feat(host-protocol): renew and complete leases

This commit is contained in:
2026-07-12 18:15:04 +08:00
parent c7f8d0c128
commit a24efc1043
4 changed files with 242 additions and 1 deletions
@@ -2,10 +2,12 @@ from __future__ import annotations
import asyncio
from collections.abc import Awaitable, Callable
from datetime import timedelta
from time import monotonic
from typing import TYPE_CHECKING
from fastapi import APIRouter, HTTPException, Request, status
from fastapi.responses import JSONResponse
from cloud.auth import (
AuthProvider,
@@ -17,6 +19,11 @@ from cloud.internal_api.models import (
ClaimResponse,
HeartbeatRequest,
HeartbeatResponse,
LeaseRenewalRequest,
LeaseRenewalResponse,
StaleLeaseConflict,
TerminalResultRequest,
TerminalResultResponse,
)
from core.models import Device, utc_now
@@ -30,10 +37,13 @@ def create_internal_router(
auth_provider: AuthProvider,
version_prefix: str = "/internal/v1",
claim_poll_interval_seconds: float = 0.1,
lease_duration_seconds: float = 60.0,
sleep: Callable[[float], Awaitable[None]] = asyncio.sleep,
) -> APIRouter:
if claim_poll_interval_seconds <= 0:
raise ValueError("claim_poll_interval_seconds must be greater than zero")
if lease_duration_seconds <= 0:
raise ValueError("lease_duration_seconds must be greater than zero")
router = APIRouter(prefix=version_prefix, tags=["host-agent"])
def authorize_host(request: Request, host_id: str) -> None:
@@ -117,9 +127,102 @@ def create_internal_router(
return ClaimResponse(timed_out=True)
await sleep(min(claim_poll_interval_seconds, remaining))
@router.post(
"/hosts/{host_id}/assignments/{task_id}/renew",
response_model=LeaseRenewalResponse,
responses={status.HTTP_409_CONFLICT: {"model": StaleLeaseConflict}},
)
def renew_assignment(
host_id: str,
task_id: str,
payload: LeaseRenewalRequest,
request: Request,
):
authorize_host(request, host_id)
_validate_assignment_identity(
host_id=host_id,
task_id=task_id,
payload_host_id=payload.host_id,
payload_task_id=payload.task_id,
)
now = utc_now()
lease_expires_at = now + timedelta(seconds=lease_duration_seconds)
renewal_status = pool.store.renew_lease(
task_id=task_id,
attempt=payload.attempt,
lease_id=payload.lease_id,
host_id=host_id,
lease_expires_at=lease_expires_at,
now=now,
)
if renewal_status == "not_found":
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail="assignment not found",
)
if renewal_status != "renewed":
return _stale_lease_conflict("assignment lease is stale or expired")
return LeaseRenewalResponse(
status="renewed",
lease_expires_at=lease_expires_at,
)
@router.post(
"/hosts/{host_id}/assignments/{task_id}/result",
response_model=TerminalResultResponse,
responses={status.HTTP_409_CONFLICT: {"model": StaleLeaseConflict}},
)
def report_result(
host_id: str,
task_id: str,
payload: TerminalResultRequest,
request: Request,
):
authorize_host(request, host_id)
_validate_assignment_identity(
host_id=host_id,
task_id=task_id,
payload_host_id=payload.host_id,
payload_task_id=payload.task_id,
)
result_status = pool.store.record_task_result(
task_id=task_id,
attempt=payload.attempt,
lease_id=payload.lease_id,
host_id=host_id,
status=payload.status,
failure_reason=payload.failure_reason,
terminal_result=payload.result,
completed_at=utc_now(),
)
if result_status == "conflict":
return _stale_lease_conflict("assignment lease is stale or superseded")
return TerminalResultResponse(status=result_status)
return router
def _validate_assignment_identity(
*,
host_id: str,
task_id: str,
payload_host_id: str,
payload_task_id: str,
) -> None:
if payload_host_id != host_id or payload_task_id != task_id:
raise HTTPException(
status_code=status.HTTP_422_UNPROCESSABLE_CONTENT,
detail="assignment identity must match the request path",
)
def _stale_lease_conflict(detail: str) -> JSONResponse:
return JSONResponse(
status_code=status.HTTP_409_CONFLICT,
content=StaleLeaseConflict(detail=detail).model_dump(),
)
def _validate_snapshot(
pool: DevicePool,
*,
@@ -70,3 +70,8 @@ class TerminalResultRequest(BaseModel):
class TerminalResultResponse(BaseModel):
status: Literal["recorded", "already_recorded"]
class StaleLeaseConflict(BaseModel):
code: Literal["stale_lease"] = "stale_lease"
detail: str