morningman commented on PR #66729:
URL: https://github.com/apache/doris/pull/66729#issuecomment-5664896992
<!-- doris-repo-review:v1:begin -->
### Local pipeline review — ✅ PASS
```yaml
schema: doris-repo-review/v1
status: PASS
pr: apache/doris#66729
commit: 2dc1fc2b4eed6a3b592bfd367adbf9cfe261c39b
base: cfb817f9c38856b4759dd0c56172acc8f69a2261
reviewed_at: 2026-09-14T21:40+08:00
reviewer: morningman
model: claude-opus-5[1m]
effort: max
findings: {blocker: 0, major: 0, minor: 4, nit: 3}
rounds: 1
converged: true
```
**Notes for maintainers**
-
`fe/be-java-extensions/hadoop-hudi-scanner/src/main/java/org/apache/doris/hudi/HudiFileSystemScopes.java:218`
— the idle-scope sweep closes filesystems while holding hadoop's process-wide
`FileSystem.CACHE` monitor (the BE twin of the still-open FE finding on
`HudiConnector.closeAllForUGI`) and logs nothing on success; collecting under
the monitor and closing outside it in the Doris-owned `FileSystem` copy fixes
both sides at once.
-
`fe/be-java-extensions/hadoop-hudi-scanner/src/main/java/org/apache/doris/hudi/HudiFileSystemScopes.java:169`
— the sweeper's `catch (RuntimeException | LinkageError)` does not keep the
schedule alive through any other `Error`: `ScheduledThreadPoolExecutor` never
re-executes a periodic task whose run threw, so one OOM inside `sweep()`
silently ends eviction for the BE's life; nothing tests `startSweeper` at all.
-
`fe/be-java-extensions/hadoop-hudi-scanner/src/test/java/org/apache/doris/hudi/HadoopHudiPluginTest.java:301`
— no BE test pins that a scan reads under its hold's UGI (the identity
`closeAllForUGI` keys on): removing the `doAs` arm of
`executeInFileSystemScope` keeps all 18 hudi-module tests green while the sweep
closes nothing. The FE half has exactly this test
(`HudiScanPlanProviderScopeTest`); a `hadoop_conf.fs.file.impl` UGI-recording
`LocalFileSystem` gives the BE half for free.
-
`fe/fe-connector/fe-connector-hudi/src/main/java/org/apache/doris/connector/hudi/HudiScanPlanProvider.java:195`
— planning under the connector's scope fixes the inherited 11ff064 F-09, and
by the same token joins the close window `HudiConnector`'s class comment
documents: an ALTER/DROP CATALOG racing `planScan` now closes the filesystems
it is listing through ("Filesystem closed"); at the base planning was immune.
Documented and accepted, so Minor.
- Delta facts: 308 of the 350 paths are byte-identical to the `9609ff6`
head; the merge with 27 master commits was dropchecked (the 38 "missing" #67726
`HiveConnector` lines are provided by the PR's own
`throwIfClosed()`/`closeStage()`; master's JUnit 4 test was ported); the layout
gate and a `PluginRuntime` probe were reproduced on the HEAD-built jar sets. No
TeamCity line has run on this exact head yet (last `run buildall` was on
`eccac9dbad2`). The PR body and `paimon-scanner/pom.xml` still carry the jindo
single-bind caveat that `8be943ae155` retracted.
<sub>Reviewed locally with the `doris-repo-review` pipeline. Repository
policy may accept this receipt for the matching commit; it is not a human
Apache approval.</sub>
<!-- doris-repo-review:v1:end -->
--
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]
