by-framework 0.2.2.dev8__tar.gz → 0.2.2.dev10__tar.gz

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: by-framework
-Version: 0.2.2.dev8
+Version: 0.2.2.dev10
 Summary: 分布式 Agent 调度框架
 License-File: LICENSE
 Requires-Python: >=3.12

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "by-framework"
-version = "0.2.2.dev8"
+version = "0.2.2.dev10"
 description = "分布式 Agent 调度框架"
 readme = "README.md"
 requires-python = ">=3.12"

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/__init__.py

@@ -8,6 +8,7 @@ from `GatewayWorker` and running `run_worker`.
 from .admin import WorkerManager
 from .client.byai_client import ByaiGatewayClient
 from .client.client import (
+    CancelSessionResponse,
     CancelTaskResponse,
     DataStreamEntry,
     GatewayClient,
@@ -148,6 +149,7 @@ __all__ = [
     "DataStreamEntry",
     "SendMessageResponse",
     "CancelTaskResponse",
+    "CancelSessionResponse",
     "run_worker",
     "logger",
     "setup_logging",

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/admin/cli.py

@@ -31,8 +31,9 @@ app = typer.Typer(
     help=(
         "by-framework cluster admin CLI. For Redis Cluster, omit --redis-url "
         "and set REDIS_CLUSTER_HOST (comma-separated host:port list; implies "
-        "cluster mode) and REDIS_KEY_SCHEMA_VERSION=v2. REDIS_MODE=cluster + "
-        "REDIS_CLUSTER_NODES also still works."
+        "both cluster mode and REDIS_KEY_SCHEMA_VERSION=v2 unless set "
+        "explicitly). REDIS_MODE=cluster + REDIS_CLUSTER_NODES + "
+        "REDIS_KEY_SCHEMA_VERSION=v2 also still works."
     ),
 )
 worker_app = typer.Typer(no_args_is_help=True)

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/client/client.py

@@ -37,6 +37,7 @@ from by_framework.core.protocol.commands import (
 from by_framework.core.protocol.data_message import DataMessage
 from by_framework.core.protocol.message_header import MessageHeader
 from by_framework.core.protocol.responses import (
+    CancelSessionResponse,
     CancelTaskResponse,
     ExecutionStatus,
     SendMessageResponse,
@@ -581,6 +582,99 @@ class GatewayClient:
             cancelled_count=len(to_cancel),
         )
 
+    async def cancel_session(
+        self,
+        session_id: str,
+        reason: str = "",
+        requested_by: str = "client",
+        cancel_mode: str = CancelMode.GRACEFUL,
+    ) -> CancelSessionResponse:
+        """Cancel every active execution registered under a session.
+
+        Unlike cancel_task, this does not require a starting message_id:
+        it cancels the whole session's active executions in one call.
+        """
+        if self.registry is None:
+            raise ValueError("GatewayClient requires a WorkerRegistry to cancel tasks")
+
+        all_executions = await self.registry.get_all_session_executions(session_id)
+
+        if not all_executions:
+            return CancelSessionResponse(
+                success=False,
+                session_id=session_id,
+                status=ExecutionStatus.NOT_FOUND,
+                timestamp=int(time.time() * 1000),
+                cancelled_count=0,
+                already_finished_count=0,
+            )
+
+        terminal_states = {"COMPLETED", "FAILED", "CANCELLED"}
+        to_cancel = [
+            node
+            for node in all_executions
+            if node.get("status", "") not in terminal_states
+        ]
+        terminal_nodes = [
+            node for node in all_executions if node.get("status", "") in terminal_states
+        ]
+
+        # Flag terminal executions so a still-running descendant can't wake
+        # them back up via callback, without changing their status.
+        for node in terminal_nodes:
+            await self.registry.mark_cancel_requested(
+                node.get("execution_id", ""), session_id, reason
+            )
+
+        if not to_cancel:
+            return CancelSessionResponse(
+                success=False,
+                session_id=session_id,
+                status=ExecutionStatus.ALREADY_FINISHED,
+                timestamp=int(time.time() * 1000),
+                cancelled_count=0,
+                already_finished_count=len(terminal_nodes),
+            )
+
+        for node in to_cancel:
+            node_execution_id = node.get("execution_id", "")
+            node_worker_id = node.get("worker_id", "")
+            node_message_id = node.get("message_id", "")
+
+            await self.registry.mark_execution_cancelling(
+                node_execution_id, session_id, reason
+            )
+
+            if node_worker_id:
+                cancel_command = CancelTaskCommand(
+                    header=MessageHeader(
+                        message_id=f"{CANCEL_MESSAGE_ID_PREFIX}{uuid.uuid4().hex[:8]}",
+                        session_id=session_id,
+                        trace_id=node.get("trace_id") or uuid.uuid4().hex,
+                        target_agent_type=node.get("target_agent_type", ""),
+                        parent_message_id=node_message_id,
+                    ),
+                    target_message_id=node_message_id,
+                    target_execution_id=node_execution_id,
+                    target_worker_id=node_worker_id,
+                    reason=reason,
+                    requested_by=requested_by,
+                    cancel_mode=cancel_mode,
+                )
+                await self.redis.xadd(
+                    RedisKeys.worker_ctrl_stream(node_worker_id),
+                    cancel_command.to_redis_payload(),
+                )
+
+        return CancelSessionResponse(
+            success=True,
+            session_id=session_id,
+            status=ExecutionStatus.CANCEL_REQUESTED,
+            timestamp=int(time.time() * 1000),
+            cancelled_count=len(to_cancel),
+            already_finished_count=len(terminal_nodes),
+        )
+
     async def send_message(
         self,
         target_agent_type: str,
@@ -691,7 +785,28 @@ class GatewayClient:
             content=params["content"],
             extra_payload=params["extra_payload"],
         )
-        execution_id = f"{EXECUTION_ID_PREFIX}{uuid.uuid4().hex[:8]}"
+
+        # A RESUME reuses the message_id of the original AskAgentCommand so the
+        # worker can look the suspended execution back up. Reuse its
+        # execution_id too, and skip re-initializing the registry record for
+        # it below -- initialize_execution() would otherwise overwrite the
+        # message_id -> execution_id mapping and detach this resume from the
+        # execution it's meant to continue.
+        resumed_execution = None
+        if (
+            params["action_type"] == ActionType.RESUME.value
+            and self.registry
+            and hasattr(self.registry, "get_execution_by_message_id")
+        ):
+            resumed_execution = await self.registry.get_execution_by_message_id(
+                message_id, session_id=params["session_id"]
+            )
+
+        execution_id = (
+            resumed_execution["execution_id"]
+            if resumed_execution
+            else f"{EXECUTION_ID_PREFIX}{uuid.uuid4().hex[:8]}"
+        )
 
         # 3. Resolve route and optionally probe agent type/liveness
         should_dispatch_control = True
@@ -837,8 +952,14 @@ class GatewayClient:
                 error_code=ExecutionStatus.ERR_AGENT_TYPE_UNAVAILABLE,
             )
 
-        # Initialize execution tracking
-        if self.registry and hasattr(self.registry, "initialize_execution"):
+        # Initialize execution tracking. Skipped for a RESUME that reattached
+        # to an existing execution -- that execution is already tracked, and
+        # re-initializing it here would overwrite its message_id mapping.
+        if (
+            not resumed_execution
+            and self.registry
+            and hasattr(self.registry, "initialize_execution")
+        ):
             try:
                 await self.registry.initialize_execution(
                     {

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/common/config.py

@@ -9,6 +9,7 @@ from dataclasses import dataclass, field
 from typing import Literal, Optional
 
 from by_framework.common.constants import RedisKeys
+from by_framework.common.logger import logger
 
 
 @dataclass(frozen=True)
@@ -36,6 +37,9 @@ class RedisConfig:
         to switch to cluster mode, no separate REDIS_MODE=cluster needed.
         An explicit REDIS_MODE still takes precedence when set, so existing
         explicit REDIS_MODE=cluster + REDIS_CLUSTER_NODES setups keep working.
+
+        REDIS_DATABASE replaces REDIS_DB (which still works as a deprecated
+        fallback, logging a warning, during the transition period).
         """
         password = os.environ.get("REDIS_PASSWORD", "")
         username = os.environ.get("REDIS_USERNAME") or None
@@ -51,10 +55,15 @@ class RedisConfig:
             for node in cluster_nodes_str.split(","):
                 node_host, node_port = node.rsplit(":", 1)
                 cluster_nodes.append((node_host, int(node_port)))
+        db_str = os.environ.get("REDIS_DATABASE")
+        if db_str is None:
+            db_str = os.environ.get("REDIS_DB")
+            if db_str is not None:
+                logger.warning("REDIS_DB is deprecated, use REDIS_DATABASE instead")
         return cls(
             host=os.environ.get("REDIS_HOST", "localhost"),
             port=int(os.environ.get("REDIS_PORT", "6379")),
-            db=int(os.environ.get("REDIS_DB", "0")),
+            db=int(db_str) if db_str is not None else 0,
             password=password,
             username=username,
             max_connections=int(max_connections) if max_connections else None,

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/common/constants.py

@@ -12,10 +12,20 @@ from typing import Optional
 def get_key_schema_version() -> str:
     """Return the configured Redis key schema version ("v1" or "v2").
 
-    Controlled by REDIS_KEY_SCHEMA_VERSION, defaulting to "v1" (the current
-    unprefixed key format). Cluster mode requires "v2" (see redis_client.init_redis).
+    Controlled by REDIS_KEY_SCHEMA_VERSION, which always wins when set
+    explicitly. When it isn't set, REDIS_CLUSTER_HOST being configured
+    implies "v2" (cluster mode requires v2 - v1 keys have no hash tags and
+    hit CROSSSLOT errors under Cluster; see redis_client.init_redis's
+    fail-fast check); otherwise it defaults to "v1" (the legacy unprefixed
+    key format). This mirrors RedisConfig.from_env()'s REDIS_MODE/
+    REDIS_CLUSTER_HOST precedence, but deliberately does NOT infer "v2"
+    from an explicit REDIS_MODE=cluster alone (without REDIS_CLUSTER_HOST)
+    - that legacy explicit-mode path still requires
+    REDIS_KEY_SCHEMA_VERSION=v2 to be set by hand.
     """
-    version = os.environ.get("REDIS_KEY_SCHEMA_VERSION", "v1")
+    version = os.environ.get("REDIS_KEY_SCHEMA_VERSION")
+    if version is None:
+        version = "v2" if os.environ.get("REDIS_CLUSTER_HOST") else "v1"
     if version not in ("v1", "v2"):
         raise ValueError(
             f"Invalid REDIS_KEY_SCHEMA_VERSION: {version!r} (must be 'v1' or 'v2')"

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/common/emitter.py

@@ -107,6 +107,22 @@ class DefaultSseLayoutBuilder(DataLayoutBuilder):
         parent_order_id: Optional[str] = None,
     ) -> Dict[str, Any]:
         """Build default BaiYing-compatible user input form payload."""
+        try:
+            parsed_prompt = json.loads(prompt)
+        except (json.JSONDecodeError, TypeError):
+            parsed_prompt = None
+
+        # Structured question prompts are already in the format expected by clients.
+        if isinstance(parsed_prompt, dict) and "questions" in parsed_prompt:
+            return self.build(
+                content=prompt,
+                role="assistant",
+                content_type=SseReasonMessageType.ask_user_question.value,
+                source_agent_type=source_agent_type,
+                order_id=order_id,
+                parent_order_id=parent_order_id,
+            )
+
         input_form = {
             "formStatus": 0,
             "pluginMachineFields": [

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/core/__init__.py

@@ -26,6 +26,7 @@ from .protocol import (
     BaiYingMessage,
     BaiYingMessageRole,
     BaseCommand,
+    CancelSessionResponse,
     CancelTaskCommand,
     CancelTaskResponse,
     DataMessage,
@@ -53,6 +54,7 @@ from .workspace import WorkspaceManager
 __all__ = [
     "SendMessageResponse",
     "CancelTaskResponse",
+    "CancelSessionResponse",
     "BaiYingMessage",
     "BaiYingMessageRole",
     "MessageContent",

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/core/protocol/__init__.py

@@ -65,6 +65,7 @@ from .message import (
 )
 from .message_header import MessageHeader
 from .responses import (
+    CancelSessionResponse,
     CancelTaskResponse,
     CancelTaskResponseDict,
     SendMessageResponse,
@@ -81,6 +82,7 @@ from .results import (
 __all__ = [
     "SendMessageResponse",
     "CancelTaskResponse",
+    "CancelSessionResponse",
     "SendMessageResponseDict",
     "CancelTaskResponseDict",
     "AgentTaskResult",

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/core/protocol/content_type.py

@@ -33,6 +33,7 @@ class SseReasonMessageType(str, Enum):
     think_code_result = "3007"  # thinking process code execution result
     task_finished = "3009"  # task finished
     task_user_input = "3013"  # user input
+    ask_user_question = "3014"  # ask user question
     task_create_file = "3010"  # create file
     task_title = "3011"  # task title
     agent_card = "2015"  # agent card

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/core/protocol/responses.py

@@ -92,3 +92,14 @@ class CancelTaskResponse:
     timestamp: int
     error: str = ""
     cancelled_count: int = 0
+
+
+@dataclass(frozen=True)
+class CancelSessionResponse:
+    success: bool
+    session_id: str
+    status: str
+    timestamp: int
+    cancelled_count: int = 0
+    already_finished_count: int = 0
+    error: str = ""

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/core/registry.py

@@ -1011,7 +1011,8 @@ class WorkerRegistry:
         old_execution = dict(current)
         current["status"] = status
         now = int(time.time() * 1000)
-        current["finished_at"] = now
+        if is_terminal_state(status):
+            current["finished_at"] = now
         current["updated_at"] = now
         if completion:
             for key, value in completion.items():

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/worker/app.py

@@ -131,6 +131,17 @@ async def _run_worker_async(
     # parties can configure either mode purely programmatically or purely
     # via env vars, without the two modes behaving inconsistently.
     env_config = RedisConfig.from_env()
+    if redis_mode is not None:
+        effective_mode = redis_mode
+    elif redis_cluster_nodes:
+        # Mirrors REDIS_CLUSTER_HOST's env-var precedence: passing
+        # redis_cluster_nodes alone is enough to imply cluster mode, no
+        # separate redis_mode="cluster" required. Without this, passing
+        # redis_cluster_nodes alone would silently stay in standalone mode
+        # and the nodes list would be dropped on the floor.
+        effective_mode = "cluster"
+    else:
+        effective_mode = env_config.mode
     effective_config = replace(
         env_config,
         host=redis_host if redis_host is not None else env_config.host,
@@ -142,7 +153,7 @@ async def _run_worker_async(
         username=(
             redis_username if redis_username is not None else env_config.username
         ),
-        mode=redis_mode if redis_mode is not None else env_config.mode,
+        mode=effective_mode,
         cluster_nodes=(
             redis_cluster_nodes
             if redis_cluster_nodes is not None
@@ -276,16 +287,23 @@ def run_worker(
     Redis configuration: every redis_* argument defaults to None, meaning
     "not specified" - when not passed, the value is read from the
     corresponding REDIS_* environment variable (REDIS_HOST, REDIS_PORT,
-    REDIS_DB, REDIS_PASSWORD, REDIS_USERNAME, REDIS_MODE, REDIS_CLUSTER_HOST,
-    REDIS_CLUSTER_NODES), falling back to "localhost"/6379/standalone if
-    none is set. Setting REDIS_CLUSTER_HOST (comma-separated "host:port"
-    list) alone is enough to switch to cluster mode - no separate
-    REDIS_MODE=cluster required, though an explicit REDIS_MODE still wins if
-    set. An explicitly-passed argument always takes precedence over its env
-    var, for every field, in both standalone and cluster mode - so either
-    mode can be configured purely programmatically (e.g.
-    ``redis_mode="cluster", redis_cluster_nodes=[("h1", 6379)]``) or purely
-    via env vars, without the two modes behaving inconsistently.
+    REDIS_DATABASE, REDIS_PASSWORD, REDIS_USERNAME, REDIS_MODE,
+    REDIS_CLUSTER_HOST, REDIS_CLUSTER_NODES), falling back to
+    "localhost"/6379/standalone if none is set. REDIS_DATABASE replaces
+    REDIS_DB (still works as a deprecated fallback, logging a warning).
+    Setting REDIS_CLUSTER_HOST (comma-separated "host:port" list) alone is
+    enough to switch to cluster mode - no separate REDIS_MODE=cluster
+    required, though an explicit REDIS_MODE still wins if set; it also
+    implies REDIS_KEY_SCHEMA_VERSION=v2 unless that's set explicitly. An
+    explicitly-passed argument always takes precedence over its env var,
+    for every field, in both standalone and cluster mode. The same
+    single-argument convenience carries over to the programmatic path:
+    passing just ``redis_cluster_nodes=[("h1", 6379)]`` (no ``redis_mode``)
+    is enough to switch to cluster mode, mirroring REDIS_CLUSTER_HOST -
+    ``redis_mode`` only needs to be passed to explicitly force a mode
+    (e.g. pin a script to standalone regardless of REDIS_CLUSTER_HOST).
+    So either mode can be configured purely programmatically or purely via
+    env vars, without the two paths behaving inconsistently.
 
     Plugin support:
     - **Auto mode**: System automatically discovers and loads all subclasses

{by_framework-0.2.2.dev8 → by_framework-0.2.2.dev10}/src/by_framework/worker/runner.py

@@ -505,6 +505,16 @@ class WorkerRunner:
                     header.message_id, session_id=header.session_id
                 )
 
+            if isinstance(command, ResumeCommand) and existing_execution is None:
+                logger.warning(
+                    "[%s] ResumeCommand did not resolve to an existing execution "
+                    "(message_id=%s, session_id=%s); starting a new, disconnected "
+                    "execution instead of continuing the suspended one.",
+                    self.worker.worker_id,
+                    header.message_id,
+                    header.session_id,
+                )
+
             # Skip terminal state replays
             if (
                 existing_execution