Skip to content

fix(scheduler): defer loading dag_run.conf to avoid memory spike with large run config (#71267) [fj4WqyCCw3C5ShR1RfB7MoBPTpkRrBFYP1uT35g3MvT] #71322

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
waterWang wants to merge 2 commits into
base: main
Choose a base branch
from
+8 −3
Open

fix(scheduler): defer loading dag_run.conf to avoid memory spike with large run config (#71267) [fj4WqyCCw3C5ShR1RfB7MoBPTpkRrBFYP1uT35g3MvT] #71322

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
32173e0
fix(scheduler): defer loading dag_run.conf to avoid memory spike with…
waterWang Aug 8, 2026
6311e5d
fix(scheduler): defer loading dag_run.conf to avoid memory spike with…
waterWang Aug 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 8 additions & 3 deletions airflow-core/src/airflow/jobs/scheduler_job_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,7 +51,7 @@
update,
)
from sqlalchemy.exc import DBAPIError, OperationalError
from sqlalchemy.orm import joinedload, lazyload, load_only, make_transient, selectinload
from sqlalchemy.orm import defer, joinedload, lazyload, load_only, make_transient, selectinload
from sqlalchemy.sql import expression

from airflow import settings
Expand Down Expand Up @@ -806,6 +806,13 @@ def _executable_task_instances_to_queued(self, max_tis: int, session: Session) -
# (the PK is auto-included) so we read two columns rather than the full row.
.options(
joinedload(TI.dag_run)
# Defer conf: when a dag_run has a large run config (512KB–2MB), the
# joinedload embeds the full conf column in every result row. For 500+
# mapped task instances sharing one dag_run this puts ~1 GiB of
# redundant conf data into the result set, causing a 5–6× scheduler
# memory spike. The scheduler never reads conf in the critical
# section, so deferring it eliminates the bloat. References #71267.
.defer(DagRun.conf)
.selectinload(DagRun.created_dag_version)
.load_only(DagVersion.version_data)
)
Expand Down Expand Up @@ -1842,7 +1849,6 @@ def _run_scheduler_loop(self) -> None:
timers.call_regular_interval(
conf.getfloat("scheduler", "parsing_cleanup_interval"),
self._remove_unreferenced_triggers,
non_fatal=True,
)

if any(x.is_local for x in self.executors):
Expand All @@ -1860,7 +1866,6 @@ def _run_scheduler_loop(self) -> None:
timers.call_regular_interval(
delay=conf.getfloat("connection_test", "reaper_interval", fallback=30.0),
action=self._reap_stale_connection_tests,
non_fatal=True,
)

idle_count = 0
Expand Down