bosswnx opened a new issue, #67403:
URL: https://github.com/apache/doris/issues/67403
### Summary
While running the full FE unit test suite locally (as part of validating
#67310, which merges latest master), I hit two test failures on **current
master itself**. Both are test-vs-code contradictions introduced by upstream
PRs landing in a conflicting order, and are unrelated to #67310 (that PR only
touches `MasterOpExecutor` / `FrontendServiceImpl` / thrift definitions, zero
overlap with these modules).
### Failure 1:
`IcebergWritePlanProviderTest.planMergePreservesExplicitlyEmptyReadAcrossConcurrentFirstAppend`
```
org.opentest4j.AssertionFailedError: an explicitly empty read must leave
RowDelta validation unbounded across the first append ==> expected: <null> but
was: <-1>
at org.apache.doris.connector.iceberg.IcebergWritePlanProviderTest
.planMergePreservesExplicitlyEmptyReadAcrossConcurrentFirstAppend(IcebergWritePlanProviderTest.java:1727)
```
- The test was added by #66345 (Aug 6) asserting `baseSnapshotId == null`
for an explicitly pinned empty read.
- #66348 (Aug 14) deliberately changed the semantics: an explicitly pinned
`-1` is kept as an OCC fence (`IcebergConnectorTransaction.applyBeginGuards`,
comment *"An explicitly pinned -1 is the empty-table generation and must remain
an OCC fence"*), and the `null`-arm was replaced by
`Long.valueOf(pinnedReadSnapshot)`.
- The old assertion was not updated, so this test has been failing on master
since Aug 14.
Reproduce: `cd fe && mvn test -pl fe-connector/fe-connector-iceberg -am
-Dtest=IcebergWritePlanProviderTest`
Suggested fix: update the assertion to `assertEquals(Long.valueOf(-1L),
txn.getBaseSnapshotId())` (matching the intent documented in #66348).
### Failure 2:
`PropertyAnalyzerTest.testAnalyzePartitionInvertedIndexFileStorageFormat`
```
org.opentest4j.AssertionFailedError: expected: <true> but was: <false>
at org.apache.doris.common.PropertyAnalyzerTest
.testAnalyzePartitionInvertedIndexFileStorageFormat(PropertyAnalyzerTest.java:412)
```
- The test was added by #66475 (Sep 1) and expects
`partition.inverted_index_storage_format = "V1"` to be rejected with the
partition-level message *"only supports V2, V3 and SNII"*.
- But #64522 (Aug 24) already made `analyzeInvertedIndexFileStorageFormat`
throw *"Inverted index V1 is deprecated and no longer allowed for new index
creation"* for any `v1` input — which fires **before** the partition-level `V1`
check, making that check (and the expected message) unreachable dead code.
Reproduce: `cd fe && mvn test -pl fe-core -am -Dtest=PropertyAnalyzerTest`
Suggested fix: either accept either rejection message in the assertion, or
have `analyzePartitionInvertedIndexFileStorageFormat` surface its own message
for `V1` before delegating to `analyzeInvertedIndexFileStorageFormat`.
### Environment
- Commit: `5fbe3600024` (master, Sep 1) — verified the failing test files
are identical to master (zero diff)
- JDK 17, Maven 3.9.16, single-fork surefire
- Failure 1 also reproduced on `d454100765c` (Aug 31 master)
Related PR: #67310 (where these were found during full local validation)
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
