Use a shared scheduler context.

Author: rjpower

lib/iris/AGENTS.md

@@ -25,7 +25,7 @@ Archived design docs (implemented, read code instead): `.agents/projects/2026*_i
 
 ```bash
 # Unit tests (run from lib/iris/)
-cd lib/iris && uv run --group dev python -m pytest --tb=short -m 'not slow and not docker and not e2e' tests/
+cd lib/iris && uv run --group dev python -m pytest --tb=short -m 'not slow and not docker and not requires_cluster' tests/
 ```
 
 See `TESTING.md` for the complete testing policy, E2E test commands, and markers.

lib/iris/src/iris/cluster/controller/budget.py

@@ -19,6 +19,7 @@
 from iris.cluster.controller.db import ControllerDB, Tx
 from iris.cluster.controller.schema import job_config_table, tasks_table
 from iris.cluster.controller.task_state import ACTIVE_TASK_STATES
+from iris.cluster.types import UserBudgetDefaults
 from iris.rpc import config_pb2, job_pb2
 
 logger = logging.getLogger(__name__)
@@ -36,26 +37,6 @@ class UserTask(Generic[T]):
 _ACTIVE_TASK_STATES = tuple(ACTIVE_TASK_STATES)
 
 
-@dataclass
-class UserBudgetDefaults:
-    """Budget settings for users without an explicit user_budgets row.
-
-    An absent row means "defaults apply" — we no longer stamp a row into the
-    table at first-submit time, so the scheduler and launch-job guard both
-    fall back to these values when the lookup misses.
-    """
-
-    budget_limit: int = 1000
-    """Max budget value applied to users without an override row.
-
-    A value of 0 means unlimited; any positive value caps accumulated spend
-    before :func:`compute_effective_band` downgrades INTERACTIVE work to BATCH.
-    """
-
-    max_band: int = job_pb2.PRIORITY_BAND_INTERACTIVE
-    """Default max priority band (proto int) for users without an override row."""
-
-
 def resource_value(cpu_millicores: int, memory_bytes: int, accelerator_count: int) -> int:
     """Compute a scalar resource value for budget tracking.