TEZ-4725: Fix flaky tests in TestAMRecoveryAggregationBroadcast#520
Open
maheshrajus wants to merge 1 commit into
Open
TEZ-4725: Fix flaky tests in TestAMRecoveryAggregationBroadcast#520maheshrajus wants to merge 1 commit into
maheshrajus wants to merge 1 commit into
Conversation
|
🎊 +1 overall
This message was automatically generated. |
abstractdog
requested changes
Jul 22, 2026
Comment on lines
+252
to
+253
| TezCounters counters = runDAGAndVerify(dag, true, | ||
| Collections.singletonList(TABLE_SCAN)); |
Contributor
There was a problem hiding this comment.
line break is not needed here
| private void waitForVertexSucceeded(DAGClient dagClient, String vertexName, | ||
| long timeoutMs) throws Exception { | ||
| long deadline = System.currentTimeMillis() + timeoutMs; | ||
| while (System.currentTimeMillis() < deadline) { |
Contributor
There was a problem hiding this comment.
try to use a monothonic clock instead of System.currentTimeMillis()
Comment on lines
+275
to
+276
| TezCounters counters = runDAGAndVerify(dag, true, | ||
| Arrays.asList(TABLE_SCAN, AGGREGATION)); |
Contributor
There was a problem hiding this comment.
line break is not needed here
Comment on lines
+363
to
+364
| waitForVertexSucceeded(dagClient, vertexName, | ||
| TimeUnit.SECONDS.toMillis(60)); |
Contributor
There was a problem hiding this comment.
line break is not needed here
Comment on lines
+395
to
+404
| if (state == VertexStatus.State.SUCCEEDED) { | ||
| return; | ||
| } | ||
| if (state == VertexStatus.State.FAILED | ||
| || state == VertexStatus.State.KILLED | ||
| || state == VertexStatus.State.ERROR) { | ||
| throw new AssertionError("Vertex " + vertexName | ||
| + " reached terminal non-success state: " + state); | ||
| } | ||
| } |
Contributor
There was a problem hiding this comment.
what about a switch case here, something like
switch (lastState) {
case SUCCEEDED -> { return; }
case FAILED, KILLED, ERROR -> throw new AssertionError("Vertex " + vertexName
+ " reached terminal non-success state: " + lastState);
...
}
| // and retry rather than crash. | ||
| String amHost = appReport.getHost(); | ||
| int amRpcPort = appReport.getRpcPort(); | ||
| if (amHost == null || amHost.equals("N/A") || amRpcPort <= 0) { |
Contributor
There was a problem hiding this comment.
I think this condition can be refactored to a separate method with appReport parameter and be reused across different codepaths
Comment on lines
+964
to
+974
| // YARN-808 gap: when the AM container is first allocated YARN briefly | ||
| // reports state=RUNNING before the AM has registered with the RM or bound | ||
| // its RPC listener. During that window the ApplicationReport contains | ||
| // sentinel values that must not be passed to | ||
| // NetUtils.createSocketAddrForHost() (which would throw | ||
| // IllegalArgumentException: port out of range) or to RPC.getProxy(): | ||
| // host == null / "N/A" : RM has not received registerApplicationMaster() | ||
| // rpcPort == 0 : protobuf wire default | ||
| // rpcPort == -1 : container up but RPC listener not yet bound | ||
| // Returning null lets callers (waitForProxy, sendAMHeartbeat) back off | ||
| // and retry rather than crash. |
Contributor
There was a problem hiding this comment.
this long comment can be put to a new method that contains the refactored condition below
except the "return null lets callers" part, which belongs to this method
Comment on lines
+283
to
+287
| // Workaround: check the default strings/sentinels set by YARN. | ||
| // port == 0 : protobuf wire default, AM has not called | ||
| // registerApplicationMaster() yet. | ||
| // port == -1 : AM container is allocated (state=RUNNING) but the | ||
| // RPC listener has not been bound yet. |
Contributor
There was a problem hiding this comment.
this comment can be put to the new method that contains the refactored condition below
Comment on lines
+754
to
+761
| private DAGClientRPCImplWithFakeReport mockReportClient(TezConfiguration conf, | ||
| String host, int rpcPort) throws IOException { | ||
| ApplicationReport report = mock(ApplicationReport.class); | ||
| when(report.getYarnApplicationState()).thenReturn(YarnApplicationState.RUNNING); | ||
| when(report.getHost()).thenReturn(host); | ||
| when(report.getRpcPort()).thenReturn(rpcPort); | ||
| return new DAGClientRPCImplWithFakeReport(mockAppId, dagIdStr, conf, report); | ||
| } |
Contributor
There was a problem hiding this comment.
it's strange that mocking and real classes are mixed: if there is a real DAGClientRPCImplWithFakeReport, it could leverage a real ApplicationReport, couldn't it?
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Root causes and fixes:
TestAMRecoveryAggregationBroadcast.testMapJoinTemporalFailure (race condition)
Replace fixed Thread.sleep(10s) before AM kill with a deterministic
waitForVertexSucceeded() helper that polls DAGClient.getVertexStatus()
every 500ms (up to 60s) until the target vertices reach SUCCEEDED state.
Each test now waits for only the vertices it logically depends on before
killing the AM, ensuring the recovery log assertions always see the
expected counts.
Make OUT_PATH unique per test run (random suffix) to eliminate cross-test.
DAGClientRPCImpl / TezClientUtils: port out of range:-1 (YARN-808 gap)
YARN sets rpcPort=-1 when an AM container is allocated (state=RUNNING)
but the AM has not yet bound its RPC listener. The existing guard only
checked rpcPort==0 (protobuf default), so rpcPort==-1 reached
NetUtils.createSocketAddrForHost(), which threw
IllegalArgumentException: port out of range:-1.
Fix DAGClientRPCImpl.createAMProxyIfNeeded(): rpcPort == 0 → rpcPort <= 0.
Fix TezClientUtils.getAMProxy(FrameworkClient,...): add the same
rpcPort <= 0 guard