[PR] Fix flaky datanode UTs under parallel surefire forks [iotdb]

["JackieTien97 (via GitHub)" <gi...@apache.org>]

JackieTien97 opened a new pull request, #17712:
URL: https://github.com/apache/iotdb/pull/17712

   ## Summary
   
   Fixes two intermittent failures in `ut-datanode` after #17698 enabled `forkCount=4`:
   
   1. **`DataRegionTest.tearDown:165` — `failed to close some ports`**
      `EnvironmentUtils.examinePorts()` checks 6667 / 5555 / 31999 / 9091 machine-wide via `connect()`. `DataNodeInternalRPCServiceImplTest` legitimately binds 6667 through `DataRegionConsensusImpl.start()` → `IoTConsensusRPCService`. A sibling fork's tearDown then sees 6667 open and fails — a cross-fork false positive, not a real leak.
      
      #17698 missed this because its grep for `ServerSocket / bind() / TServer.serve()` doesn't catch the binding done indirectly via `IConsensus.start()`.
   
   2. **`AlignedSeriesScanOperatorTest.batchTest2:423` — NPE on null TsBlock**
      `FullOuterTimeJoinOperator.next()` (and `AbstractSeriesScanOperator.next()`) legitimately return `null` or an empty TsBlock when `prepareInput()` yields on its 500ms `maxRunTime` slice. The test dereferenced `tsBlock` immediately.
   
   ## Changes
   
   - **`EnvironmentUtils.FORK_PORT_OFFSET = ((surefire.forkNumber - 1) % 30 + 1) * 1000`**
     Each fork shifts its bound ports **and** the corresponding `examinePorts()` check by this offset, so forks live in disjoint port namespaces. `examinePorts()` in fork A only sees ports bound by fork A (real leak detection preserved); fork A no longer sees fork B's bindings (cross-fork false positive removed). Modulo 30 caps the offset so the highest checked port (31999 + 30*1000 = 61999) stays valid.
   
   - **`DataNodeInternalRPCServiceImplTest`** binds `6667 + FORK_PORT_OFFSET` via the same constant — binding and check stay in sync.
   
   - **`AlignedSeriesScanOperatorTest.batchTest1/2/3`** add `if (tsBlock == null || tsBlock.isEmpty()) continue;` after each `operator.next()`, matching the pattern already used in `TreeTopKOperatorTest`, `MergeTreeSortOperatorTest`, `TreeSortOperatorTest`, etc.
   
   Audited `iotdb-core/datanode` UTs for other monitored-port bindings (`grep -rE "(\W|^)(6667|5555|31999|9091)(\W|$)"`) — no other test binds any of these ports. The only real binder is `DataNodeInternalRPCServiceImplTest`, which is updated here.
   
   ## Test plan
   
   - [ ] Spotless passes locally (`mvn spotless:check -pl iotdb-core/datanode`)
   - [ ] `ut-datanode` CI job runs without "failed to close some ports" failures
   - [ ] `AlignedSeriesScanOperatorTest.batchTest2` no longer NPEs intermittently
   - [ ] `DataNodeInternalRPCServiceImplTest` still passes (binding now on fork-specific port)


-- 
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: reviews-unsubscribe@iotdb.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Fix flaky datanode UTs under parallel surefire forks [iotdb]

["JackieTien97 (via GitHub)" <gi...@apache.org>]

JackieTien97 merged PR #17712:
URL: https://github.com/apache/iotdb/pull/17712


-- 
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: reviews-unsubscribe@iotdb.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org