feat(workbench): persist operational workflows and events

This commit is contained in:
2026-09-24 04:19:02 +08:00
parent baff7dfe9e
commit 9335bdc1ea
66 changed files with 16580 additions and 238 deletions
+570
View File
@@ -0,0 +1,570 @@
"""Project-scoped APIs for operational tasks, evidence and approved actions."""
from __future__ import annotations
from datetime import datetime
from typing import Literal
from uuid import UUID
from fastapi import APIRouter, Depends, Header, HTTPException, Query, status
from psycopg import AsyncConnection
from app.auth.permissions import (
OPERATIONAL_TASK_EXECUTE,
OPERATIONAL_TASK_EXTERNAL_SYNC,
OPERATIONAL_TASK_VIEW,
OPERATIONAL_TASK_WRITE,
permissions_for_context,
require_permission,
)
from app.auth.project_dependencies import (
ProjectContext,
get_project_context,
get_project_pg_connection,
)
from app.domain.domain_errors import DomainError
from app.domain.schemas.operational_data import TaskDataQuery, TaskDataSnapshot
from app.domain.schemas.operational_task import (
ActionDecisionCommand,
ActionExecuteCommand,
ActionProposalCommand,
ActionReconcileCommand,
AgentArtifactsCommand,
AgentArtifactsResult,
AgentRunCancelCommand,
AgentRunCommand,
AgentRunDispatchFailureCommand,
AgentRunFailureCommand,
AgentTaskContext,
DispatchDueSchedulesCommand,
DispatchDueSchedulesResult,
EvidenceRecord,
ExternalFeedbackCommand,
ExternalFeedbackResult,
HumanApprovalAction,
HumanEvidenceCommand,
NotificationAcknowledgeCommand,
OperationalTaskCreate,
OperationalTaskDetail,
OperationalTaskNotification,
OperationalTaskNotificationPage,
OperationalTaskPage,
OperationalTaskRetryCommand,
OperationalTaskRun,
OperationalTaskRunPage,
OperationalTaskSchedule,
OperationalTaskScheduleCreate,
OperationalTaskSchedulePage,
OperationalTaskScheduleUpdate,
RecoverExpiredLeasesCommand,
RecoverExpiredLeasesResult,
TaskStatus,
TaskDataQueryResult,
TaskType,
)
from app.infra.db.postgresql.operational_tasks import OperationalTaskRepository
from app.infra.integrations.operational_actions import (
OperationalActionDispatcher,
get_operational_action_dispatcher,
)
from app.infra.integrations.operational_data import TaskDataError, read_task_data
from app.services.operational_tasks import OperationalTaskService
router = APIRouter()
def _service(ctx: ProjectContext, conn: AsyncConnection) -> OperationalTaskService:
return OperationalTaskService(
OperationalTaskRepository(conn, ctx.project_id),
ctx,
permissions_for_context(ctx),
)
def _raise_domain(exc: DomainError):
raise HTTPException(status_code=exc.status, detail=exc.payload()) from exc
@router.post(
"/operational-tasks",
response_model=OperationalTaskDetail,
status_code=status.HTTP_201_CREATED,
)
async def create_operational_task(
command: OperationalTaskCreate,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_WRITE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).create(command)
except DomainError as exc:
_raise_domain(exc)
@router.get("/operational-tasks", response_model=OperationalTaskPage)
async def list_operational_tasks(
task_type: TaskType | None = None,
task_status: TaskStatus | None = Query(default=None, alias="status"),
limit: int = Query(default=50, ge=1, le=100),
offset: int = Query(default=0, ge=0),
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
items, next_offset = await _service(ctx, conn).list(
task_type=task_type, status=task_status, limit=limit, offset=offset
)
return OperationalTaskPage(
items=items, limit=limit, offset=offset, next_offset=next_offset
)
except DomainError as exc:
_raise_domain(exc)
@router.get("/operational-task-runs", response_model=OperationalTaskRunPage)
async def list_operational_task_runs(
run_status: Literal[
"queued", "claimed", "running", "succeeded", "failed", "timed_out", "cancelled"
] | None = Query(default=None, alias="status"),
limit: int = Query(default=50, ge=1, le=100),
cursor: str | None = None,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
cursor_created_at = None
cursor_run_id = None
if cursor:
try:
timestamp, run_id = cursor.rsplit("|", 1)
cursor_created_at = datetime.fromisoformat(timestamp)
cursor_run_id = UUID(run_id)
except (ValueError, TypeError) as exc:
raise HTTPException(status_code=400, detail="Invalid run cursor") from exc
try:
items, next_cursor = await _service(ctx, conn).list_runs(
status=run_status,
limit=limit,
cursor_created_at=cursor_created_at,
cursor_run_id=cursor_run_id,
)
return OperationalTaskRunPage(items=items, next_cursor=next_cursor)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/schedules",
response_model=OperationalTaskSchedule,
status_code=status.HTTP_201_CREATED,
)
async def create_operational_schedule(
command: OperationalTaskScheduleCreate,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_WRITE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).create_schedule(command)
except DomainError as exc:
_raise_domain(exc)
@router.get(
"/operational-tasks/schedules", response_model=OperationalTaskSchedulePage
)
async def list_operational_schedules(
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return OperationalTaskSchedulePage(items=await _service(ctx, conn).list_schedules())
except DomainError as exc:
_raise_domain(exc)
@router.patch(
"/operational-tasks/schedules/{schedule_id}",
response_model=OperationalTaskSchedule,
)
async def update_operational_schedule(
schedule_id: UUID,
command: OperationalTaskScheduleUpdate,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_WRITE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).update_schedule(schedule_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/schedules/dispatch-due",
response_model=DispatchDueSchedulesResult,
)
async def dispatch_due_operational_schedules(
command: DispatchDueSchedulesCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).dispatch_due_schedules(command.limit)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/leases/recover",
response_model=RecoverExpiredLeasesResult,
)
async def recover_operational_task_leases(
command: RecoverExpiredLeasesCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).recover_expired_leases(command.limit)
except DomainError as exc:
_raise_domain(exc)
@router.get(
"/operational-task-notifications",
response_model=OperationalTaskNotificationPage,
)
async def list_operational_task_notifications(
unread_only: bool = False,
limit: int = Query(default=50, ge=1, le=100),
offset: int = Query(default=0, ge=0),
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
items, next_offset = await _service(ctx, conn).list_notifications(
unread_only=unread_only, limit=limit, offset=offset
)
return OperationalTaskNotificationPage(
items=items, limit=limit, offset=offset, next_offset=next_offset
)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-task-notifications/{notification_id}/acknowledge",
response_model=OperationalTaskNotification,
)
async def acknowledge_operational_task_notification(
notification_id: UUID,
command: NotificationAcknowledgeCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).acknowledge_notification(notification_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.get("/operational-tasks/{task_id}", response_model=OperationalTaskDetail)
async def get_operational_task(
task_id: UUID,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).detail(task_id)
except DomainError as exc:
_raise_domain(exc)
@router.post("/operational-tasks/{task_id}/retry", response_model=OperationalTaskDetail)
async def retry_operational_task(
task_id: UUID,
command: OperationalTaskRetryCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_WRITE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).retry_task(task_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/runs/{run_id}/claim",
response_model=OperationalTaskRun,
)
async def claim_operational_task_run(
task_id: UUID,
run_id: UUID,
command: AgentRunCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).claim_run(task_id, run_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/runs/{run_id}/heartbeat",
response_model=OperationalTaskRun,
)
async def heartbeat_operational_task_run(
task_id: UUID,
run_id: UUID,
command: AgentRunCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).heartbeat_run(task_id, run_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/runs/{run_id}/complete",
response_model=OperationalTaskDetail,
)
async def complete_operational_task_run(
task_id: UUID,
run_id: UUID,
command: AgentRunCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).complete_run(task_id, run_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/runs/{run_id}/fail",
response_model=OperationalTaskDetail,
)
async def fail_operational_task_run(
task_id: UUID,
run_id: UUID,
command: AgentRunFailureCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).fail_run(task_id, run_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/runs/{run_id}/cancel",
response_model=OperationalTaskDetail,
)
async def cancel_operational_task_run(
task_id: UUID,
run_id: UUID,
command: AgentRunCancelCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_WRITE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).cancel_run(task_id, run_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/runs/{run_id}/dispatch-failure",
response_model=OperationalTaskRun,
)
async def record_operational_task_run_dispatch_failure(
task_id: UUID,
run_id: UUID,
command: AgentRunDispatchFailureCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).record_dispatch_failure(task_id, run_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.get(
"/operational-tasks/{task_id}/agent-context", response_model=AgentTaskContext
)
async def get_operational_task_agent_context(
task_id: UUID,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).agent_context(task_id)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/agent-artifacts",
response_model=AgentArtifactsResult,
)
async def publish_operational_task_artifacts(
task_id: UUID,
command: AgentArtifactsCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).publish_artifacts(task_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/data-query", response_model=TaskDataQueryResult
)
async def query_operational_task_data(
task_id: UUID,
query: TaskDataQuery,
authorization: str | None = Header(default=None, alias="Authorization"),
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXECUTE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
service = _service(ctx, conn)
try:
await service.detail(task_id)
snapshot = await read_task_data(
context=ctx, query=query, credential=authorization
)
evidence, stored_snapshot = await service.record_data_snapshot(task_id, snapshot)
return TaskDataQueryResult(evidence=evidence, snapshot=stored_snapshot)
except TaskDataError as exc:
raise HTTPException(status_code=exc.status, detail=exc.detail) from exc
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/human-evidence",
response_model=OperationalTaskDetail,
)
async def add_operational_task_human_evidence(
task_id: UUID,
command: HumanEvidenceCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_WRITE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).add_human_evidence(task_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.get(
"/operational-tasks/{task_id}/evidence/{evidence_id}/snapshot",
response_model=TaskDataSnapshot,
)
async def get_operational_task_data_snapshot(
task_id: UUID,
evidence_id: str,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).data_snapshot(task_id, evidence_id)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/actions",
response_model=HumanApprovalAction,
status_code=status.HTTP_201_CREATED,
)
async def propose_operational_task_action(
task_id: UUID,
command: ActionProposalCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_WRITE)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).propose_action(task_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/actions/{action_id}/decision",
response_model=HumanApprovalAction,
)
async def decide_operational_task_action(
task_id: UUID,
action_id: UUID,
command: ActionDecisionCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).decide_action(task_id, action_id, command)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/actions/{action_id}/execute",
response_model=HumanApprovalAction,
)
async def execute_operational_task_action(
task_id: UUID,
action_id: UUID,
command: ActionExecuteCommand,
authorization: str | None = Header(default=None, alias="Authorization"),
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
dispatcher: OperationalActionDispatcher = Depends(get_operational_action_dispatcher),
):
try:
return await _service(ctx, conn).submit_action(
task_id, action_id, command, dispatcher, authorization
)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/actions/{action_id}/reconcile",
response_model=HumanApprovalAction,
)
async def reconcile_operational_task_action(
task_id: UUID,
action_id: UUID,
command: ActionReconcileCommand,
authorization: str | None = Header(default=None, alias="Authorization"),
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
dispatcher: OperationalActionDispatcher = Depends(get_operational_action_dispatcher),
):
try:
return await _service(ctx, conn).reconcile_action(
task_id, action_id, command, dispatcher, authorization
)
except DomainError as exc:
_raise_domain(exc)
@router.post(
"/operational-tasks/{task_id}/external-feedback",
response_model=ExternalFeedbackResult,
)
async def synchronize_operational_task_feedback(
task_id: UUID,
command: ExternalFeedbackCommand,
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_EXTERNAL_SYNC)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
try:
return await _service(ctx, conn).external_feedback(task_id, command)
except DomainError as exc:
_raise_domain(exc)
+153
View File
@@ -0,0 +1,153 @@
"""Authenticated workbench snapshot and resumable Server-Sent Events."""
from __future__ import annotations
import asyncio
import json
from datetime import datetime, timezone
from fastapi import APIRouter, Depends, Header, HTTPException, Query
from fastapi.responses import StreamingResponse
from psycopg import AsyncConnection
from psycopg.conninfo import conninfo_to_dict
from app.api.v1.endpoints.workbench_runtime import workbench_runtime
from app.auth.permissions import OPERATIONAL_TASK_VIEW, permissions_for_context, require_permission
from app.auth.project_dependencies import (
ProjectContext,
get_metadata_repository,
get_project_business_routing,
get_project_pg_connection,
)
from app.core.config import settings
from app.domain.schemas.workbench_events import WorkbenchEvent, WorkbenchSyncSnapshot
from app.infra.db.postgresql.operational_tasks import OperationalTaskRepository
from app.infra.db.postgresql.workbench_events import WorkbenchEventRepository
from app.infra.db.metadb.repositories.metadata_repository import MetadataRepository
from app.infra.db.project_routing import ActiveProjectRouting
from app.services.operational_tasks import OperationalTaskService
from app.infra.db.postgresql.workbench_event_hub import workbench_event_hubs
router = APIRouter()
def _service(ctx: ProjectContext, conn: AsyncConnection) -> OperationalTaskService:
return OperationalTaskService(
OperationalTaskRepository(conn, ctx.project_id),
ctx,
permissions_for_context(ctx),
)
@router.get("/workbench/sync-snapshot", response_model=WorkbenchSyncSnapshot)
async def workbench_sync_snapshot(
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
conn: AsyncConnection = Depends(get_project_pg_connection),
):
async with conn.transaction(isolation_level="repeatable read", read_only=True):
cursor = await WorkbenchEventRepository(conn, ctx.project_id).current_cursor()
runtime = await workbench_runtime(ctx=ctx, conn=conn)
tasks, _ = await _service(ctx, conn).list(
task_type=None, status=None, limit=50, offset=0
)
notifications, _ = await _service(ctx, conn).list_notifications(
unread_only=False, limit=100, offset=0
)
return WorkbenchSyncSnapshot(
schema="workbench-sync-snapshot@1",
event_cursor=cursor,
runtime=runtime,
tasks=tasks,
notifications=notifications,
generated_at=datetime.now(timezone.utc),
)
@router.get("/workbench/events")
async def workbench_events(
after: int | None = Query(default=None, ge=0),
last_event_id: str | None = Header(default=None, alias="Last-Event-ID"),
ctx: ProjectContext = Depends(require_permission(OPERATIONAL_TASK_VIEW)),
routing: ActiveProjectRouting = Depends(get_project_business_routing),
metadata_repo: MetadataRepository = Depends(get_metadata_repository),
):
if not settings.WORKBENCH_EVENTS_ENABLED:
raise HTTPException(status_code=503, detail="Workbench events are disabled")
cursor = after
if cursor is None and last_event_id:
try:
cursor = int(last_event_id)
except ValueError as exc:
raise HTTPException(status_code=400, detail="Invalid Last-Event-ID") from exc
cursor = cursor or 0
current_identity = _database_identity(routing.business_dsn)
try:
active_routings = await metadata_repo.list_active_project_db_routings("biz_data")
finally:
# FastAPI keeps yield dependencies alive for a StreamingResponse. The
# stream only needs immutable routing values, so release the shared
# metadata checkout before the five-minute SSE body begins.
await metadata_repo.session.close()
if any(
item.project_id != ctx.project_id
and _database_identity(item.dsn) == current_identity
for item in active_routings
):
raise HTTPException(
status_code=503,
detail="Workbench events require one dedicated business database per project",
)
async def stream():
started = asyncio.get_running_loop().time()
last_keepalive = started
async with workbench_event_hubs.subscribe(ctx.project_id, routing.business_dsn) as hub:
oldest = await hub.oldest_cursor()
latest = await hub.current_cursor()
if cursor > 0 and (
oldest is None or cursor < oldest - 1 or cursor > latest
):
payload = {"schema": "workbench-event@1", "cursor": cursor}
yield f"event: reset_required\ndata: {json.dumps(payload, separators=(',', ':'))}\n\n"
return
current = cursor
wake_generation = hub.generation
while asyncio.get_running_loop().time() - started < 300:
rows = await hub.read_after(current)
for row in rows:
event = WorkbenchEvent(
schema="workbench-event@1",
**row,
)
current = event.event_seq
data = event.model_dump_json(by_alias=True)
yield f"id: {current}\nevent: workbench_event\ndata: {data}\n\n"
if rows:
continue
now = asyncio.get_running_loop().time()
if now - last_keepalive >= 15:
yield f": keepalive {int(now)}\n\n"
last_keepalive = now
# LISTEN is only a wake-up path. The timeout also reconciles
# against the durable cursor if a notification was missed.
wake_generation = await hub.wait_for_wake(wake_generation, timeout=1)
yield "event: reconnect\ndata: {}\n\n"
return StreamingResponse(
stream(),
media_type="text/event-stream",
headers={
"Cache-Control": "no-cache, no-transform",
"X-Accel-Buffering": "no",
"Content-Encoding": "identity",
},
)
def _database_identity(dsn: str) -> tuple[str, str, str]:
values = conninfo_to_dict(dsn)
return (
values.get("host") or values.get("hostaddr") or "localhost",
values.get("port") or "5432",
values.get("dbname") or values.get("user") or "",
)