Reduce API server memory usage by eliminating SerializedDAG loads on task start#60803
Merged
Merged
Conversation
SerializedDAG loads on task start
kaxil
marked this pull request as ready for review
January 19, 2026 22:42
kaxil
force-pushed
the
move-upstream-map-indexes-to-sdk
branch
from
January 19, 2026 23:03
869eb6a to
804d0a7
Compare
kaxil
marked this pull request as draft
January 19, 2026 23:25
kaxil
force-pushed
the
move-upstream-map-indexes-to-sdk
branch
from
January 19, 2026 23:51
804d0a7 to
fa76ca6
Compare
kaxil
marked this pull request as ready for review
January 19, 2026 23:52
…n task start Remove server-side computation of upstream_map_indexes from the Execution API. The Task SDK now computes these values locally when resolving XCom arguments for mapped task groups. This eliminates the need to load SerializedDAG on the API server for the /run endpoint, reducing memory usage and improving performance. Changes: - Add task_mapping.py with SDK-side computation logic - Remove upstream_map_indexes field from TIRunContext - Add Cadwyn migration for backward compatibility with older SDKs - Update xcom_arg.py and expandinput.py for lazy computation
kaxil
force-pushed
the
move-upstream-map-indexes-to-sdk
branch
from
January 19, 2026 23:52
fa76ca6 to
8bf24ec
Compare
amoghrajesh
approved these changes
Jan 20, 2026
amoghrajesh
left a comment
Contributor
There was a problem hiding this comment.
Makes sense. Didn't realise we had a todo in there
jason810496
pushed a commit
to jason810496/airflow
that referenced
this pull request
Jan 22, 2026
amoghrajesh
pushed a commit
to astronomer/airflow
that referenced
this pull request
Jan 22, 2026
suii2210
pushed a commit
to suii2210/airflow
that referenced
this pull request
Jan 26, 2026
shreyas-dev
pushed a commit
to shreyas-dev/airflow
that referenced
this pull request
Jan 29, 2026
jhgoebbert
pushed a commit
to jhgoebbert/airflow_Owen-CH-Leung
that referenced
this pull request
Feb 8, 2026
Member
|
Nice! |
choo121600
pushed a commit
to choo121600/airflow
that referenced
this pull request
Feb 22, 2026
Subham-KRLX
pushed a commit
to Subham-KRLX/airflow
that referenced
this pull request
Mar 4, 2026
Ankurdeewan
pushed a commit
to Ankurdeewan/airflow
that referenced
this pull request
Mar 15, 2026
1 task
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Remove server-side computation of upstream_map_indexes from the Execution API. The Task SDK now computes these values locally when resolving XCom arguments for mapped task groups.
Previously, every
/runrequest loaded the fullSerializedDAG(5-50MB) just to computeupstream_map_indexesfor XCom resolution in mapped task groups. This contributed to memory pressure on the API server.Now the Task SDK computes these values lazily when resolving XCom arguments, using the existing
GetTICountAPI to query task instance counts.Screenshots from tests:

Was generative AI tooling used to co-author this PR?
{pr_number}.significant.rstor{issue_number}.significant.rst, in airflow-core/newsfragments.