prefect-client 3.6.1.dev1__tar.gz → 3.6.2.dev1__tar.gz

{prefect_client-3.6.1.dev1 → prefect_client-3.6.2.dev1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: prefect-client
-Version: 3.6.1.dev1
+Version: 3.6.2.dev1
 Summary: Workflow orchestration and management.
 Project-URL: Changelog, https://github.com/PrefectHQ/prefect/releases
 Project-URL: Documentation, https://docs.prefect.io

prefect_client-3.6.2.dev1/src/prefect/_build_info.py

@@ -0,0 +1,5 @@
+# Generated by versioningit
+__version__ = "3.6.2.dev1"
+__build_date__ = "2025-11-13 08:09:13.475253+00:00"
+__git_commit__ = "144cdba00ccee55b1c252f80268ebdf73356dfcd"
+__dirty__ = False

{prefect_client-3.6.1.dev1 → prefect_client-3.6.2.dev1}/src/prefect/context.py

@@ -75,6 +75,10 @@ def serialize_context(
     tags_context = TagsContext.get()
     settings_context = SettingsContext.get()
 
+    # Serialize deployment ContextVars for cross-process context propagation
+    deployment_id = _deployment_id.get()
+    deployment_params = _deployment_parameters.get()
+
     return {
         "flow_run_context": flow_run_context.serialize() if flow_run_context else {},
         "task_run_context": task_run_context.serialize() if task_run_context else {},
@@ -85,6 +89,8 @@ def serialize_context(
         ).serialize()
         if asset_ctx_kwargs
         else {},
+        "deployment_id": str(deployment_id) if deployment_id else None,
+        "deployment_parameters": deployment_params,
     }
 
 
@@ -138,6 +144,15 @@ def hydrated_context(
             # Set up asset context
             if asset_context := serialized_context.get("asset_context"):
                 stack.enter_context(AssetContext(**asset_context))
+            # Restore deployment ContextVars for cross-process context propagation
+            if deployment_id_str := serialized_context.get("deployment_id"):
+                from uuid import UUID
+
+                deployment_id_token = _deployment_id.set(UUID(deployment_id_str))
+                stack.callback(_deployment_id.reset, deployment_id_token)
+            if deployment_params := serialized_context.get("deployment_parameters"):
+                deployment_params_token = _deployment_parameters.set(deployment_params)
+                stack.callback(_deployment_parameters.reset, deployment_params_token)
         yield
 
 
@@ -743,6 +758,13 @@ class SettingsContext(ContextModel):
             return None
 
 
+# Root deployment context vars for O(1) access in nested flows
+_deployment_id: ContextVar[UUID | None] = ContextVar("deployment_id", default=None)
+_deployment_parameters: ContextVar[dict[str, Any] | None] = ContextVar(
+    "deployment_parameters", default=None
+)
+
+
 def get_run_context() -> Union[FlowRunContext, TaskRunContext]:
     """
     Get the current run context from within a task or flow function.

{prefect_client-3.6.1.dev1 → prefect_client-3.6.2.dev1}/src/prefect/flow_engine.py

@@ -53,6 +53,8 @@ from prefect.context import (
     SettingsContext,
     SyncClientContext,
     TagsContext,
+    _deployment_id,
+    _deployment_parameters,
     get_settings_context,
     hydrated_context,
     serialize_context,
@@ -626,6 +628,13 @@ class FlowRunEngine(BaseFlowRunEngine[P, R]):
                     else should_persist_result(),
                 )
             )
+            # Set deployment context vars only if this is the top-level deployment run
+            # (nested flows will inherit via ContextVar propagation)
+            if self.flow_run.deployment_id and not _deployment_id.get():
+                id_token = _deployment_id.set(self.flow_run.deployment_id)
+                params_token = _deployment_parameters.set(self.flow_run.parameters)
+                stack.callback(_deployment_id.reset, id_token)
+                stack.callback(_deployment_parameters.reset, params_token)
             stack.enter_context(ConcurrencyContextV1())
             stack.enter_context(ConcurrencyContext())
             if lease_id := self.state.state_details.deployment_concurrency_lease_id:
@@ -1203,6 +1212,13 @@ class AsyncFlowRunEngine(BaseFlowRunEngine[P, R]):
                     else should_persist_result(),
                 )
             )
+            # Set deployment context vars only if this is the top-level deployment run
+            # (nested flows will inherit via ContextVar propagation)
+            if self.flow_run.deployment_id and not _deployment_id.get():
+                id_token = _deployment_id.set(self.flow_run.deployment_id)
+                params_token = _deployment_parameters.set(self.flow_run.parameters)
+                stack.callback(_deployment_id.reset, id_token)
+                stack.callback(_deployment_parameters.reset, params_token)
             stack.enter_context(ConcurrencyContextV1())
             stack.enter_context(ConcurrencyContext())
             if lease_id := self.state.state_details.deployment_concurrency_lease_id:

{prefect_client-3.6.1.dev1 → prefect_client-3.6.2.dev1}/src/prefect/runtime/deployment.py

@@ -32,7 +32,7 @@ from typing import TYPE_CHECKING, Any, Callable, List, Optional
 
 from prefect._internal.concurrency.api import create_call, from_sync
 from prefect.client.orchestration import get_client
-from prefect.context import FlowRunContext
+from prefect.context import FlowRunContext, _deployment_id, _deployment_parameters
 
 from .flow_run import _get_flow_run
 
@@ -105,6 +105,10 @@ async def _get_deployment(deployment_id: str) -> "DeploymentResponse":
 
 
 def get_id() -> Optional[str]:
+    # Check deployment context var first (avoids API calls in nested flows)
+    if deployment_id := _deployment_id.get():
+        return str(deployment_id)
+
     flow_run = FlowRunContext.get()
     deployment_id = getattr(flow_run, "deployment_id", None)
     if deployment_id is None:
@@ -123,6 +127,10 @@ def get_id() -> Optional[str]:
 
 
 def get_parameters() -> dict[str, Any]:
+    # Check deployment context var first (avoids API calls in nested flows)
+    if params := _deployment_parameters.get():
+        return params
+
     run_id = get_flow_run_id()
     if run_id is None:
         return {}

{prefect_client-3.6.1.dev1 → prefect_client-3.6.2.dev1}/src/prefect/server/api/concurrency_limits_v2.py

@@ -17,12 +17,20 @@ from prefect.server.database import PrefectDBInterface, provide_database_interfa
 from prefect.server.schemas import actions
 from prefect.server.utilities.schemas import PrefectBaseModel
 from prefect.server.utilities.server import PrefectRouter
+from prefect.settings.context import get_current_settings
 from prefect.types._concurrency import ConcurrencyLeaseHolder
+from prefect.utilities.math import clamped_poisson_interval
 
 router: PrefectRouter = PrefectRouter(
     prefix="/v2/concurrency_limits", tags=["Concurrency Limits V2"]
 )
 
+# Max retry delay (in seconds) for non-tag concurrency limits when slot occupancy is high.
+# Tag-based limits use the legacy tag_concurrency_slot_wait_seconds setting (default 30s)
+# to restore V1 behavior. Non-tag limits use this higher value to allow more uniform queues
+# to check more frequently while still preventing excessive delays from long-running tasks.
+DEFAULT_SLOT_WAIT_SECONDS = 30.0
+
 
 @router.post("/", status_code=status.HTTP_201_CREATED)
 async def create_concurrency_limit_v2(
@@ -217,6 +225,24 @@ async def _generate_concurrency_locked_response(
     limits: list[schemas.core.ConcurrencyLimitV2],
     slots: int,
 ) -> HTTPException:
+    """
+    Generate a 423 Locked response when concurrency slots cannot be acquired.
+
+    Calculates an appropriate Retry-After header value based on the blocking limit's
+    characteristics. For limits without slot decay, caps avg_slot_occupancy_seconds
+    at a configured maximum to prevent excessive retry delays from long-running tasks:
+
+    - Tag-based limits (name starts with "tag:"): Capped at tag_concurrency_slot_wait_seconds
+      (default 30s) to restore V1 behavior that users relied on
+    - Other limits: Capped at DEFAULT_SLOT_WAIT_SECONDS (30s) to allow more uniform queues
+      while still preventing astronomical delays
+
+    Low average occupancies are always respected (e.g., 2s stays 2s, not forced higher).
+    Limits with slot decay use the decay rate directly without capping.
+
+    The final retry value includes jitter via clamped_poisson_interval to prevent
+    thundering herd when many tasks retry simultaneously.
+    """
     active_limits = [limit for limit in limits if bool(limit.active)]
 
     await models.concurrency_limits_v2.bulk_update_denied_slots(
@@ -234,13 +260,20 @@ async def _generate_concurrency_locked_response(
     blocking_limit = max((limit for limit in active_limits), key=num_blocking_slots)
     blocking_slots = num_blocking_slots(blocking_limit)
 
-    wait_time_per_slot = (
-        blocking_limit.avg_slot_occupancy_seconds
-        if blocking_limit.slot_decay_per_second == 0.0
-        else (1.0 / blocking_limit.slot_decay_per_second)
-    )
+    if blocking_limit.slot_decay_per_second == 0.0:
+        settings = get_current_settings()
+        max_wait = (
+            settings.server.tasks.tag_concurrency_slot_wait_seconds
+            if blocking_limit.name.startswith("tag:")
+            else DEFAULT_SLOT_WAIT_SECONDS
+        )
+        wait_time_per_slot = min(blocking_limit.avg_slot_occupancy_seconds, max_wait)
+    else:
+        wait_time_per_slot = 1.0 / blocking_limit.slot_decay_per_second
 
-    retry_after = wait_time_per_slot * blocking_slots
+    retry_after = clamped_poisson_interval(
+        average_interval=wait_time_per_slot * blocking_slots
+    )
 
     return HTTPException(
         status_code=status.HTTP_423_LOCKED,

prefect_client-3.6.1.dev1/src/prefect/_build_info.py

@@ -1,5 +0,0 @@
-# Generated by versioningit
-__version__ = "3.6.1.dev1"
-__build_date__ = "2025-11-07 20:18:14.119205+00:00"
-__git_commit__ = "0c000c0667d4d565f9d4b4fab30e01b27cbffb72"
-__dirty__ = False