agentengine-sdk-python 0.8.0__py3-none-any.whl → 0.8.2__py3-none-any.whl

ksadk/server/routes/sessions.py

@@ -7,7 +7,7 @@ from typing import Any
 
 from fastapi import HTTPException
 
-from ksadk.server.factory import get_runner, get_state
+from ksadk.server.factory import get_runtime_execution, get_state
 from ksadk.sessions import SessionEvent
 from ksadk.tools.gateway import tool_approval_capability
 from ksadk.toolsets import describe_agentengine_tools
@@ -23,6 +23,7 @@ from .common import (
     _build_native_terminal_capability,
     _ensure_session,
     _hydrate_session,
+    _resolve_agent_ui_spec,
 )
 from .models import (
     CreateSessionActionRequest,
@@ -58,45 +59,42 @@ from .streaming import _cancel_detached_streams_for_session
 @ui_bootstrap_router.post("/agentengine/api/v1/GetAgentUiBootstrap")
 async def get_agent_ui_bootstrap(request: UiBootstrapRequest):
     state = get_state()
-    runner = state.runner
-    if runner is not None:
-        runner = get_runner()
-    agent_id = request.AgentId or (_runtime_agent_id(runner) if runner else "default-agent")
-    description = getattr(runner.detection_result, "description", "") if runner else ""
-    framework = ""
-    if runner:
-        detection_type = getattr(getattr(runner, "detection_result", None), "type", None)
-        framework = str(getattr(detection_type, "value", detection_type) or "").strip().lower()
+    executor, launch_context = get_runtime_execution()
+    detection = launch_context.detection
+    agent_id = request.AgentId or _runtime_agent_id(launch_context)
+    description = str(getattr(detection, "description", "") or "")
+    framework = launch_context.runtime_type.strip().lower()
     workspace_enabled = workspace_files_enabled(default=True)
-    ui_spec = deps.resolve_agent_ui_spec()
-    runtime_capabilities = (
-        runner.get_runtime_capabilities()
-        if runner and callable(getattr(runner, "get_runtime_capabilities", None))
-        else {}
+    ui_spec = _resolve_agent_ui_spec()
+    runtime_capabilities = executor.native_capabilities(launch_context)
+    runtime_capability_matrix = executor.capability_matrix(launch_context)
+    resume_capability = (
+        runtime_capabilities.get("ResumeRun")
+        if isinstance(runtime_capabilities, Mapping)
+        else None
+    )
+    resume_capability = resume_capability if isinstance(resume_capability, Mapping) else {}
+    checkpoint_capability = (
+        runtime_capabilities.get("Checkpoint")
+        if isinstance(runtime_capabilities, Mapping)
+        else None
     )
+    checkpoint_capability = (
+        checkpoint_capability if isinstance(checkpoint_capability, Mapping) else {}
+    )
+    cancel_capability = (
+        runtime_capabilities.get("CancelRun")
+        if isinstance(runtime_capabilities, Mapping)
+        else None
+    )
+    cancel_capability = cancel_capability if isinstance(cancel_capability, Mapping) else {}
     checkpoint_resume_capability = {
-        "Supported": bool(
-            (runtime_capabilities.get("ResumeRun") or {}).get("Supported")
-            if isinstance(runtime_capabilities, Mapping)
-            else False
-        ),
-        "Checkpoint": (
-            (runtime_capabilities.get("Checkpoint") or {})
-            if isinstance(runtime_capabilities, Mapping)
-            else {}
-        ),
-        "ResumeRun": (
-            (runtime_capabilities.get("ResumeRun") or {})
-            if isinstance(runtime_capabilities, Mapping)
-            else {}
-        ),
+        "Supported": bool(resume_capability.get("Supported")),
+        "Checkpoint": checkpoint_capability,
+        "ResumeRun": resume_capability,
     }
     checkpoint_resume_supported = bool(checkpoint_resume_capability["Supported"])
-    cancel_run_supported = bool(
-        (runtime_capabilities.get("CancelRun") or {}).get("Supported")
-        if isinstance(runtime_capabilities, Mapping)
-        else False
-    )
+    cancel_run_supported = bool(cancel_capability.get("Supported"))
     responses_transport = {
         "Protocol": "responses",
         "Runtime": "ksadk",
@@ -137,7 +135,7 @@ async def get_agent_ui_bootstrap(request: UiBootstrapRequest):
         {
             "Agent": {
                 "AgentId": agent_id,
-                "Name": runner.detection_result.name if runner else agent_id,
+                "Name": str(getattr(detection, "name", "") or agent_id),
                 "Description": description or "",
                 "Framework": framework,
             },
@@ -151,6 +149,7 @@ async def get_agent_ui_bootstrap(request: UiBootstrapRequest):
                 "StopRun": cancel_run_supported,
                 "ResumeRun": checkpoint_resume_supported,
                 "RuntimeCapabilities": runtime_capabilities,
+                "RuntimeCapabilityMatrix": runtime_capability_matrix,
                 "CheckpointResumeCapability": checkpoint_resume_capability,
                 "RunLifecycle": {
                     "Enabled": True,
@@ -258,15 +257,13 @@ async def delete_session_action(request: SessionIdRequest):
 
 @sessions_router.post("/agentengine/api/v1/ListSessionEvents")
 async def list_session_events_action(request: ListSessionEventsActionRequest):
-    runner = get_state().runner
+    _executor, launch_context = get_runtime_execution()
     service = deps.resolve_session_service()
     session_id = str(request.SessionId or "").strip()
     if not session_id:
         # 未传 SessionId：返回该 agent 全部会话的事件（存储层跨会话查询，真分页无截断；
         # seq 游标是会话内序号，跨会话模式下不适用，直接忽略）
-        agent_id = str(request.AgentId or "").strip() or (
-            _runtime_agent_id(runner) if runner else "default-agent"
-        )
+        agent_id = str(request.AgentId or "").strip() or _runtime_agent_id(launch_context)
         events = await service.get_events_for_agent(
             agent_id,
             user_id=request.UserId,

ksadk/server/routes/streaming.py

@@ -4,18 +4,24 @@ from __future__ import annotations
 
 import asyncio
 import logging
+import os
 from collections.abc import AsyncIterator, Mapping
 from typing import Any
 
 from fastapi import HTTPException
+from fastapi.responses import StreamingResponse
 
+from ksadk.conversations.runtime_persistence import append_run_status_event
 from ksadk.server.factory import get_state
 
-from . import dependencies as deps
 from .models import _RUN_TERMINAL_STATUSES
 from .projection import _latest_invocation_status
 
 logger = logging.getLogger(__name__)
+SSE_HEARTBEAT_INTERVAL_SECONDS = max(
+    0.01,
+    float(os.getenv("AGENTENGINE_SSE_HEARTBEAT_SECONDS", "15")),
+)
 
 _RESERVED_UI_PATHS = {"/", "/chat", "/build", "/deploy"}
 _CUSTOM_API_PROXY_ENV_KEYS = ("KSADK_USER_BACKEND_URL", "LUOLUO_USER_BACKEND_URL")
@@ -33,6 +39,13 @@ _HOP_BY_HOP_HEADERS = {
 }
 
 
+def _observe_detached_task_result(task: asyncio.Task[Any]) -> None:
+    """Consume a detached task exception after ``_consume`` has logged it."""
+    if task.cancelled():
+        return
+    task.exception()
+
+
 class _DetachedSSEStream:
     _MAX_BACKLOG_CHUNKS = 256
 
@@ -58,6 +71,7 @@ class _DetachedSSEStream:
         self._task = asyncio.create_task(self._consume())
         self._registry.streams.add(self._task)
         self._task.add_done_callback(self._registry.streams.discard)
+        self._task.add_done_callback(_observe_detached_task_result)
         if self.invocation_id:
             self._registry.streams_by_invocation[self.invocation_id] = self
             self._task.add_done_callback(
@@ -70,7 +84,7 @@ class _DetachedSSEStream:
         if not self.session_id or not self.invocation_id:
             return False
         status = await _latest_invocation_status(
-            deps.resolve_session_service(),
+            get_state().resolve_session_service(),
             self.session_id,
             self.invocation_id,
         )
@@ -78,6 +92,7 @@ class _DetachedSSEStream:
 
     async def _consume(self) -> None:
         terminal_fallback_status: str | None = None
+        terminal_fallback_detail: str | None = None
         try:
             async for chunk in self._source:
                 self._backlog.append(chunk)
@@ -93,8 +108,9 @@ class _DetachedSSEStream:
         except asyncio.CancelledError:
             terminal_fallback_status = "cancelled"
             raise
-        except Exception:
+        except Exception as exc:
             terminal_fallback_status = "failed"
+            terminal_fallback_detail = f"{type(exc).__name__}: {exc}"[:2048]
             logger.exception("Detached SSE stream failed")
             raise
         finally:
@@ -108,15 +124,16 @@ class _DetachedSSEStream:
             if terminal_fallback_status and self.session_id:
                 try:
                     if not await self._has_terminal_run_status():
-                        await deps.conversation().append_run_status_event(
+                        await append_run_status_event(
                             session_id=self.session_id,
                             author="system",
                             status=terminal_fallback_status,
                             invocation_id=self.invocation_id or "",
                             detail=(
-                                f"background_{terminal_fallback_status}:{self.invocation_id or ''}"
+                                terminal_fallback_detail
+                                or f"background_{terminal_fallback_status}:{self.invocation_id or ''}"
                             ),
-                            session_service_provider=deps.resolve_session_service,
+                            session_service_provider=get_state().resolve_session_service,
                             run_mode=self._run_mode,
                             run_trigger=self._run_trigger,
                         )
@@ -151,7 +168,7 @@ class _DetachedSSEStream:
                 try:
                     chunk = await asyncio.wait_for(
                         queue.get(),
-                        timeout=deps.heartbeat_interval(),
+                        timeout=SSE_HEARTBEAT_INTERVAL_SECONDS,
                     )
                 except asyncio.TimeoutError:
                     yield ": ping\n\n"
@@ -221,3 +238,31 @@ def _reject_if_detached_resume_active(resume_key: tuple[str, str] | None) -> Non
             "run_id": resume_key[1],
         },
     )
+
+
+def detached_streaming_response(
+    source: AsyncIterator[str],
+    *,
+    invocation_id: str | None = None,
+    session_id: str | None = None,
+    resume_key: tuple[str, str] | None = None,
+    run_mode: str = "unknown",
+    run_trigger: str = "unknown",
+) -> StreamingResponse:
+    """Create an app-owned detached SSE response with reconnectable backlog."""
+
+    detached = _DetachedSSEStream(
+        source,
+        invocation_id=invocation_id,
+        session_id=session_id,
+        run_mode=run_mode,
+        run_trigger=run_trigger,
+    )
+    registry = detached._registry
+    if invocation_id and resume_key:
+        registry.resume_keys_by_invocation[invocation_id] = resume_key
+        registry.active_resume_invocation_by_key[resume_key] = invocation_id
+        detached._task.add_done_callback(
+            lambda _task: _clear_detached_resume_key(registry, invocation_id, resume_key)
+        )
+    return StreamingResponse(detached.iter_for_client(), media_type="text/event-stream")

ksadk/server/routes/workspace.py

@@ -4,6 +4,7 @@ from __future__ import annotations
 
 import io
 import logging
+import time
 import uuid
 import zipfile
 from pathlib import PurePosixPath
@@ -16,7 +17,8 @@ from pydantic import BaseModel
 
 from ksadk.conversations.attachment_storage import AttachmentStorageService
 from ksadk.conversations.model_context import normalize_model_metadata
-from ksadk.server.factory import get_state
+from ksadk.runtime.adapter import CancelResult
+from ksadk.server.factory import get_runtime_execution, get_state
 from ksadk_runtime_common.workspace_files.preview import (
     build_workspace_file_base_href,
     build_workspace_preview_csp,
@@ -26,7 +28,6 @@ from ksadk_runtime_common.workspace_files.preview import (
 from . import dependencies as deps
 from .common import (
     _action_response,
-    _resolve_active_runner,
     _resolve_current_model,
     _workspace_root_dir,
     _workspace_runtime_request,
@@ -134,6 +135,7 @@ async def delete_workspace_file_action(request: WorkspaceDeleteActionRequest):
 
 @control_router.post("/agentengine/api/v1/CancelRun")
 async def cancel_run_action(request: CancelRunActionRequest):
+    executor, launch_context = get_runtime_execution()
     detached = get_state().stream_registry.streams_by_invocation.get(request.InvocationId)
     service = deps.resolve_session_service()
     scoped_session_id = str(request.SessionId or "").strip()
@@ -145,6 +147,15 @@ async def cancel_run_action(request: CancelRunActionRequest):
         )
     if not scoped_session_id and detached is not None:
         scoped_session_id = detached_session_id
+    handle = (
+        executor.find_handle(
+            launch_context.runtime_type,
+            request.InvocationId,
+            scoped_session_id,
+        )
+        if scoped_session_id
+        else None
+    )
     if scoped_session_id:
         await _require_action_session(
             service,
@@ -152,7 +163,7 @@ async def cancel_run_action(request: CancelRunActionRequest):
             agent_id=request.AgentId,
             user_id=request.UserId,
         )
-        if detached is None and not await _session_contains_invocation(
+        if detached is None and handle is None and not await _session_contains_invocation(
             service,
             scoped_session_id,
             request.InvocationId,
@@ -171,33 +182,31 @@ async def cancel_run_action(request: CancelRunActionRequest):
             user_id=request.UserId,
         ):
             raise HTTPException(status_code=404, detail="Invocation not found")
-    found = detached is not None
+    found = detached is not None or handle is not None
     cancel_requested = False
     if detached is not None:
         cancel_requested = detached.cancel()
-    runner_cancel_status = "not_found" if found else "unsupported"
-    active_runner = _resolve_active_runner()
-    if active_runner is not None:
+    runtime_cancel_status = "detached_task_cancelled" if cancel_requested else "not_running"
+    if handle is not None and detached is None:
         try:
-            runner_result = active_runner.request_cancel(request.InvocationId)
-            if isinstance(runner_result, str) and runner_result:
-                runner_cancel_status = runner_result
-            elif runner_result is True:
-                runner_cancel_status = "accepted"
-            elif runner_result is False and not found:
-                runner_cancel_status = "not_found"
-        except Exception as exc:
-            runner_cancel_status = "error"
+            runtime_result = await executor.cancel(handle)
+            runtime_cancel_status = runtime_result.value
+        except Exception as exc:  # noqa: BLE001
+            runtime_cancel_status = CancelResult.FAILED.value
             logger.warning("CancelRun failed: %s", exc)
-    runner_accepted = runner_cancel_status in {"accepted", "cancelling", "cancelled"}
-    status = "cancelling" if found or runner_accepted else runner_cancel_status
+    runtime_accepted = runtime_cancel_status in {
+        "detached_task_cancelled",
+        CancelResult.INTERRUPTED_ACTIVE_TURN.value,
+        CancelResult.PENDING_CANCEL_RECORDED.value,
+    }
+    status = "cancelling" if runtime_accepted else runtime_cancel_status
     return _action_response(
         "CancelRun",
         {
-            "Cancelled": bool(cancel_requested or runner_accepted),
+            "Cancelled": bool(cancel_requested or runtime_accepted),
             "Found": found,
             "Status": status,
-            "RunnerCancelStatus": runner_cancel_status,
+            "RuntimeCancelStatus": runtime_cancel_status,
         },
     )
 
@@ -324,6 +333,11 @@ def _normalize_model_catalog_items(raw_models: list[Any]) -> list[dict[str, Any]
     return sorted(normalized_by_id.values(), key=lambda item: item["id"])
 
 
+_MODELS_CATALOG_TTL_SECONDS = 60.0
+# api_base -> (monotonic_ts, payload)。进程内短 TTL 缓存，见 _build_models_payload。
+_MODELS_CATALOG_CACHE: dict[str, tuple[float, dict[str, Any]]] = {}
+
+
 async def _build_models_payload() -> dict[str, Any]:
     import os
 
@@ -344,6 +358,18 @@ async def _build_models_payload() -> dict[str, Any]:
     if not api_base:
         return _fallback_catalog()
 
+    # 上游 /v1/models 是一次真实外网往返（数百毫秒），会话页初始化会
+    # 调用本接口，不能每次都同步打上游。做短 TTL 进程内缓存：
+    # 命中直接返回；上游失败时优先回退到过期的缓存值，避免模型列表
+    # 因上游抖动整体不可用。
+    cache_key = api_base.rstrip("/")
+    cached = _MODELS_CATALOG_CACHE.get(cache_key)
+    now = time.monotonic()
+    if cached is not None and now - cached[0] < _MODELS_CATALOG_TTL_SECONDS:
+        payload = dict(cached[1])
+        payload["cached"] = True
+        return payload
+
     try:
         base_url = api_base.rstrip("/")
         if base_url.endswith("/v1"):
@@ -368,9 +394,16 @@ async def _build_models_payload() -> dict[str, Any]:
                 str(item.get("id") or "").strip() != current_model for item in models
             ):
                 models = _normalize_model_catalog_items([*models, current_model])
-            return {"data": models, "current": current_model, "source": source}
+            payload = {"data": models, "current": current_model, "source": source}
+            _MODELS_CATALOG_CACHE[cache_key] = (now, payload)
+            return dict(payload)
     except Exception as e:
         logger.error(f"Failed to fetch models: {e}")
+        if cached is not None:
+            payload = dict(cached[1])
+            payload["stale"] = True
+            payload["error"] = str(e)
+            return payload
         fallback = _fallback_catalog()
         fallback["error"] = str(e)
         return fallback

ksadk/server/static/assets/{ArtifactsPanel-Cjnqt-8W.js → ArtifactsPanel-DSU1sWD_.js}

@@ -1 +1 @@
-import{bt as e,ht as t,lt as n,n as r,t as i,ut as a,wt as o,xt as s,yt as c}from"./index-QSwCHuB8.js";var l=o(s(),1),u=c();function d(){let o=e(e=>e.visible),s=e(e=>e.content),c=e(e=>e.type),d=e(e=>e.hide),f=(0,l.useRef)(null);if(r(f),!o||!s)return null;let p=i(s);return(0,u.jsxs)(`aside`,{className:n(`flex h-full flex-shrink-0 flex-col border-l border-slate-200/30 bg-white dark:border-slate-800/40 dark:bg-slate-950`,`w-[min(42rem,50vw)]`),children:[(0,u.jsxs)(`div`,{className:`flex h-14 flex-shrink-0 items-center justify-between border-b border-slate-200/30 px-4 dark:border-slate-800/40`,children:[(0,u.jsxs)(`div`,{className:`flex min-w-0 items-center gap-2`,children:[(0,u.jsx)(`span`,{className:`text-sm font-semibold text-slate-900 dark:text-slate-100`,children:`Artifact`}),(0,u.jsx)(`span`,{className:`rounded-md bg-slate-100 px-1.5 py-0.5 font-mono text-[11px] text-slate-500 dark:bg-slate-900 dark:text-slate-400`,children:c})]}),(0,u.jsxs)(`div`,{className:`flex items-center gap-1.5`,children:[(0,u.jsxs)(`button`,{type:`button`,onClick:()=>{let e=new Blob([s],{type:`text/html;charset=utf-8`}),t=URL.createObjectURL(e),n=document.createElement(`a`);n.href=t,n.download=`artifact.${c===`svg`?`svg`:`html`}`,document.body.appendChild(n),n.click(),document.body.removeChild(n),URL.revokeObjectURL(t)},className:`inline-flex items-center gap-1.5 rounded-lg px-2.5 py-1.5 text-xs font-medium text-slate-600 transition hover:bg-slate-100 dark:text-slate-300 dark:hover:bg-slate-900`,title:`Download`,children:[(0,u.jsx)(t,{className:`h-3.5 w-3.5`}),`Download`]}),(0,u.jsx)(`button`,{type:`button`,onClick:d,className:`rounded-lg p-1.5 text-slate-500 transition hover:bg-slate-100 hover:text-slate-900 dark:text-slate-400 dark:hover:bg-slate-900 dark:hover:text-slate-100`,"aria-label":`Close Artifact`,title:`Close`,children:(0,u.jsx)(a,{className:`h-4 w-4`})})]})]}),(0,u.jsx)(`div`,{className:`min-h-0 flex-1`,children:(0,u.jsx)(`iframe`,{ref:f,srcDoc:p,sandbox:`allow-scripts allow-downloads`,title:`Artifact Preview`,className:`h-full w-full border-0 bg-white`})})]})}export{d as ArtifactsPanel};
+import{St as e,Tt as t,bt as n,gt as r,lt as i,n as a,t as o,ut as s,xt as c}from"./index-8ipRcQ-M.js";var l=t(e(),1),u=n();function d(){let e=c(e=>e.visible),t=c(e=>e.content),n=c(e=>e.type),d=c(e=>e.hide),f=(0,l.useRef)(null);if(a(f),!e||!t)return null;let p=o(t);return(0,u.jsxs)(`aside`,{className:i(`flex h-full flex-shrink-0 flex-col border-l border-slate-200/30 bg-white dark:border-slate-800/40 dark:bg-slate-950`,`w-[min(42rem,50vw)]`),children:[(0,u.jsxs)(`div`,{className:`flex h-14 flex-shrink-0 items-center justify-between border-b border-slate-200/30 px-4 dark:border-slate-800/40`,children:[(0,u.jsxs)(`div`,{className:`flex min-w-0 items-center gap-2`,children:[(0,u.jsx)(`span`,{className:`text-sm font-semibold text-slate-900 dark:text-slate-100`,children:`Artifact`}),(0,u.jsx)(`span`,{className:`rounded-md bg-slate-100 px-1.5 py-0.5 font-mono text-[11px] text-slate-500 dark:bg-slate-900 dark:text-slate-400`,children:n})]}),(0,u.jsxs)(`div`,{className:`flex items-center gap-1.5`,children:[(0,u.jsxs)(`button`,{type:`button`,onClick:()=>{let e=new Blob([t],{type:`text/html;charset=utf-8`}),r=URL.createObjectURL(e),i=document.createElement(`a`);i.href=r,i.download=`artifact.${n===`svg`?`svg`:`html`}`,document.body.appendChild(i),i.click(),document.body.removeChild(i),URL.revokeObjectURL(r)},className:`inline-flex items-center gap-1.5 rounded-lg px-2.5 py-1.5 text-xs font-medium text-slate-600 transition hover:bg-slate-100 dark:text-slate-300 dark:hover:bg-slate-900`,title:`Download`,children:[(0,u.jsx)(r,{className:`h-3.5 w-3.5`}),`Download`]}),(0,u.jsx)(`button`,{type:`button`,onClick:d,className:`rounded-lg p-1.5 text-slate-500 transition hover:bg-slate-100 hover:text-slate-900 dark:text-slate-400 dark:hover:bg-slate-900 dark:hover:text-slate-100`,"aria-label":`Close Artifact`,title:`Close`,children:(0,u.jsx)(s,{className:`h-4 w-4`})})]})]}),(0,u.jsx)(`div`,{className:`min-h-0 flex-1`,children:(0,u.jsx)(`iframe`,{ref:f,srcDoc:p,sandbox:`allow-scripts allow-downloads`,title:`Artifact Preview`,className:`h-full w-full border-0 bg-white`})})]})}export{d as ArtifactsPanel};