[BREAKING] Python: Refactor Checkpointing for runner and runner context (#1645)

* Refactor Checkpointing for runner and runner context

* exception

* Fix formatting

* Comments

* rename

* Add detailed doc string
This commit is contained in:
Tao Chen
2025-10-23 20:34:55 -07:00
committed by GitHub
Unverified
parent 3aa682082a
commit 31701dbb92
14 changed files with 280 additions and 281 deletions
@@ -20,9 +20,7 @@ def test_workflow_checkpoint_default_values():
assert checkpoint.timestamp != ""
assert checkpoint.messages == {}
assert checkpoint.shared_state == {}
assert checkpoint.executor_states == {}
assert checkpoint.iteration_count == 0
assert checkpoint.max_iterations == 100
assert checkpoint.metadata == {}
assert checkpoint.version == "1.0"
@@ -35,9 +33,7 @@ def test_workflow_checkpoint_custom_values():
timestamp=custom_timestamp,
messages={"executor1": [{"data": "test"}]},
shared_state={"key": "value"},
executor_states={"executor1": {"state": "active"}},
iteration_count=5,
max_iterations=50,
metadata={"test": True},
version="2.0",
)
@@ -47,9 +43,7 @@ def test_workflow_checkpoint_custom_values():
assert checkpoint.timestamp == custom_timestamp
assert checkpoint.messages == {"executor1": [{"data": "test"}]}
assert checkpoint.shared_state == {"key": "value"}
assert checkpoint.executor_states == {"executor1": {"state": "active"}}
assert checkpoint.iteration_count == 5
assert checkpoint.max_iterations == 50
assert checkpoint.metadata == {"test": True}
assert checkpoint.version == "2.0"
@@ -290,7 +284,6 @@ async def test_file_checkpoint_storage_json_serialization():
workflow_id="complex-workflow",
messages={"executor1": [{"data": {"nested": {"value": 42}}, "source_id": "test", "target_id": None}]},
shared_state={"list": [1, 2, 3], "dict": {"a": "b", "c": {"d": "e"}}, "bool": True, "null": None},
executor_states={"executor1": {"state": "active", "config": {"timeout": 30, "retries": 3}}},
)
# Save and load
@@ -300,7 +293,6 @@ async def test_file_checkpoint_storage_json_serialization():
assert loaded is not None
assert loaded.messages == checkpoint.messages
assert loaded.shared_state == checkpoint.shared_state
assert loaded.executor_states == checkpoint.executor_states
# Verify the JSON file is properly formatted
file_path = Path(temp_dir) / f"{checkpoint.checkpoint_id}.json"
@@ -8,6 +8,7 @@ from typing import Any
from agent_framework._workflows._checkpoint import WorkflowCheckpoint
from agent_framework._workflows._checkpoint_encoding import encode_checkpoint_value
from agent_framework._workflows._checkpoint_summary import get_checkpoint_summary
from agent_framework._workflows._const import EXECUTOR_STATE_KEY
from agent_framework._workflows._events import RequestInfoEvent, WorkflowEvent
from agent_framework._workflows._request_info_executor import (
PendingRequestDetails,
@@ -16,10 +17,7 @@ from agent_framework._workflows._request_info_executor import (
RequestInfoMessage,
RequestResponse,
)
from agent_framework._workflows._runner_context import (
Message,
WorkflowState,
)
from agent_framework._workflows._runner_context import Message
from agent_framework._workflows._shared_state import SharedState
from agent_framework._workflows._workflow_context import WorkflowContext
@@ -29,9 +27,6 @@ PENDING_STATE_KEY = RequestInfoExecutor._PENDING_SHARED_STATE_KEY # pyright: ig
class _StubRunnerContext:
"""Minimal runner context stub for exercising WorkflowContext helpers."""
def __init__(self, stored_state: dict[str, Any] | None = None) -> None:
self._state = stored_state or {}
async def send_message(self, message: Message) -> None: # pragma: no cover - unused in tests
return None
@@ -53,31 +48,27 @@ class _StubRunnerContext:
async def next_event(self) -> WorkflowEvent: # pragma: no cover - unused
raise RuntimeError("Not implemented in stub context")
async def get_executor_state(self, executor_id: str) -> dict[str, Any] | None: # pragma: no cover - trivial
return self._state
async def set_executor_state(self, executor_id: str, state: dict[str, Any]) -> None: # pragma: no cover - unused
self._state = state
def has_checkpointing(self) -> bool: # pragma: no cover - unused
return False
def set_workflow_id(self, workflow_id: str) -> None: # pragma: no cover - unused
pass
def reset_for_new_run(self, workflow_shared_state: SharedState | None = None) -> None: # pragma: no cover - unused
def reset_for_new_run(self) -> None: # pragma: no cover - unused
pass
async def create_checkpoint(self, metadata: dict[str, Any] | None = None) -> str: # pragma: no cover - unused
async def create_checkpoint(
self,
shared_state: SharedState,
iteration_count: int,
metadata: dict[str, Any] | None = None,
) -> str: # pragma: no cover - unused
raise RuntimeError("Checkpointing not supported in stub context")
async def load_checkpoint(self, checkpoint_id: str) -> WorkflowCheckpoint | None: # pragma: no cover - unused
return None
async def get_workflow_state(self) -> WorkflowState: # pragma: no cover - unused
return {} # type: ignore[return-value]
async def set_workflow_state(self, state: WorkflowState) -> None: # pragma: no cover - unused
async def apply_checkpoint(self, checkpoint: WorkflowCheckpoint) -> None: # pragma: no cover - unused
pass
def set_streaming(self, streaming: bool) -> None: # pragma: no cover - unused
@@ -120,8 +111,8 @@ async def test_rehydrate_falls_back_when_request_type_missing() -> None:
},
)
runner_ctx = _StubRunnerContext({PENDING_STATE_KEY: {request_id: snapshot}})
ctx: WorkflowContext[Any] = WorkflowContext("request_info", ["workflow"], SharedState(), runner_ctx)
ctx: WorkflowContext[Any] = WorkflowContext("request_info", ["workflow"], SharedState(), _StubRunnerContext())
await ctx.set_executor_state({PENDING_STATE_KEY: {request_id: snapshot}})
executor = RequestInfoExecutor(id="request_info")
@@ -143,8 +134,8 @@ async def test_has_pending_request_detects_snapshot() -> None:
},
)
runner_ctx = _StubRunnerContext({PENDING_STATE_KEY: {request_id: snapshot}})
ctx: WorkflowContext[Any] = WorkflowContext("request_info", ["workflow"], SharedState(), runner_ctx)
ctx: WorkflowContext[Any] = WorkflowContext("request_info", ["workflow"], SharedState(), _StubRunnerContext())
await ctx.set_executor_state({PENDING_STATE_KEY: {request_id: snapshot}})
executor = RequestInfoExecutor(id="request_info")
@@ -152,9 +143,8 @@ async def test_has_pending_request_detects_snapshot() -> None:
async def test_has_pending_request_false_when_snapshot_absent() -> None:
shared_state = SharedState()
runner_ctx = _StubRunnerContext({"pending_requests": {}})
ctx: WorkflowContext[Any] = WorkflowContext("request_info", ["workflow"], shared_state, runner_ctx)
ctx: WorkflowContext[Any] = WorkflowContext("request_info", ["workflow"], SharedState(), _StubRunnerContext())
await ctx.set_executor_state({PENDING_STATE_KEY: {}})
executor = RequestInfoExecutor(id="request_info")
@@ -196,7 +186,6 @@ def test_pending_requests_from_checkpoint_and_summary() -> None:
}
}
},
executor_states={},
iteration_count=1,
)
@@ -284,10 +273,13 @@ async def test_run_persists_pending_requests_in_runner_state() -> None:
await executor.execute(approval, ctx.source_executor_ids, shared_state, runner_ctx)
# Runner state should include both pending snapshot and serialized request events
assert PENDING_STATE_KEY in runner_ctx._state # pyright: ignore[reportPrivateUsage]
assert approval.request_id in runner_ctx._state[PENDING_STATE_KEY] # pyright: ignore[reportPrivateUsage]
assert await shared_state.has(EXECUTOR_STATE_KEY)
executor_state = await shared_state.get(EXECUTOR_STATE_KEY)
assert executor.id in executor_state
assert PENDING_STATE_KEY in executor_state[executor.id]
assert approval.request_id in executor_state[executor.id][PENDING_STATE_KEY]
response_ctx: WorkflowContext[None] = WorkflowContext("request_info", ["source"], shared_state, runner_ctx)
await executor.handle_response("approved", approval.request_id, response_ctx) # type: ignore
assert runner_ctx._state[PENDING_STATE_KEY] == {} # pyright: ignore[reportPrivateUsage]
assert executor_state[executor.id][PENDING_STATE_KEY] == {}
@@ -414,9 +414,7 @@ async def test_workflow_run_stream_from_checkpoint_with_external_storage(simple_
workflow_id="test-workflow",
messages={},
shared_state={},
executor_states={},
iteration_count=0,
max_iterations=100,
)
checkpoint_id = await storage.save_checkpoint(test_checkpoint)
@@ -451,9 +449,7 @@ async def test_workflow_run_from_checkpoint_non_streaming(simple_executor: Execu
workflow_id="test-workflow",
messages={},
shared_state={},
executor_states={},
iteration_count=0,
max_iterations=100,
)
checkpoint_id = await storage.save_checkpoint(test_checkpoint)
@@ -484,9 +480,7 @@ async def test_workflow_run_stream_from_checkpoint_with_responses(simple_executo
workflow_id="test-workflow",
messages={},
shared_state={},
executor_states={},
iteration_count=0,
max_iterations=100,
)
checkpoint_id = await storage.save_checkpoint(test_checkpoint)
@@ -525,7 +519,7 @@ class StateTrackingExecutor(Executor):
"""An executor that tracks state in shared state to test context reset behavior."""
@handler
async def handle_message(self, message: StateTrackingMessage, ctx: WorkflowContext[Any, list]) -> None:
async def handle_message(self, message: StateTrackingMessage, ctx: WorkflowContext[Any, list[Any]]) -> None:
"""Handle the message and track it in shared state."""
# Get existing messages from shared state
try:
@@ -6,7 +6,7 @@ import pytest
from opentelemetry import trace
from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
from agent_framework import WorkflowBuilder
from agent_framework import InMemoryCheckpointStorage, WorkflowBuilder
from agent_framework._workflows._executor import Executor, handler
from agent_framework._workflows._runner_context import InProcRunnerContext, Message
from agent_framework._workflows._shared_state import SharedState
@@ -426,7 +426,7 @@ async def test_workflow_error_handling_in_tracing(span_exporter: InMemorySpanExp
@pytest.mark.parametrize("enable_otel", [False], indirect=True)
async def test_message_trace_context_serialization(span_exporter: InMemorySpanExporter) -> None:
"""Test that message trace context is properly serialized/deserialized."""
ctx = InProcRunnerContext()
ctx = InProcRunnerContext(InMemoryCheckpointStorage())
# Create message with trace context
message = Message(
@@ -439,16 +439,18 @@ async def test_message_trace_context_serialization(span_exporter: InMemorySpanEx
await ctx.send_message(message)
# Get context state (which serializes messages)
state = await ctx.get_workflow_state()
# Create a checkpoint that includes the message
checkpoint_id = await ctx.create_checkpoint(SharedState(), 0)
checkpoint = await ctx.load_checkpoint(checkpoint_id)
assert checkpoint is not None
# Check serialized message includes trace context
serialized_msg = state["messages"]["source"][0]
serialized_msg = checkpoint.messages["source"][0]
assert serialized_msg["trace_contexts"] == [{"traceparent": "00-trace-span-01"}]
assert serialized_msg["source_span_ids"] == ["span123"]
# Test deserialization
await ctx.set_workflow_state(state)
await ctx.apply_checkpoint(checkpoint)
restored_messages = await ctx.drain_messages()
restored_msg = list(restored_messages.values())[0][0]