[improvement](be) Limit packed file writes to rowset first segment

[liaoxin01]

What problem does this PR solve?

Issue Number: None

Related PR: None

Problem Summary: Only merge the first segment in a rowset into packed files. Later segments bypass packed files so large imports and memory-pressure-triggered multi-segment flushes do not keep feeding packed-file buffering.

Release note

None

Check List (For Author)

• Test: Unit Test / Manual test
  • Unit Test: Attempted ./run-be-ut.sh --run --filter=PackedFileSystemTest.* -j ${DORIS_PARALLELISM:-8},
  • Manual test: Ran clang++ -fsyntax-only for be/src/io/fs/packed_file_system.cpp and be/test/io/fs/packed_file_system_test.cpp
• Behavior changed: Yes (packed file now applies only to the first segment in a rowset)
• Does this need documentation: No

[Thearas]

Thank you for your contribution to Apache Doris.
Don't know what should be done next? See How to process your PR.

Please clearly describe your PR:

1. What problem was fixed (it's best to include specific error reporting information). How it was fixed.
2. Which behaviors were modified. What was the previous behavior, what is it now, why was it modified, and what possible impacts might there be.
3. What features were added. Why was this function added?
4. Which code was refactored and why was this part of the code refactored?
5. Which functions were optimized and what is the difference before and after the optimization?

[liaoxin01]

run buildall

[Copilot]

Pull request overview

This PR changes BE packed-file behavior so only the first segment of a rowset is eligible for packed-file aggregation, while later segments write directly to the underlying filesystem to avoid prolonged buffering during large imports or multi-segment flushes.

Changes:

• Add filename-based gating in PackedFileSystem::create_file_impl() to bypass PackedFileWriter for non-first segments.
• Introduce helper logic to detect “first segment” based on .dat/.idx filenames.
• Add unit tests asserting packed vs direct writer selection for segment 0 vs later segments.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
be/src/io/fs/packed_file_system.cpp	Adds should_use_packed_writer() and conditionally skips PackedFileWriter for later segments.
be/test/io/fs/packed_file_system_test.cpp	Adds tests to validate packed-writer usage for first segment and direct-writer usage for later segments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[liaoxin01]

run buildall

[hello-stephen]

BE UT Coverage Report

Increment line coverage 90.91% (20/22) 🎉

Increment coverage report
Complete coverage report

Category	Coverage
Function Coverage	53.03% (20143/37985)
Line Coverage	36.57% (189352/517849)
Region Coverage	32.82% (146949/447809)
Branch Coverage	33.96% (64357/189533)

[hello-stephen]

BE Regression && UT Coverage Report

Increment line coverage 90.91% (20/22) 🎉

Increment coverage report
Complete coverage report

Category	Coverage
Function Coverage	73.64% (27392/37199)
Line Coverage	57.30% (295836/516265)
Region Coverage	54.49% (246272/451941)
Branch Coverage	56.17% (106787/190115)

[liaoxin01]

/review

[github-actions]

Found 1 blocking issue.

1. be/src/io/fs/packed_file_system.cpp: should_use_packed_writer() only inspects file.filename(). For cloud StorageResource::remote_segment_path() with path_version == 1, segment paths are data/{shard}/{tablet_id}/{rowset_id}/{seg_id}.dat and V2 index paths are .../{rowset_id}/{seg_id}.idx. The basename becomes just 1.dat / 1.idx, so rfind('_') fails and the helper returns true, meaning later segments still get wrapped in PackedFileWriter. That leaves the multi-segment cloud path unchanged and the PR goal is not met on this supported path format. Please parse the segment-id decision from the full path or otherwise handle path_version == 1, and add a regression/unit test for that path shape.

Critical checkpoint conclusions:

• Goal and proof: Partially accomplished. The new tests prove the behavior only for local-style names; they do not cover the supported cloud remote path v1 format where the change currently fails.
• Change scope: Small and focused, but the filename heuristic is too narrow for all supported rowset path shapes.
• Concurrency: No new concurrency or lock-safety concerns in this patch.
• Lifecycle/static init: No special lifecycle or static initialization concerns introduced.
• Config: No configuration changes.
• Compatibility: No storage-format or symbol compatibility changes.
• Parallel paths: Not fully handled. Local/path_version0 names and cloud path_version1 names are functionally parallel here, but only the former is covered.
• Special conditions: The new conditional lacks coverage for the non-intuitive remote path layout.
• Test coverage: Added BE unit tests are useful but incomplete; missing remote path_version1 cases for both .dat and V2 .idx.
• Observability: Existing logging is sufficient; no additional observability appears necessary.
• Transaction/persistence: No direct persistence protocol change, but wrong writer selection changes whether packed-slice metadata is produced.
• Data write correctness: The intended write-path behavior change is not guaranteed for all supported cloud paths.
• FE/BE variable passing: Not applicable.
• Performance: The intended memory-usage improvement is not realized on the affected cloud path.
• Other issues: None beyond the path-format gap above.

V1 inverted-index tablets are excluded before that in be/src/storage/rowset/rowset_writer_context.h, where packed_file_active is forced off for V1.

[github-actions]

PR approved by at least one committer and no changes requested.

[github-actions]

PR approved by anyone and no changes requested.

[gavinchou]

LGTM