report:
* ea90d57725eb8d2b1a8849dc1a5d746c419c5671 UNKNOWN
<details>
<summary>Bot commands</summary>
The @flinkbot bot supports the following commands:
- `@flinkbot run azure` re-run the last Azure build
</details>
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] rkhachatryan commented on a diff in pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "rkhachatryan (via GitHub)" <gi...@apache.org>.
rkhachatryan commented on code in PR #22584:
URL: https://github.com/apache/flink/pull/22584#discussion_r1194923077
##########
flink-tests/src/test/java/org/apache/flink/test/checkpointing/UnalignedCheckpointRescaleITCase.java:
##########
@@ -459,46 +478,61 @@ public void processBroadcastElement(Long value, Context ctx, Collector<Long> out
}
}
- @Parameterized.Parameters(name = "{0} {1} from {2} to {3}, buffersPerChannel = {4}")
+ @Parameterized.Parameters(
+ name = "{0} {1} from {2} to {3}, sourceSleepMs = {4}, buffersPerChannel = {5}")
public static Object[][] getScaleFactors() {
+ // We use `sourceSleepMs` > 0 to test rescaling without backpressure and only very few
+ // captured in-flight records, see FLINK-31963.
Object[][] parameters =
new Object[][] {
- new Object[] {"downscale", Topology.KEYED_DIFFERENT_PARALLELISM, 12, 7},
- new Object[] {"upscale", Topology.KEYED_DIFFERENT_PARALLELISM, 7, 12},
- new Object[] {"downscale", Topology.KEYED_BROADCAST, 7, 2},
- new Object[] {"upscale", Topology.KEYED_BROADCAST, 2, 7},
- new Object[] {"downscale", Topology.BROADCAST, 5, 2},
- new Object[] {"upscale", Topology.BROADCAST, 2, 5},
- new Object[] {"upscale", Topology.PIPELINE, 1, 2},
- new Object[] {"upscale", Topology.PIPELINE, 2, 3},
- new Object[] {"upscale", Topology.PIPELINE, 3, 7},
- new Object[] {"upscale", Topology.PIPELINE, 4, 8},
- new Object[] {"upscale", Topology.PIPELINE, 20, 21},
- new Object[] {"downscale", Topology.PIPELINE, 2, 1},
- new Object[] {"downscale", Topology.PIPELINE, 3, 2},
- new Object[] {"downscale", Topology.PIPELINE, 7, 3},
- new Object[] {"downscale", Topology.PIPELINE, 8, 4},
- new Object[] {"downscale", Topology.PIPELINE, 21, 20},
- new Object[] {"no scale", Topology.PIPELINE, 1, 1},
- new Object[] {"no scale", Topology.PIPELINE, 3, 3},
- new Object[] {"no scale", Topology.PIPELINE, 7, 7},
- new Object[] {"no scale", Topology.PIPELINE, 20, 20},
- new Object[] {"upscale", Topology.UNION, 1, 2},
- new Object[] {"upscale", Topology.UNION, 2, 3},
- new Object[] {"upscale", Topology.UNION, 3, 7},
- new Object[] {"downscale", Topology.UNION, 2, 1},
- new Object[] {"downscale", Topology.UNION, 3, 2},
- new Object[] {"downscale", Topology.UNION, 7, 3},
- new Object[] {"no scale", Topology.UNION, 1, 1},
- new Object[] {"no scale", Topology.UNION, 7, 7},
- new Object[] {"upscale", Topology.MULTI_INPUT, 1, 2},
- new Object[] {"upscale", Topology.MULTI_INPUT, 2, 3},
- new Object[] {"upscale", Topology.MULTI_INPUT, 3, 7},
- new Object[] {"downscale", Topology.MULTI_INPUT, 2, 1},
- new Object[] {"downscale", Topology.MULTI_INPUT, 3, 2},
- new Object[] {"downscale", Topology.MULTI_INPUT, 7, 3},
- new Object[] {"no scale", Topology.MULTI_INPUT, 1, 1},
- new Object[] {"no scale", Topology.MULTI_INPUT, 7, 7},
+ new Object[] {"downscale", Topology.KEYED_DIFFERENT_PARALLELISM, 12, 7, 0L},
Review Comment:
nit: I'd consider combining a limited set of manually-crafted cases with a randomly generated ones (different on each run).
That would increase the coverage a bit given that there are a lot of runs on the CI.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] StefanRRichter merged pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "StefanRRichter (via GitHub)" <gi...@apache.org>.
StefanRRichter merged PR #22584:
URL: https://github.com/apache/flink/pull/22584
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] StefanRRichter commented on a diff in pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "StefanRRichter (via GitHub)" <gi...@apache.org>.
StefanRRichter commented on code in PR #22584:
URL: https://github.com/apache/flink/pull/22584#discussion_r1193925009
##########
flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperation.java:
##########
@@ -136,19 +136,24 @@ public void assignStates() {
// repartition state
for (TaskStateAssignment stateAssignment : vertexAssignments.values()) {
- if (stateAssignment.hasNonFinishedState) {
+ if (stateAssignment.hasNonFinishedState
+ // FLINK-31963: We need to run repartitioning for stateless operators that have
+ // upstream output or downstream input states.
+ || stateAssignment.hasUpstreamOutputStates()
+ || stateAssignment.hasDownstreamInputStates()) {
assignAttemptState(stateAssignment);
}
}
// actually assign the state
for (TaskStateAssignment stateAssignment : vertexAssignments.values()) {
- // If upstream has output states, even the empty task state should be assigned for the
- // current task in order to notify this task that the old states will send to it which
- // likely should be filtered.
+ // If upstream has output states or downstream has input states, even the empty task
+ // state should be assigned for the current task in order to notify this task that the
+ // old states will send to it which likely should be filtered.
if (stateAssignment.hasNonFinishedState
|| stateAssignment.isFullyFinished
- || stateAssignment.hasUpstreamOutputStates()) {
+ || stateAssignment.hasUpstreamOutputStates()
+ || stateAssignment.hasDownstreamInputStates()) {
Review Comment:
I was considering that, and then I saw that most of the heavy work for the functions is already cached. Do you think it's still worth it?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] rkhachatryan commented on a diff in pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "rkhachatryan (via GitHub)" <gi...@apache.org>.
rkhachatryan commented on code in PR #22584:
URL: https://github.com/apache/flink/pull/22584#discussion_r1194862690
##########
flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperation.java:
##########
@@ -136,19 +136,24 @@ public void assignStates() {
// repartition state
for (TaskStateAssignment stateAssignment : vertexAssignments.values()) {
- if (stateAssignment.hasNonFinishedState) {
+ if (stateAssignment.hasNonFinishedState
+ // FLINK-31963: We need to run repartitioning for stateless operators that have
+ // upstream output or downstream input states.
+ || stateAssignment.hasUpstreamOutputStates()
+ || stateAssignment.hasDownstreamInputStates()) {
assignAttemptState(stateAssignment);
}
}
// actually assign the state
for (TaskStateAssignment stateAssignment : vertexAssignments.values()) {
- // If upstream has output states, even the empty task state should be assigned for the
- // current task in order to notify this task that the old states will send to it which
- // likely should be filtered.
+ // If upstream has output states or downstream has input states, even the empty task
+ // state should be assigned for the current task in order to notify this task that the
+ // old states will send to it which likely should be filtered.
if (stateAssignment.hasNonFinishedState
|| stateAssignment.isFullyFinished
- || stateAssignment.hasUpstreamOutputStates()) {
+ || stateAssignment.hasUpstreamOutputStates()
+ || stateAssignment.hasDownstreamInputStates()) {
Review Comment:
I think it is, it's still an iteration over all upstream subtasks for every downstream subtask in the worst case, and it would be more consistent with the other checks.
But I'm also fine with the current way.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] rkhachatryan commented on a diff in pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "rkhachatryan (via GitHub)" <gi...@apache.org>.
rkhachatryan commented on code in PR #22584:
URL: https://github.com/apache/flink/pull/22584#discussion_r1193902234
##########
flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperation.java:
##########
@@ -136,19 +136,24 @@ public void assignStates() {
// repartition state
for (TaskStateAssignment stateAssignment : vertexAssignments.values()) {
- if (stateAssignment.hasNonFinishedState) {
+ if (stateAssignment.hasNonFinishedState
+ // FLINK-31963: We need to run repartitioning for stateless operators that have
+ // upstream output or downstream input states.
+ || stateAssignment.hasUpstreamOutputStates()
+ || stateAssignment.hasDownstreamInputStates()) {
assignAttemptState(stateAssignment);
}
}
// actually assign the state
for (TaskStateAssignment stateAssignment : vertexAssignments.values()) {
- // If upstream has output states, even the empty task state should be assigned for the
- // current task in order to notify this task that the old states will send to it which
- // likely should be filtered.
+ // If upstream has output states or downstream has input states, even the empty task
+ // state should be assigned for the current task in order to notify this task that the
+ // old states will send to it which likely should be filtered.
if (stateAssignment.hasNonFinishedState
|| stateAssignment.isFullyFinished
- || stateAssignment.hasUpstreamOutputStates()) {
+ || stateAssignment.hasUpstreamOutputStates()
+ || stateAssignment.hasDownstreamInputStates()) {
Review Comment:
Should we compute `hasUpstreamOutputStates` and `hasDownstreamInputStates` once, similar to `isFullyFinished`?
(that would at least be consistent, and also faster)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] StefanRRichter commented on a diff in pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "StefanRRichter (via GitHub)" <gi...@apache.org>.
StefanRRichter commented on code in PR #22584:
URL: https://github.com/apache/flink/pull/22584#discussion_r1193856106
##########
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperationTest.java:
##########
@@ -785,6 +789,130 @@ public void testOnlyUpstreamChannelStateAssignment()
}
}
+ /** FLINK-31963: Tests rescaling for stateless operators and upstream result partition state. */
+ @Test
+ public void testOnlyUpstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState upstreamOpState =
+ OperatorSubtaskState.builder()
+ .setResultSubpartitionState(
+ new StateObjectCollection<>(
+ asList(
+ createNewResultSubpartitionStateHandle(10, random),
+ createNewResultSubpartitionStateHandle(
+ 10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(upstreamOpState, null, 5, 7);
+ }
+
+ /** FLINK-31963: Tests rescaling for stateless operators and downstream input channel state. */
+ @Test
+ public void testOnlyDownstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState downstreamOpState =
+ OperatorSubtaskState.builder()
+ .setInputChannelState(
+ new StateObjectCollection<>(
+ asList(
+ createNewInputChannelStateHandle(10, random),
+ createNewInputChannelStateHandle(10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(null, downstreamOpState, 5, 5);
+ }
+
+ private void testOnlyUpstreamOrDownstreamRescalingInternal(
+ @Nullable OperatorSubtaskState upstreamOpState,
+ @Nullable OperatorSubtaskState downstreamOpState,
+ int expectedUpstreamCount,
+ int expectedDownstreamCount)
+ throws JobException, JobExecutionException {
+
+ if ((upstreamOpState == null && downstreamOpState == null)
+ || (upstreamOpState != null && downstreamOpState != null)) {
+ // Either upstream or downstream state must exist, but not both.
+ return;
+ }
+
+ // Start from parallelism 5 for both operators
+ int upstreamParallelism = 5;
+ int downstreamParallelism = 5;
+
+ // Build states
+ List<OperatorID> operatorIds = buildOperatorIds(2);
+ Map<OperatorID, OperatorState> states = new HashMap<>();
+ OperatorState upstreamState =
+ new OperatorState(operatorIds.get(0), upstreamParallelism, MAX_P);
+ OperatorState downstreamState =
+ new OperatorState(operatorIds.get(1), downstreamParallelism, MAX_P);
+
+ states.put(operatorIds.get(0), upstreamState);
+ states.put(operatorIds.get(1), downstreamState);
+
+ if (upstreamOpState != null) {
+ upstreamState.putState(0, upstreamOpState);
+ // rescale downstream 5 -> 3
+ downstreamParallelism = 3;
+ }
+
+ if (downstreamOpState != null) {
+ downstreamState.putState(0, downstreamOpState);
+ // rescale upstream 5 -> 3
+ upstreamParallelism = 3;
+ }
+
+ List<OperatorIdWithParallelism> opIdWithParallelism = new ArrayList<>(2);
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(0), upstreamParallelism));
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(1), downstreamParallelism));
+
+ Map<OperatorID, ExecutionJobVertex> vertices =
+ buildVertices(opIdWithParallelism, RANGE, ROUND_ROBIN);
+
+ // Run state assignment
+ new StateAssignmentOperation(0, new HashSet<>(vertices.values()), states, false)
+ .assignStates();
+
+ // Check results
+ ExecutionJobVertex upstreamExecutionJobVertex = vertices.get(operatorIds.get(0));
+ ExecutionJobVertex downstreamExecutionJobVertex = vertices.get(operatorIds.get(1));
+
+ List<TaskStateSnapshot> upstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(upstreamExecutionJobVertex);
+ List<TaskStateSnapshot> downstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(downstreamExecutionJobVertex);
+
+ checkMappings(
+ upstreamRescalingDescriptors,
+ TaskStateSnapshot::getOutputRescalingDescriptor,
+ expectedUpstreamCount);
+
+ checkMappings(
+ downstreamRescalingDescriptors,
+ TaskStateSnapshot::getInputRescalingDescriptor,
+ expectedDownstreamCount);
Review Comment:
Hm, but what is this improving?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] StefanRRichter commented on a diff in pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "StefanRRichter (via GitHub)" <gi...@apache.org>.
StefanRRichter commented on code in PR #22584:
URL: https://github.com/apache/flink/pull/22584#discussion_r1193845132
##########
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperationTest.java:
##########
@@ -785,6 +789,130 @@ public void testOnlyUpstreamChannelStateAssignment()
}
}
+ /** FLINK-31963: Tests rescaling for stateless operators and upstream result partition state. */
+ @Test
+ public void testOnlyUpstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState upstreamOpState =
+ OperatorSubtaskState.builder()
+ .setResultSubpartitionState(
+ new StateObjectCollection<>(
+ asList(
+ createNewResultSubpartitionStateHandle(10, random),
+ createNewResultSubpartitionStateHandle(
+ 10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(upstreamOpState, null, 5, 7);
+ }
+
+ /** FLINK-31963: Tests rescaling for stateless operators and downstream input channel state. */
+ @Test
+ public void testOnlyDownstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState downstreamOpState =
+ OperatorSubtaskState.builder()
+ .setInputChannelState(
+ new StateObjectCollection<>(
+ asList(
+ createNewInputChannelStateHandle(10, random),
+ createNewInputChannelStateHandle(10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(null, downstreamOpState, 5, 5);
+ }
+
+ private void testOnlyUpstreamOrDownstreamRescalingInternal(
+ @Nullable OperatorSubtaskState upstreamOpState,
+ @Nullable OperatorSubtaskState downstreamOpState,
+ int expectedUpstreamCount,
+ int expectedDownstreamCount)
+ throws JobException, JobExecutionException {
+
+ if ((upstreamOpState == null && downstreamOpState == null)
+ || (upstreamOpState != null && downstreamOpState != null)) {
+ // Either upstream or downstream state must exist, but not both.
+ return;
+ }
+
+ // Start from parallelism 5 for both operators
+ int upstreamParallelism = 5;
+ int downstreamParallelism = 5;
+
+ // Build states
+ List<OperatorID> operatorIds = buildOperatorIds(2);
+ Map<OperatorID, OperatorState> states = new HashMap<>();
+ OperatorState upstreamState =
+ new OperatorState(operatorIds.get(0), upstreamParallelism, MAX_P);
+ OperatorState downstreamState =
+ new OperatorState(operatorIds.get(1), downstreamParallelism, MAX_P);
+
+ states.put(operatorIds.get(0), upstreamState);
+ states.put(operatorIds.get(1), downstreamState);
+
+ if (upstreamOpState != null) {
+ upstreamState.putState(0, upstreamOpState);
+ // rescale downstream 5 -> 3
+ downstreamParallelism = 3;
+ }
+
+ if (downstreamOpState != null) {
+ downstreamState.putState(0, downstreamOpState);
+ // rescale upstream 5 -> 3
+ upstreamParallelism = 3;
+ }
+
+ List<OperatorIdWithParallelism> opIdWithParallelism = new ArrayList<>(2);
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(0), upstreamParallelism));
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(1), downstreamParallelism));
+
+ Map<OperatorID, ExecutionJobVertex> vertices =
+ buildVertices(opIdWithParallelism, RANGE, ROUND_ROBIN);
+
+ // Run state assignment
+ new StateAssignmentOperation(0, new HashSet<>(vertices.values()), states, false)
+ .assignStates();
+
+ // Check results
+ ExecutionJobVertex upstreamExecutionJobVertex = vertices.get(operatorIds.get(0));
+ ExecutionJobVertex downstreamExecutionJobVertex = vertices.get(operatorIds.get(1));
+
+ List<TaskStateSnapshot> upstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(upstreamExecutionJobVertex);
+ List<TaskStateSnapshot> downstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(downstreamExecutionJobVertex);
+
+ checkMappings(
+ upstreamRescalingDescriptors,
+ TaskStateSnapshot::getOutputRescalingDescriptor,
+ expectedUpstreamCount);
+
+ checkMappings(
+ downstreamRescalingDescriptors,
+ TaskStateSnapshot::getInputRescalingDescriptor,
+ expectedDownstreamCount);
+ }
+
+ private void checkMappings(
+ List<TaskStateSnapshot> taskStateSnapshots,
+ Function<TaskStateSnapshot, InflightDataRescalingDescriptor> extractFun,
+ int expectedCount) {
+ Assert.assertEquals(
+ expectedCount,
+ taskStateSnapshots.stream()
+ .map(extractFun)
+ .mapToInt(
+ x -> {
+ int len = x.getOldSubtaskIndexes(0).length;
+ // Assert that there is a mapping.
+ Assert.assertTrue(len > 0);
+ return len;
+ })
+ .sum());
Review Comment:
I was thinking about this and decided to keep the test targeted at just checking that a remapping has happened. I'd hope there are already tests that check the correctness of such remappings thoroughly.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] pnowojski commented on a diff in pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "pnowojski (via GitHub)" <gi...@apache.org>.
pnowojski commented on code in PR #22584:
URL: https://github.com/apache/flink/pull/22584#discussion_r1193934653
##########
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperationTest.java:
##########
@@ -785,6 +789,130 @@ public void testOnlyUpstreamChannelStateAssignment()
}
}
+ /** FLINK-31963: Tests rescaling for stateless operators and upstream result partition state. */
+ @Test
+ public void testOnlyUpstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState upstreamOpState =
+ OperatorSubtaskState.builder()
+ .setResultSubpartitionState(
+ new StateObjectCollection<>(
+ asList(
+ createNewResultSubpartitionStateHandle(10, random),
+ createNewResultSubpartitionStateHandle(
+ 10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(upstreamOpState, null, 5, 7);
+ }
+
+ /** FLINK-31963: Tests rescaling for stateless operators and downstream input channel state. */
+ @Test
+ public void testOnlyDownstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState downstreamOpState =
+ OperatorSubtaskState.builder()
+ .setInputChannelState(
+ new StateObjectCollection<>(
+ asList(
+ createNewInputChannelStateHandle(10, random),
+ createNewInputChannelStateHandle(10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(null, downstreamOpState, 5, 5);
+ }
+
+ private void testOnlyUpstreamOrDownstreamRescalingInternal(
+ @Nullable OperatorSubtaskState upstreamOpState,
+ @Nullable OperatorSubtaskState downstreamOpState,
+ int expectedUpstreamCount,
+ int expectedDownstreamCount)
+ throws JobException, JobExecutionException {
+
+ if ((upstreamOpState == null && downstreamOpState == null)
+ || (upstreamOpState != null && downstreamOpState != null)) {
+ // Either upstream or downstream state must exist, but not both.
+ return;
+ }
+
+ // Start from parallelism 5 for both operators
+ int upstreamParallelism = 5;
+ int downstreamParallelism = 5;
+
+ // Build states
+ List<OperatorID> operatorIds = buildOperatorIds(2);
+ Map<OperatorID, OperatorState> states = new HashMap<>();
+ OperatorState upstreamState =
+ new OperatorState(operatorIds.get(0), upstreamParallelism, MAX_P);
+ OperatorState downstreamState =
+ new OperatorState(operatorIds.get(1), downstreamParallelism, MAX_P);
+
+ states.put(operatorIds.get(0), upstreamState);
+ states.put(operatorIds.get(1), downstreamState);
+
+ if (upstreamOpState != null) {
+ upstreamState.putState(0, upstreamOpState);
+ // rescale downstream 5 -> 3
+ downstreamParallelism = 3;
+ }
+
+ if (downstreamOpState != null) {
+ downstreamState.putState(0, downstreamOpState);
+ // rescale upstream 5 -> 3
+ upstreamParallelism = 3;
+ }
+
+ List<OperatorIdWithParallelism> opIdWithParallelism = new ArrayList<>(2);
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(0), upstreamParallelism));
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(1), downstreamParallelism));
+
+ Map<OperatorID, ExecutionJobVertex> vertices =
+ buildVertices(opIdWithParallelism, RANGE, ROUND_ROBIN);
+
+ // Run state assignment
+ new StateAssignmentOperation(0, new HashSet<>(vertices.values()), states, false)
+ .assignStates();
+
+ // Check results
+ ExecutionJobVertex upstreamExecutionJobVertex = vertices.get(operatorIds.get(0));
+ ExecutionJobVertex downstreamExecutionJobVertex = vertices.get(operatorIds.get(1));
+
+ List<TaskStateSnapshot> upstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(upstreamExecutionJobVertex);
+ List<TaskStateSnapshot> downstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(downstreamExecutionJobVertex);
+
+ checkMappings(
+ upstreamRescalingDescriptors,
+ TaskStateSnapshot::getOutputRescalingDescriptor,
+ expectedUpstreamCount);
+
+ checkMappings(
+ downstreamRescalingDescriptors,
+ TaskStateSnapshot::getInputRescalingDescriptor,
+ expectedDownstreamCount);
Review Comment:
When reading the code of `checkMappings` it's tricky to understand what does the `extractFn` do. But feel free to ignore this comment.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [flink] pnowojski commented on a diff in pull request #22584: [FLINK-31963][state] Fix rescaling bug in recovery from unaligned checkpoints.
Posted by "pnowojski (via GitHub)" <gi...@apache.org>.
pnowojski commented on code in PR #22584:
URL: https://github.com/apache/flink/pull/22584#discussion_r1193812795
##########
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperationTest.java:
##########
@@ -785,6 +789,130 @@ public void testOnlyUpstreamChannelStateAssignment()
}
}
+ /** FLINK-31963: Tests rescaling for stateless operators and upstream result partition state. */
+ @Test
+ public void testOnlyUpstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState upstreamOpState =
+ OperatorSubtaskState.builder()
+ .setResultSubpartitionState(
+ new StateObjectCollection<>(
+ asList(
+ createNewResultSubpartitionStateHandle(10, random),
+ createNewResultSubpartitionStateHandle(
+ 10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(upstreamOpState, null, 5, 7);
+ }
+
+ /** FLINK-31963: Tests rescaling for stateless operators and downstream input channel state. */
+ @Test
+ public void testOnlyDownstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState downstreamOpState =
+ OperatorSubtaskState.builder()
+ .setInputChannelState(
+ new StateObjectCollection<>(
+ asList(
+ createNewInputChannelStateHandle(10, random),
+ createNewInputChannelStateHandle(10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(null, downstreamOpState, 5, 5);
+ }
+
+ private void testOnlyUpstreamOrDownstreamRescalingInternal(
+ @Nullable OperatorSubtaskState upstreamOpState,
+ @Nullable OperatorSubtaskState downstreamOpState,
+ int expectedUpstreamCount,
+ int expectedDownstreamCount)
+ throws JobException, JobExecutionException {
+
+ if ((upstreamOpState == null && downstreamOpState == null)
+ || (upstreamOpState != null && downstreamOpState != null)) {
+ // Either upstream or downstream state must exist, but not both.
+ return;
+ }
+
+ // Start from parallelism 5 for both operators
+ int upstreamParallelism = 5;
+ int downstreamParallelism = 5;
+
+ // Build states
+ List<OperatorID> operatorIds = buildOperatorIds(2);
+ Map<OperatorID, OperatorState> states = new HashMap<>();
+ OperatorState upstreamState =
+ new OperatorState(operatorIds.get(0), upstreamParallelism, MAX_P);
+ OperatorState downstreamState =
+ new OperatorState(operatorIds.get(1), downstreamParallelism, MAX_P);
+
+ states.put(operatorIds.get(0), upstreamState);
+ states.put(operatorIds.get(1), downstreamState);
+
+ if (upstreamOpState != null) {
+ upstreamState.putState(0, upstreamOpState);
+ // rescale downstream 5 -> 3
+ downstreamParallelism = 3;
+ }
+
+ if (downstreamOpState != null) {
+ downstreamState.putState(0, downstreamOpState);
+ // rescale upstream 5 -> 3
+ upstreamParallelism = 3;
+ }
+
+ List<OperatorIdWithParallelism> opIdWithParallelism = new ArrayList<>(2);
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(0), upstreamParallelism));
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(1), downstreamParallelism));
+
+ Map<OperatorID, ExecutionJobVertex> vertices =
+ buildVertices(opIdWithParallelism, RANGE, ROUND_ROBIN);
+
+ // Run state assignment
+ new StateAssignmentOperation(0, new HashSet<>(vertices.values()), states, false)
+ .assignStates();
+
+ // Check results
+ ExecutionJobVertex upstreamExecutionJobVertex = vertices.get(operatorIds.get(0));
+ ExecutionJobVertex downstreamExecutionJobVertex = vertices.get(operatorIds.get(1));
+
+ List<TaskStateSnapshot> upstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(upstreamExecutionJobVertex);
+ List<TaskStateSnapshot> downstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(downstreamExecutionJobVertex);
+
+ checkMappings(
+ upstreamRescalingDescriptors,
+ TaskStateSnapshot::getOutputRescalingDescriptor,
+ expectedUpstreamCount);
+
+ checkMappings(
+ downstreamRescalingDescriptors,
+ TaskStateSnapshot::getInputRescalingDescriptor,
+ expectedDownstreamCount);
+ }
+
+ private void checkMappings(
+ List<TaskStateSnapshot> taskStateSnapshots,
+ Function<TaskStateSnapshot, InflightDataRescalingDescriptor> extractFun,
+ int expectedCount) {
+ Assert.assertEquals(
+ expectedCount,
+ taskStateSnapshots.stream()
+ .map(extractFun)
+ .mapToInt(
+ x -> {
+ int len = x.getOldSubtaskIndexes(0).length;
+ // Assert that there is a mapping.
+ Assert.assertTrue(len > 0);
+ return len;
+ })
+ .sum());
Review Comment:
Instead of asserting length of the mappings, should we assert the actual mappings? 🤔
##########
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperationTest.java:
##########
@@ -785,6 +789,130 @@ public void testOnlyUpstreamChannelStateAssignment()
}
}
+ /** FLINK-31963: Tests rescaling for stateless operators and upstream result partition state. */
+ @Test
+ public void testOnlyUpstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState upstreamOpState =
+ OperatorSubtaskState.builder()
+ .setResultSubpartitionState(
+ new StateObjectCollection<>(
+ asList(
+ createNewResultSubpartitionStateHandle(10, random),
+ createNewResultSubpartitionStateHandle(
+ 10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(upstreamOpState, null, 5, 7);
+ }
+
+ /** FLINK-31963: Tests rescaling for stateless operators and downstream input channel state. */
+ @Test
+ public void testOnlyDownstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState downstreamOpState =
+ OperatorSubtaskState.builder()
+ .setInputChannelState(
+ new StateObjectCollection<>(
+ asList(
+ createNewInputChannelStateHandle(10, random),
+ createNewInputChannelStateHandle(10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(null, downstreamOpState, 5, 5);
+ }
+
+ private void testOnlyUpstreamOrDownstreamRescalingInternal(
+ @Nullable OperatorSubtaskState upstreamOpState,
+ @Nullable OperatorSubtaskState downstreamOpState,
+ int expectedUpstreamCount,
+ int expectedDownstreamCount)
+ throws JobException, JobExecutionException {
+
+ if ((upstreamOpState == null && downstreamOpState == null)
+ || (upstreamOpState != null && downstreamOpState != null)) {
+ // Either upstream or downstream state must exist, but not both.
+ return;
+ }
+
+ // Start from parallelism 5 for both operators
+ int upstreamParallelism = 5;
+ int downstreamParallelism = 5;
+
+ // Build states
+ List<OperatorID> operatorIds = buildOperatorIds(2);
+ Map<OperatorID, OperatorState> states = new HashMap<>();
+ OperatorState upstreamState =
+ new OperatorState(operatorIds.get(0), upstreamParallelism, MAX_P);
+ OperatorState downstreamState =
+ new OperatorState(operatorIds.get(1), downstreamParallelism, MAX_P);
+
+ states.put(operatorIds.get(0), upstreamState);
+ states.put(operatorIds.get(1), downstreamState);
+
+ if (upstreamOpState != null) {
+ upstreamState.putState(0, upstreamOpState);
+ // rescale downstream 5 -> 3
+ downstreamParallelism = 3;
+ }
+
+ if (downstreamOpState != null) {
+ downstreamState.putState(0, downstreamOpState);
+ // rescale upstream 5 -> 3
+ upstreamParallelism = 3;
+ }
+
+ List<OperatorIdWithParallelism> opIdWithParallelism = new ArrayList<>(2);
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(0), upstreamParallelism));
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(1), downstreamParallelism));
+
+ Map<OperatorID, ExecutionJobVertex> vertices =
+ buildVertices(opIdWithParallelism, RANGE, ROUND_ROBIN);
+
+ // Run state assignment
+ new StateAssignmentOperation(0, new HashSet<>(vertices.values()), states, false)
+ .assignStates();
+
+ // Check results
+ ExecutionJobVertex upstreamExecutionJobVertex = vertices.get(operatorIds.get(0));
+ ExecutionJobVertex downstreamExecutionJobVertex = vertices.get(operatorIds.get(1));
+
+ List<TaskStateSnapshot> upstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(upstreamExecutionJobVertex);
+ List<TaskStateSnapshot> downstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(downstreamExecutionJobVertex);
Review Comment:
Rename `RescalingDescriptors` -> `TaskStateSnapshots`? You are obtaining descriptors from the `TaskStateSnapshot` in the next step within `checkMappings`.
##########
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperationTest.java:
##########
@@ -785,6 +789,130 @@ public void testOnlyUpstreamChannelStateAssignment()
}
}
+ /** FLINK-31963: Tests rescaling for stateless operators and upstream result partition state. */
+ @Test
+ public void testOnlyUpstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState upstreamOpState =
+ OperatorSubtaskState.builder()
+ .setResultSubpartitionState(
+ new StateObjectCollection<>(
+ asList(
+ createNewResultSubpartitionStateHandle(10, random),
+ createNewResultSubpartitionStateHandle(
+ 10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(upstreamOpState, null, 5, 7);
+ }
+
+ /** FLINK-31963: Tests rescaling for stateless operators and downstream input channel state. */
+ @Test
+ public void testOnlyDownstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState downstreamOpState =
+ OperatorSubtaskState.builder()
+ .setInputChannelState(
+ new StateObjectCollection<>(
+ asList(
+ createNewInputChannelStateHandle(10, random),
+ createNewInputChannelStateHandle(10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(null, downstreamOpState, 5, 5);
+ }
+
+ private void testOnlyUpstreamOrDownstreamRescalingInternal(
+ @Nullable OperatorSubtaskState upstreamOpState,
+ @Nullable OperatorSubtaskState downstreamOpState,
+ int expectedUpstreamCount,
+ int expectedDownstreamCount)
+ throws JobException, JobExecutionException {
+
+ if ((upstreamOpState == null && downstreamOpState == null)
+ || (upstreamOpState != null && downstreamOpState != null)) {
+ // Either upstream or downstream state must exist, but not both.
+ return;
+ }
Review Comment:
`checkArgument(..., "Either upstream or downstream state must exist, but not both")`?
##########
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperationTest.java:
##########
@@ -785,6 +789,130 @@ public void testOnlyUpstreamChannelStateAssignment()
}
}
+ /** FLINK-31963: Tests rescaling for stateless operators and upstream result partition state. */
+ @Test
+ public void testOnlyUpstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState upstreamOpState =
+ OperatorSubtaskState.builder()
+ .setResultSubpartitionState(
+ new StateObjectCollection<>(
+ asList(
+ createNewResultSubpartitionStateHandle(10, random),
+ createNewResultSubpartitionStateHandle(
+ 10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(upstreamOpState, null, 5, 7);
+ }
+
+ /** FLINK-31963: Tests rescaling for stateless operators and downstream input channel state. */
+ @Test
+ public void testOnlyDownstreamChannelRescaleStateAssignment()
+ throws JobException, JobExecutionException {
+ Random random = new Random();
+ OperatorSubtaskState downstreamOpState =
+ OperatorSubtaskState.builder()
+ .setInputChannelState(
+ new StateObjectCollection<>(
+ asList(
+ createNewInputChannelStateHandle(10, random),
+ createNewInputChannelStateHandle(10, random))))
+ .build();
+ testOnlyUpstreamOrDownstreamRescalingInternal(null, downstreamOpState, 5, 5);
+ }
+
+ private void testOnlyUpstreamOrDownstreamRescalingInternal(
+ @Nullable OperatorSubtaskState upstreamOpState,
+ @Nullable OperatorSubtaskState downstreamOpState,
+ int expectedUpstreamCount,
+ int expectedDownstreamCount)
+ throws JobException, JobExecutionException {
+
+ if ((upstreamOpState == null && downstreamOpState == null)
+ || (upstreamOpState != null && downstreamOpState != null)) {
+ // Either upstream or downstream state must exist, but not both.
+ return;
+ }
+
+ // Start from parallelism 5 for both operators
+ int upstreamParallelism = 5;
+ int downstreamParallelism = 5;
+
+ // Build states
+ List<OperatorID> operatorIds = buildOperatorIds(2);
+ Map<OperatorID, OperatorState> states = new HashMap<>();
+ OperatorState upstreamState =
+ new OperatorState(operatorIds.get(0), upstreamParallelism, MAX_P);
+ OperatorState downstreamState =
+ new OperatorState(operatorIds.get(1), downstreamParallelism, MAX_P);
+
+ states.put(operatorIds.get(0), upstreamState);
+ states.put(operatorIds.get(1), downstreamState);
+
+ if (upstreamOpState != null) {
+ upstreamState.putState(0, upstreamOpState);
+ // rescale downstream 5 -> 3
+ downstreamParallelism = 3;
+ }
+
+ if (downstreamOpState != null) {
+ downstreamState.putState(0, downstreamOpState);
+ // rescale upstream 5 -> 3
+ upstreamParallelism = 3;
+ }
+
+ List<OperatorIdWithParallelism> opIdWithParallelism = new ArrayList<>(2);
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(0), upstreamParallelism));
+ opIdWithParallelism.add(
+ new OperatorIdWithParallelism(operatorIds.get(1), downstreamParallelism));
+
+ Map<OperatorID, ExecutionJobVertex> vertices =
+ buildVertices(opIdWithParallelism, RANGE, ROUND_ROBIN);
+
+ // Run state assignment
+ new StateAssignmentOperation(0, new HashSet<>(vertices.values()), states, false)
+ .assignStates();
+
+ // Check results
+ ExecutionJobVertex upstreamExecutionJobVertex = vertices.get(operatorIds.get(0));
+ ExecutionJobVertex downstreamExecutionJobVertex = vertices.get(operatorIds.get(1));
+
+ List<TaskStateSnapshot> upstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(upstreamExecutionJobVertex);
+ List<TaskStateSnapshot> downstreamRescalingDescriptors =
+ getRescalingDescriptorsFromVertex(downstreamExecutionJobVertex);
+
+ checkMappings(
+ upstreamRescalingDescriptors,
+ TaskStateSnapshot::getOutputRescalingDescriptor,
+ expectedUpstreamCount);
+
+ checkMappings(
+ downstreamRescalingDescriptors,
+ TaskStateSnapshot::getInputRescalingDescriptor,
+ expectedDownstreamCount);
Review Comment:
nit: instead of lambda functions I would accept a little bit of code deduplication and replace those calls with:
```
checkMappings(
upstreamTaskStateSnapshots.stream().map(TaskStateSnapshot::getOutputRescalingDescriptor),
expectedUpstreamCount);
checkMappings(
downstreamTaskStateSnapshots.stream().map(TaskStateSnapshot::getInputRescalingDescriptor),
expectedDownstreamCount);
```
##########
flink-tests/src/test/java/org/apache/flink/test/checkpointing/UnalignedCheckpointRescaleITCase.java:
##########
@@ -459,46 +478,59 @@ public void processBroadcastElement(Long value, Context ctx, Collector<Long> out
}
}
- @Parameterized.Parameters(name = "{0} {1} from {2} to {3}, buffersPerChannel = {4}")
+ @Parameterized.Parameters(
+ name = "{0} {1} from {2} to {3}, sourceSleepMs = {4}, buffersPerChannel = {5}")
Review Comment:
I would add a comment above this line explaining why do we want to have non zero `sourceSleepMs` sometimes. That we want to test the rescaling without backpressure with only occasional a couple of captured in-flight records .
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org