Skip to content

Speed up TaskGroup.topological_sort with int-indexed projected sweep - #67288

Merged
shahar1 merged 8 commits into
apache:mainfrom
shahar1:taskgroup-topo-sort-projection
May 27, 2026
Merged

Speed up TaskGroup.topological_sort with int-indexed projected sweep#67288
shahar1 merged 8 commits into
apache:mainfrom
shahar1:taskgroup-topo-sort-projection

Conversation

@shahar1

@shahar1 shahar1 commented May 21, 2026

Copy link
Copy Markdown
Contributor

Summary

Replaces the modified-Kahn body of TaskGroup.topological_sort (and the mirror in SerializedTaskGroup) with a projected sweep:

  • Each child's per-task upstream_task_ids is projected onto sibling-level integer indices once up front (_project_child_deps).
  • The sweep (_sweep_projection) then runs against the projection with a bytearray-backed emission flag - per-edge work happens once per sort instead of once per outer-loop pass.

Emission order is identical to the previous implementation. The existing order-sensitive tests (test_topological_sort1/2, test_topological_nested_groups, test_topological_group_dep) cover the contract.

Benchmark

Min of five runs, N=2000 children, ms per call. Source: https://gist.github.com/shahar1/9c61dc9f34f7e77cd29cfb9d67af7ceb

Shape main PR speedup
chain (forward-declared) 1.39 0.75 1.84x
diamond (root → many → sink) 2.20 0.86 2.56x
independent (no deps) 0.65 0.35 1.84x
layered (4-layer full bipartite) 230.96 67.15 3.44x
nested (depth-3, walked across all groups) 17.78 4.77 3.73x
rev-chain (adversarial reverse insertion) 894.04 114.16 7.83x

The 7.8x on rev-chain comes from the projection precompute alone (one projection vs N projections); algorithmic complexity is unchanged and remains O(N²) on that shape until the follow-up PR.

The nested row times a recursive walk across all 39 nested groups (depth=3, ~222 tasks each); the root-only timing was dominated by setup overhead and was previously labelled ~noise in this table.

Test plan

  • task-sdk/tests/task_sdk/definitions/test_taskgroup.py + test_dag.py — 145 passed (18 new shape-correctness cases)
  • airflow-core/tests/unit/utils/test_task_group.py — 20 passed (1 new serialization round-trip case)
  • airflow-core/tests/unit/serialization/ (task-group / topological filter) — 4 passed
  • prek run mypy-task-sdk — passed
  • prek run mypy-airflow-core — passed
  • prek run --from-ref upstream/main --stage pre-commit — passed

Was generative AI tooling used to co-author this PR?
  • Yes — Claude Code (Opus 4.7)

Generated-by: Claude Code (Opus 4.7) following the guidelines

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR optimizes TaskGroup.topological_sort (and the serialized equivalent) by pre-projecting upstream dependencies to sibling indices once per call, then performing a sweep using an efficient emitted-flag structure. This aims to reduce repeated per-edge work during sorting while preserving the existing emission order.

Changes:

  • Replaced the previous modified-Kahn loop in TaskGroup.topological_sort with a “project then sweep” implementation, including explicit cycle detection.
  • Mirrored the same algorithm into SerializedTaskGroup.topological_sort for consistency/performance.
  • Added Task SDK tests that validate topological correctness across multiple DAG shapes, plus a newsfragment entry.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File Description
task-sdk/tests/task_sdk/definitions/test_taskgroup.py Adds cross-shape correctness tests for TaskGroup.topological_sort.
task-sdk/src/airflow/sdk/definitions/taskgroup.py Implements projected dependency computation and sweep-based topo sort in Task SDK TaskGroup.
airflow-core/src/airflow/serialization/definitions/taskgroup.py Mirrors the new topo sort algorithm for serialized task groups and adds cycle detection behavior.
airflow-core/newsfragments/67288.improvement.rst Documents the TaskGroup.topological_sort performance improvement.

Comment thread task-sdk/tests/task_sdk/definitions/test_taskgroup.py Outdated
Comment thread airflow-core/src/airflow/serialization/definitions/taskgroup.py Outdated
shahar1 added 6 commits May 21, 2026 23:21
The previous modified-Kahn implementation re-derived each child's upstream
edges (materializing ``upstream_list`` and walking ``parent_group``) on every
outer-loop pass. Project per-task upstream IDs onto sibling-level integer
indices once up front, then run a greedy multi-pass sweep against that
projection with a ``bytearray`` emission flag. Emission order is identical to
the previous implementation; existing order-sensitive tests cover the contract.

Same change is mirrored in SerializedTaskGroup.
Adds a round-trip test for SerializedTaskGroup.topological_sort (the
serialization variant was previously untested), rewrites the newsfragment
in user-facing terms, and cleans up a stale reference and type annotation
in the task-sdk shape tests.
Per Copilot review feedback, the previous wording said cycles are caught
at "deserialization time" — but DAG.check_cycle runs at DAG parse time
(via dagbag loading), not during from_dict/from_json. Reword to describe
cycles reaching this code path as malformed serialized data, with the
defensive ValueError still raised on detection.
Tighten _project_child_deps and the topological_sort body: drop the
per-edge child_idx comparison in favour of a single set.discard at the
end, inline the get_task/.task_group chain, and replace the pre-alloc
loop with a list comprehension. Net -15 lines per file with a small
(~10%) speedup on the layered shape and no regression elsewhere.

The hot _sweep_projection inner loop is left intact — earlier attempts
to extract its duplicated body into a closure cost 10-30% on
independent / layered shapes and were reverted.
@shahar1
shahar1 force-pushed the taskgroup-topo-sort-projection branch from 62325c1 to 411f90e Compare May 21, 2026 20:22
@shahar1
shahar1 requested a review from Copilot May 21, 2026 20:30

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Comment thread task-sdk/tests/task_sdk/definitions/test_taskgroup.py
Comment thread airflow-core/newsfragments/67288.improvement.rst
Drop n=500 from the parametrize grid. The algorithm has no n-dependent
branches, so n=100 covers every code path; n=500 only re-runs the same
loops with more iterations and added ~0.7s to the file without
exercising new correctness behaviour (per Copilot review feedback on PR
apache#67288).
@shahar1
shahar1 marked this pull request as ready for review May 21, 2026 20:46

@kaxil kaxil left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@shahar1
shahar1 merged commit 7bf71a3 into apache:main May 27, 2026
143 checks passed
@shahar1
shahar1 deleted the taskgroup-topo-sort-projection branch May 27, 2026 18:39
shahar1 added a commit to shahar1/airflow that referenced this pull request May 29, 2026
vatsrahul1001 pushed a commit that referenced this pull request May 29, 2026
…ected sweep (#67288) (#67687)

* Speed up TaskGroup.topological_sort with int-indexed projected sweep (backport #67288)

* Remove __getitem__ and NodeNotFound tests not available in v3-2-test

These tests depend on TaskGroup.__getitem__ and NodeNotFound exception
which don't exist in v3-2-test. Keep topological_sort shape correctness
tests which are for the actual optimization.

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Haiku 4.5 <noreply@anthropic.com>
vatsrahul1001 pushed a commit that referenced this pull request May 29, 2026
…ected sweep (#67288) (#67687)

* Speed up TaskGroup.topological_sort with int-indexed projected sweep (backport #67288)

* Remove __getitem__ and NodeNotFound tests not available in v3-2-test

These tests depend on TaskGroup.__getitem__ and NodeNotFound exception
which don't exist in v3-2-test. Keep topological_sort shape correctness
tests which are for the actual optimization.

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Haiku 4.5 <noreply@anthropic.com>
shahar1 added a commit to shahar1/airflow that referenced this pull request May 29, 2026
@vatsrahul1001 vatsrahul1001 added this to the Airflow 3.3.0 milestone Jun 15, 2026
jason810496 pushed a commit to hkc-8010/my-airflow-repository that referenced this pull request Jul 28, 2026
…deps

TaskGroup._project_child_deps only looked at a group's own upstream_task_ids,
which stays empty for a direct group-to-group dependency (list or individual
`>>`) and for a task-level dependency that crosses into another group's entry
task. Both cases sorted the group as if it had no upstream at all.

Now also pulls in the group's upstream_group_ids and its root tasks' upstream
task ids before projecting sibling dependencies. Applied to both the
serialization-layer sort and the mirrored design-time sort in task-sdk.

closes: apache#65291

Related: apache#67964 (closed for inactivity, written against the
topological_sort implementation before PR apache#67288/apache#67688 rewrote it) and
apache#65639 (draft, same issue, also predates the rewrite).
jason810496 added a commit that referenced this pull request Jul 28, 2026
…dependencies (#69933)

* Fix grid/graph view topological sort for group-level and cross-group deps

TaskGroup._project_child_deps only looked at a group's own upstream_task_ids,
which stays empty for a direct group-to-group dependency (list or individual
`>>`) and for a task-level dependency that crosses into another group's entry
task. Both cases sorted the group as if it had no upstream at all.

Now also pulls in the group's upstream_group_ids and its root tasks' upstream
task ids before projecting sibling dependencies. Applied to both the
serialization-layer sort and the mirrored design-time sort in task-sdk.

closes: #65291

Related: #67964 (closed for inactivity, written against the
topological_sort implementation before PR #67288/#67688 rewrote it) and
#65639 (draft, same issue, also predates the rewrite).

* Address review feedback: cache get_task_group_dict, describe test intent not issue numbers

viiccwen pointed out that fetching the group map inside topological_sort() rebuilds
the whole DAG's group tree on every nested group's own call, turning a render with
G groups into an O(G^2) cost. get_task_group_dict() is now memoized per DAG instance
(kept behind a small private helper since methodtools.lru_cache has no type stubs and
would otherwise widen the public method's return type to Any for every caller).

Also reworded test comments/docstrings that cited issue numbers to describe what's
actually being verified instead.

* Hoist common logic into shared lib

* Remove caching on get_task_group_dict

* Remove stale get_task_group_dict cache tests

The cache these tests asserted was removed in the previous commit, so the
identity check and the _get_task_group_dict_cached.cache_info() assertions
no longer apply.

* Add call-level task group memo to reduce calc

* Tidy Typy

* Add test for task group memoing

---------

Co-authored-by: TP <uranusjr@apache.org>
Co-authored-by: LIU ZHE YOU <zhu424.dev@gmail.com>
pierrejeambrun pushed a commit that referenced this pull request Jul 28, 2026
…dependencies (#69933) (#70591)

* Fix grid/graph view topological sort for group-level and cross-group deps

TaskGroup._project_child_deps only looked at a group's own upstream_task_ids,
which stays empty for a direct group-to-group dependency (list or individual
`>>`) and for a task-level dependency that crosses into another group's entry
task. Both cases sorted the group as if it had no upstream at all.

Now also pulls in the group's upstream_group_ids and its root tasks' upstream
task ids before projecting sibling dependencies. Applied to both the
serialization-layer sort and the mirrored design-time sort in task-sdk.

closes: #65291

Related: #67964 (closed for inactivity, written against the
topological_sort implementation before PR #67288/#67688 rewrote it) and
#65639 (draft, same issue, also predates the rewrite).

* Address review feedback: cache get_task_group_dict, describe test intent not issue numbers

viiccwen pointed out that fetching the group map inside topological_sort() rebuilds
the whole DAG's group tree on every nested group's own call, turning a render with
G groups into an O(G^2) cost. get_task_group_dict() is now memoized per DAG instance
(kept behind a small private helper since methodtools.lru_cache has no type stubs and
would otherwise widen the public method's return type to Any for every caller).

Also reworded test comments/docstrings that cited issue numbers to describe what's
actually being verified instead.

* Hoist common logic into shared lib

* Remove caching on get_task_group_dict

* Remove stale get_task_group_dict cache tests

The cache these tests asserted was removed in the previous commit, so the
identity check and the _get_task_group_dict_cached.cache_info() assertions
no longer apply.

* Add call-level task group memo to reduce calc

* Tidy Typy

* Add test for task group memoing

---------

Co-authored-by: TP <uranusjr@apache.org>
Co-authored-by: LIU ZHE YOU <zhu424.dev@gmail.com>
(cherry picked from commit d7aa929)

# Conflicts:
#	airflow-core/src/airflow/api_fastapi/core_api/services/ui/task_group.py

Co-authored-by: Hemkumar Chheda <95332229+hkc-8010@users.noreply.github.com>
vatsrahul1001 pushed a commit that referenced this pull request Aug 5, 2026
…dependencies (#69933) (#70591)

* Fix grid/graph view topological sort for group-level and cross-group deps

TaskGroup._project_child_deps only looked at a group's own upstream_task_ids,
which stays empty for a direct group-to-group dependency (list or individual
`>>`) and for a task-level dependency that crosses into another group's entry
task. Both cases sorted the group as if it had no upstream at all.

Now also pulls in the group's upstream_group_ids and its root tasks' upstream
task ids before projecting sibling dependencies. Applied to both the
serialization-layer sort and the mirrored design-time sort in task-sdk.

closes: #65291

Related: #67964 (closed for inactivity, written against the
topological_sort implementation before PR #67288/#67688 rewrote it) and
#65639 (draft, same issue, also predates the rewrite).

* Address review feedback: cache get_task_group_dict, describe test intent not issue numbers

viiccwen pointed out that fetching the group map inside topological_sort() rebuilds
the whole DAG's group tree on every nested group's own call, turning a render with
G groups into an O(G^2) cost. get_task_group_dict() is now memoized per DAG instance
(kept behind a small private helper since methodtools.lru_cache has no type stubs and
would otherwise widen the public method's return type to Any for every caller).

Also reworded test comments/docstrings that cited issue numbers to describe what's
actually being verified instead.

* Hoist common logic into shared lib

* Remove caching on get_task_group_dict

* Remove stale get_task_group_dict cache tests

The cache these tests asserted was removed in the previous commit, so the
identity check and the _get_task_group_dict_cached.cache_info() assertions
no longer apply.

* Add call-level task group memo to reduce calc

* Tidy Typy

* Add test for task group memoing

---------

Co-authored-by: TP <uranusjr@apache.org>
Co-authored-by: LIU ZHE YOU <zhu424.dev@gmail.com>
(cherry picked from commit d7aa929)

# Conflicts:
#	airflow-core/src/airflow/api_fastapi/core_api/services/ui/task_group.py

Co-authored-by: Hemkumar Chheda <95332229+hkc-8010@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants