feat: extend operational investigation workflows
Generic Container CI/CD / test-build-publish (push) Successful in 3m53s
Server CI/CD v2 / build-test-publish-and-deploy (push) Successful in 3m52s

This commit is contained in:
2026-10-30 17:30:19 +08:00
parent 63af998045
commit 05183aff93
82 changed files with 4644 additions and 330 deletions
+17 -8
View File
@@ -39,12 +39,15 @@ def envelope(source: str, data, *, job_id="list", page=None) -> QueryResult:
def public_job(row: dict) -> dict:
fields = {"job_id", "status", "tool", "created_at", "started_at", "finished_at", "updated_at", "error"}
fields = {"job_id", "status", "tool", "created_at", "started_at", "finished_at", "updated_at", "error", "queue_expires_at", "result_expires_at"}
from fastapi.encoders import jsonable_encoder
return {**jsonable_encoder({key: row[key] for key in fields if key in row}), "poll_after_ms": 1500}
expiry = row.get("result_expires_at")
available = row.get("status") == "succeeded" and (expiry is None or expiry > datetime.now(timezone.utc))
return {**jsonable_encoder({key: row[key] for key in fields if key in row}),
"result_available": available, "poll_after_ms": 1500}
async def submit_calculation(scope, binding, tool, args, call_id: str) -> QueryResult:
async def submit_calculation(scope, binding, tool, args, call_id: str, *, job_timeout_seconds: int = 3600) -> QueryResult:
from app.mcp.query_service import authorize_query
await authorize_query(scope, binding, tool, args)
if binding.purpose == "task_reader":
@@ -68,7 +71,7 @@ async def submit_calculation(scope, binding, tool, args, call_id: str) -> QueryR
if not retryable_capacity_rejection(previous["result"]):
raise HTTPException(409, "Operation already started; inspect the existing calculation or mutation status")
check_digest(args, await preview_change(scope, binding, tool.name, args))
parameters = args.model_dump(mode="json", exclude={"context_handle"})
parameters = args.model_dump(mode="json", exclude={"context_handle", "timeout_seconds"})
request_id = str(args.request_id) if tool.mutating else call_id
key = str(uuid5(NAMESPACE_URL, ":".join([binding.subject, binding.client_id, binding.native_session_id, str(binding.task_id), str(binding.run_id), request_id])))
repo = CalculationJobRepository(scope.connection, scope.project.project_id)
@@ -84,12 +87,12 @@ async def submit_calculation(scope, binding, tool, args, call_id: str) -> QueryR
await verify_task(scope, binding)
row = await repo.enqueue(
owner_user_id=scope.project.user_id, request_key=key, tool=tool.name,
parameters=parameters, binding=durable_binding(binding), timeout_seconds=args.timeout_seconds,
parameters=parameters, binding=durable_binding(binding), timeout_seconds=job_timeout_seconds,
)
else:
row = await repo.enqueue(
owner_user_id=scope.project.user_id, request_key=key, tool=tool.name,
parameters=parameters, binding=durable_binding(binding), timeout_seconds=args.timeout_seconds,
parameters=parameters, binding=durable_binding(binding), timeout_seconds=job_timeout_seconds,
)
await scope.connection.commit()
return envelope(tool.name, public_job(row), job_id=row["job_id"])
@@ -107,7 +110,9 @@ async def authorize_original(scope, row):
from app.mcp.contexts import verify_task
original = restore_binding(row["binding"])
tool = TOOLS[row["tool"]]
args = tool.model.model_validate({**row["parameters"], "context_handle": "durable-calculation-context-handle"})
# Legacy queue rows stored the old HTTP timeout. It does not govern durable execution.
parameters = {key: value for key, value in row["parameters"].items() if key != "timeout_seconds"}
args = tool.model.model_validate({**parameters, "context_handle": "durable-calculation-context-handle"})
if (str(scope.project.user_id) != str(row["owner_user_id"])
or str(scope.project.project_id) != str(original.project_id)):
raise HTTPException(403, "Calculation owner is no longer authorized")
@@ -125,7 +130,8 @@ async def read_calculation_job(scope, binding, tool, args) -> QueryResult:
})
rows = await repo.list(owner_user_id=scope.project.user_id, binding=identity_scope, limit=args.limit + 1, offset=args.offset, status=args.status)
return envelope(tool.name, [public_job(row) for row in rows[:args.limit]], page=PageInfo(limit=args.limit, offset=args.offset, has_more=len(rows) > args.limit))
row = await repo.get(args.job_id, owner_user_id=scope.project.user_id)
row = await repo.get(args.job_id, owner_user_id=scope.project.user_id,
include_result=tool.name == "get_calculation_result")
if row is None:
raise HTTPException(404, "Calculation was not found")
check_binding(binding, row)
@@ -137,6 +143,9 @@ async def read_calculation_job(scope, binding, tool, args) -> QueryResult:
await scope.connection.commit()
if tool.name != "get_calculation_result" or row["status"] != "succeeded":
return envelope(tool.name, public_job(row), job_id=row["job_id"])
expiry = row.get("result_expires_at")
if row.get("result") is None or (expiry is not None and expiry <= datetime.now(timezone.utc)):
raise HTTPException(410, "Calculation result retention period has expired")
result = QueryResult.model_validate(row["result"])
result = await attach_calculation_evidence(scope, row, result)
value = result.data