fix(tracing): disable autoflush on async_sessionmaker to prevent span FK violations

[ashutoshdharibm]

Summary

SQLAlchemy's default autoflush=True causes implicit flushes between individual session.merge() calls inside _flush_to_database(). When this fires, a child span is written to the DB before its parent span, producing a FK violation on span.parent_span_id → span.id — even though topological_sort_spans() (added in #12242) has already sorted spans into the correct order.

The topological sort was correct all along. The problem is SQLAlchemy's autoflush defeating it mid-loop.

Root cause

In DatabaseService.__init__(), both async_sessionmaker() call-sites use the default autoflush=True. Inside _flush_to_database() in native.py:

async with session_scope() as session:
    for span in topological_sort_spans(spans):
        session.merge(span)        # ← autoflush can fire here mid-loop
        await session.flush()

When autoflush fires between two merge() calls, a child span whose parent hasn't been merged yet lands in the DB first → IntegrityError.

Fix

Pass autoflush=False to both async_sessionmaker() call-sites in DatabaseService.__init__(). No writes are lost because session_scope() calls await session.commit() explicitly on exit.

# before
self.async_session = async_sessionmaker(engine, expire_on_commit=False)

# after
self.async_session = async_sessionmaker(engine, expire_on_commit=False, autoflush=False)

Tests

4 regression tests added in test_autoflush_disabled.py:

Test	Purpose
test_async_session_maker_has_autoflush_false	Guards factory config
test_default_autoflush_is_true_without_fix	Documents the unsafe baseline
test_topo_sorted_merge_with_autoflush_false_does_not_raise	E2E: topological merge + FK constraint with autoflush disabled
test_session_maker_factory_produces_autoflush_false_sessions	Confirms produced sessions inherit the flag

All 4 pass locally.

Relationship to prior work

• fix: Avoid foreign key violation on span table with topological sort #12242 added topological_sort_spans() to fix this — it is correct but SQLAlchemy autoflush defeats it at runtime
• This PR completes the fix by preventing autoflush from firing mid-loop

Fixes: DSLF-524

Summary by CodeRabbit

• Bug Fixes

  • Improved database session handling to prevent foreign-key errors when saving related records.
  • Ensured parent and child span records can be merged and committed successfully.

• Tests

  • Added regression coverage for session configuration and span persistence.
  • Verified behavior with foreign-key enforcement enabled.

[coderabbitai]

Walkthrough

Async database sessionmakers now use autoflush=False during initialization and engine reload. Regression tests verify the setting and confirm parent-child span merges commit successfully with foreign-key enforcement.

Changes

Async session autoflush fix

Layer / File(s)	Summary
Sessionmaker configuration src/backend/base/langflow/services/database/service.py	DatabaseService sets autoflush=False for sessionmakers created during initialization and engine reload.
Foreign-key regression coverage src/backend/tests/unit/services/database/test_autoflush_disabled.py	Tests verify explicit and default autoflush behavior, configure foreign-key-enforced SQLite, and validate successful parent-child span persistence.

Estimated code review effort: 2 (Simple) | ~15 minutes

Suggested labels: bug

---

Important

Pre-merge checks failed

Please resolve all errors before merging. Addressing warnings is optional.

❌ Failed checks (1 error, 1 warning)

Check name	Status	Explanation	Resolution
Test Coverage For New Implementations	❌ Error	The new tests are real, but they recreate sessionmakers directly and never instantiate DatabaseService or call reload_engine, so the changed code isn’t covered.	Update tests to use a real DatabaseService fixture, assert its initial async_session_maker, then call reload_engine() and verify the rebuilt factory still has autoflush=False.
Test Quality And Coverage	⚠️ Warning	Tests mostly recreate async_sessionmaker manually; none instantiate DatabaseService or exercise reload_engine, so they don't guard the changed code paths.	Use a real DatabaseService fixture, assert its initial async_session_maker/autoflush, call reload_engine(), and assert the rebuilt factory still produces autoflush=False sessions.

✅ Passed checks (7 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main change: disabling autoflush on async_sessionmaker to avoid span foreign-key violations.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Test File Naming And Structure	✅ Passed	The backend test file is correctly named under unit tests, uses pytest classes/methods with clear names, has yield-based fixture cleanup, and covers both positive and negative cases.
Excessive Mock Usage Warning	✅ Passed	No mocks are used in the added database tests; they exercise real async SQLite engines and SQLModel sessions instead of mocked core logic.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

src/backend/tests/unit/services/database/test_autoflush_disabled.py (1)

134-135: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Remove the redundant asyncio marker.

pytest-asyncio auto mode already runs these async test methods. Based on learnings, pytest-asyncio is configured with asyncio_mode = 'auto'; tests should avoid unnecessary pytest.mark.asyncio decorators.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/backend/tests/unit/services/database/test_autoflush_disabled.py` around
lines 134 - 135, Remove the redundant `@pytest.mark.asyncio` decorator from the
TestAutoflushFKViolationRegression class, relying on the configured
pytest-asyncio auto mode while leaving the async test methods unchanged.

Source: Learnings

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/backend/tests/unit/services/database/test_autoflush_disabled.py`:
- Around line 58-64: Update
src/backend/tests/unit/services/database/test_autoflush_disabled.py at lines
58-64, 182-186, and 214-219 to test the real DatabaseService rather than
independently recreated async_sessionmaker instances: use
DatabaseService.async_session_maker for the initial-session assertions, run the
merge regression through that service-created factory, and call
DatabaseService.reload_engine() before asserting the replacement factory still
has autoflush=False.

---

Nitpick comments:
In `@src/backend/tests/unit/services/database/test_autoflush_disabled.py`:
- Around line 134-135: Remove the redundant `@pytest.mark.asyncio` decorator from
the TestAutoflushFKViolationRegression class, relying on the configured
pytest-asyncio auto mode while leaving the async test methods unchanged.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ffced404-848d-40e1-9ba0-e5ccfb0025bb

📥 Commits

Reviewing files that changed from the base of the PR and between c1030a3 and bccee43.

📒 Files selected for processing (2)

• src/backend/base/langflow/services/database/service.py
• src/backend/tests/unit/services/database/test_autoflush_disabled.py

[coderabbitai]

🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Test DatabaseService rather than recreated session factories.

These tests can all pass if DatabaseService.__init__() or reload_engine() stops passing autoflush=False. Use a real DatabaseService fixture, assert a session from its initial factory, call reload_engine(), then assert a session from the rebuilt factory.

• src/backend/tests/unit/services/database/test_autoflush_disabled.py#L58-L64: replace the standalone factory with the initial DatabaseService.async_session_maker.
• src/backend/tests/unit/services/database/test_autoflush_disabled.py#L182-L186: run the merge regression through the service-created session factory.
• src/backend/tests/unit/services/database/test_autoflush_disabled.py#L214-L219: exercise DatabaseService.reload_engine() and assert its replacement factory retains autoflush=False.

As per coding guidelines, new backend bug fixes must include tests that cover the changed behavior rather than placeholders.

📍 Affects 1 file

• src/backend/tests/unit/services/database/test_autoflush_disabled.py#L58-L64 (this comment)
• src/backend/tests/unit/services/database/test_autoflush_disabled.py#L182-L186
• src/backend/tests/unit/services/database/test_autoflush_disabled.py#L214-L219

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/backend/tests/unit/services/database/test_autoflush_disabled.py` around
lines 58 - 64, Update
src/backend/tests/unit/services/database/test_autoflush_disabled.py at lines
58-64, 182-186, and 214-219 to test the real DatabaseService rather than
independently recreated async_sessionmaker instances: use
DatabaseService.async_session_maker for the initial-session assertions, run the
merge regression through that service-created factory, and call
DatabaseService.reload_engine() before asserting the replacement factory still
has autoflush=False.

Source: Coding guidelines