mirror of
https://github.com/microsoft/agent-framework.git
synced 2026-06-16 21:04:09 +08:00
Add create checkpoint to workflow interface
This commit is contained in:
@@ -242,6 +242,7 @@ from ._workflows._agent_executor import (
|
||||
)
|
||||
from ._workflows._agent_utils import resolve_agent_id
|
||||
from ._workflows._checkpoint import (
|
||||
CheckpointID,
|
||||
CheckpointStorage,
|
||||
FileCheckpointStorage,
|
||||
InMemoryCheckpointStorage,
|
||||
@@ -285,7 +286,6 @@ from ._workflows._functional import (
|
||||
workflow,
|
||||
)
|
||||
from ._workflows._request_info_mixin import response_handler
|
||||
from ._workflows._runner import Runner
|
||||
from ._workflows._runner_context import (
|
||||
InProcRunnerContext,
|
||||
RunnerContext,
|
||||
@@ -379,6 +379,7 @@ __all__ = [
|
||||
"ChatResponse",
|
||||
"ChatResponseUpdate",
|
||||
"CheckResult",
|
||||
"CheckpointID",
|
||||
"CheckpointStorage",
|
||||
"ClassSkill",
|
||||
"CompactionProvider",
|
||||
@@ -470,7 +471,6 @@ __all__ = [
|
||||
"RoleLiteral",
|
||||
"RubricScore",
|
||||
"RunContext",
|
||||
"Runner",
|
||||
"RunnerContext",
|
||||
"SecretString",
|
||||
"SelectiveToolCallCompactionStrategy",
|
||||
|
||||
@@ -65,7 +65,7 @@ class Runner:
|
||||
|
||||
# Checkpointing related attributes
|
||||
self._resumed_from_checkpoint = False
|
||||
self._previous_checkpoint_id: CheckpointID | None = None
|
||||
self.previous_checkpoint_id: CheckpointID | None = None
|
||||
|
||||
@property
|
||||
def context(self) -> RunnerContext:
|
||||
@@ -113,12 +113,12 @@ class Runner:
|
||||
for event in await self._ctx.drain_events():
|
||||
yield event
|
||||
|
||||
# Create the first checkpoint. Checkpoints are usually considered to be created at the end of an iteration,
|
||||
# we can think of the first checkpoint as being created at the end of a "superstep 0" which captures the
|
||||
# states after which the start executor has run. Note that we execute the start executor outside of the
|
||||
# main iteration loop.
|
||||
# Create a checkpoint before a run starts. Checkpoints are usually considered to be created at the
|
||||
# end of an iteration, we can think of this checkpoint as being created at the end of a "superstep 0"
|
||||
# which captures the states after which the start executor has run. Note that we execute the start
|
||||
# executor outside of the main iteration loop.
|
||||
if await self._ctx.has_messages() and not self._resumed_from_checkpoint:
|
||||
await self._create_checkpoint_if_enabled()
|
||||
await self.create_checkpoint_if_enabled()
|
||||
|
||||
while self._iteration < self._max_iterations:
|
||||
logger.info(f"Starting superstep {self._iteration + 1}")
|
||||
@@ -165,7 +165,7 @@ class Runner:
|
||||
self._state.commit()
|
||||
|
||||
# Create checkpoint after each superstep iteration
|
||||
await self._create_checkpoint_if_enabled()
|
||||
await self.create_checkpoint_if_enabled()
|
||||
|
||||
yield WorkflowEvent.superstep_completed(iteration=self._iteration)
|
||||
|
||||
@@ -231,7 +231,7 @@ class Runner:
|
||||
]
|
||||
await asyncio.gather(*tasks)
|
||||
|
||||
async def _create_checkpoint_if_enabled(self) -> None:
|
||||
async def create_checkpoint_if_enabled(self) -> None:
|
||||
"""Create a checkpoint if checkpointing is enabled and attach a label and metadata."""
|
||||
if not self._ctx.has_checkpointing():
|
||||
return
|
||||
@@ -249,7 +249,7 @@ class Runner:
|
||||
self._workflow_name,
|
||||
self._graph_signature_hash,
|
||||
self._state,
|
||||
self._previous_checkpoint_id,
|
||||
self.previous_checkpoint_id,
|
||||
self._iteration,
|
||||
)
|
||||
|
||||
@@ -257,9 +257,9 @@ class Runner:
|
||||
"Created checkpoint: %s with parent checkpoint at iteration %d: %s",
|
||||
checkpoint_id,
|
||||
self._iteration,
|
||||
self._previous_checkpoint_id,
|
||||
self.previous_checkpoint_id,
|
||||
)
|
||||
self._previous_checkpoint_id = checkpoint_id
|
||||
self.previous_checkpoint_id = checkpoint_id
|
||||
except Exception as e:
|
||||
logger.warning(
|
||||
"Failed to create checkpoint at iteration %d: %s. "
|
||||
@@ -267,7 +267,7 @@ class Runner:
|
||||
"The next successfully-created checkpoint will be parented to the last successful checkpoint: %s",
|
||||
self._iteration,
|
||||
e,
|
||||
self._previous_checkpoint_id,
|
||||
self.previous_checkpoint_id,
|
||||
)
|
||||
|
||||
async def restore_from_checkpoint(
|
||||
@@ -396,7 +396,7 @@ class Runner:
|
||||
"""
|
||||
self._resumed_from_checkpoint = True
|
||||
self._iteration = checkpoint.iteration_count
|
||||
self._previous_checkpoint_id = checkpoint.checkpoint_id
|
||||
self.previous_checkpoint_id = checkpoint.checkpoint_id
|
||||
|
||||
async def _set_executor_state(self, executor_id: str, state: dict[str, Any]) -> None:
|
||||
"""Store executor state in state under a reserved key.
|
||||
|
||||
@@ -18,9 +18,9 @@ from typing import TYPE_CHECKING, Any, Literal, overload
|
||||
|
||||
from .._sessions import ContextProvider
|
||||
from .._types import ResponseStream
|
||||
from ..exceptions import WorkflowException, WorkflowRunnerException
|
||||
from ..exceptions import WorkflowCheckpointException, WorkflowException, WorkflowRunnerException
|
||||
from ..observability import OtelAttr, capture_exception, create_workflow_span
|
||||
from ._checkpoint import CheckpointStorage
|
||||
from ._checkpoint import CheckpointID, CheckpointStorage
|
||||
from ._const import DEFAULT_MAX_ITERATIONS, GLOBAL_KWARGS_KEY, WORKFLOW_RUN_KWARGS_KEY
|
||||
from ._edge import (
|
||||
EdgeGroup,
|
||||
@@ -474,6 +474,43 @@ class Workflow(DictConvertible):
|
||||
"""Get the list of executors in the workflow."""
|
||||
return list(self.executors.values())
|
||||
|
||||
async def create_checkpoint(self, checkpoint_storage: CheckpointStorage | None) -> CheckpointID:
|
||||
"""Create a checkpoint of the current workflow state in the provided storage.
|
||||
|
||||
Args:
|
||||
checkpoint_storage: The CheckpointStorage instance where the checkpoint will be stored.
|
||||
If None, will use the workflow's default checkpoint storage if configured, or raise
|
||||
if checkpointing is not enabled.
|
||||
|
||||
Notes:
|
||||
- Checkpoints can only be created when the workflow is idle (not actively running).
|
||||
- Checkpoints are automatically created at the end of each superstep if a checkpoint storage is configured.
|
||||
Use this method only when necessary, for example to capture the initial state of the workflow prior to the
|
||||
first run.
|
||||
- Creating a checkpoint manually will alter the checkpoint lineage. The new checkpoint will become the
|
||||
parent of the next checkpoint created automatically (if checkpointing is enabled by providing a storage).
|
||||
"""
|
||||
if self._is_run_active():
|
||||
raise WorkflowException(
|
||||
"Cannot create checkpoint while a workflow run is active. "
|
||||
"Checkpointing is only allowed between runs when the workflow is idle."
|
||||
)
|
||||
|
||||
if checkpoint_storage is None and not self._runner.context.has_checkpointing():
|
||||
raise WorkflowCheckpointException(
|
||||
"Checkpoint storage must be provided to create a checkpoint when checkpointing is not enabled."
|
||||
)
|
||||
if checkpoint_storage is not None:
|
||||
self._runner.context.set_runtime_checkpoint_storage(checkpoint_storage)
|
||||
|
||||
try:
|
||||
await self._runner.create_checkpoint_if_enabled()
|
||||
if self._runner.previous_checkpoint_id is None:
|
||||
raise WorkflowCheckpointException("Failed to create checkpoint.")
|
||||
return self._runner.previous_checkpoint_id
|
||||
finally:
|
||||
self._runner.context.clear_runtime_checkpoint_storage()
|
||||
|
||||
async def _run_workflow_with_tracing(
|
||||
self,
|
||||
initial_executor_fn: Callable[[], Awaitable[None]] | None = None,
|
||||
@@ -1165,6 +1202,15 @@ class Workflow(DictConvertible):
|
||||
**kwargs,
|
||||
)
|
||||
|
||||
def _is_run_active(self) -> bool:
|
||||
"""Check if a workflow run is currently active.
|
||||
|
||||
Returns:
|
||||
True if a run is active, False otherwise.
|
||||
"""
|
||||
existing_stream = self._active_run() if self._active_run is not None else None
|
||||
return existing_stream is not None
|
||||
|
||||
async def reset_for_new_run(self) -> None:
|
||||
"""Reset the workflow for a new run that is independent from prior runs.
|
||||
|
||||
|
||||
@@ -921,7 +921,7 @@ async def test_runner_mark_resumed_sets_previous_checkpoint_id():
|
||||
)
|
||||
|
||||
# Pre-condition: nothing to chain back to
|
||||
assert runner._previous_checkpoint_id is None # pyright: ignore[reportPrivateUsage]
|
||||
assert runner.previous_checkpoint_id is None
|
||||
|
||||
resumed_checkpoint = WorkflowCheckpoint(
|
||||
checkpoint_id="resumed-cp-id",
|
||||
@@ -933,7 +933,7 @@ async def test_runner_mark_resumed_sets_previous_checkpoint_id():
|
||||
|
||||
assert runner._resumed_from_checkpoint is True # pyright: ignore[reportPrivateUsage]
|
||||
assert runner._iteration == 3 # pyright: ignore[reportPrivateUsage]
|
||||
assert runner._previous_checkpoint_id == "resumed-cp-id" # pyright: ignore[reportPrivateUsage]
|
||||
assert runner.previous_checkpoint_id == "resumed-cp-id"
|
||||
|
||||
|
||||
async def test_runner_post_resume_checkpoint_chains_to_resumed_checkpoint():
|
||||
|
||||
Reference in New Issue
Block a user