feat: add durable WorldQuant backtests with UI and AI confirmation

This commit is contained in:
yuxuanhui
2026-09-08 10:06:00 +08:00
parent 404a4d8a04
commit a4b93200c5
34 changed files with 4437 additions and 23 deletions
+607
View File
@@ -0,0 +1,607 @@
"""Transactional research interface. Callers own authorization and commit boundaries."""
from collections import Counter, defaultdict
from uuid import uuid4
from fastapi import HTTPException
from fastapi.encoders import jsonable_encoder
from sqlalchemy import func, select, update
from ..models import (
Account,
BacktestConfig,
BacktestDraft,
BacktestEvent,
BacktestItem,
BacktestPreview,
BacktestResult,
BacktestRun,
SimulationAttempt,
now,
)
from .contracts import Candidate, DraftInput, PreviewInput, Source, fingerprint, group_key
def uid():
return str(uuid4())
async def event(db, run, kind, payload):
"""Append a run-local cursor under the run row lock, in the result's transaction."""
run.event_seq += 1
run.updated_at = now()
db.add(BacktestEvent(run_id=run.id, seq=run.event_seq, kind=kind, payload=payload))
async def locked_run(db, run_id):
run = await db.scalar(select(BacktestRun).where(BacktestRun.id == run_id).with_for_update())
if not run:
raise HTTPException(404, "回测运行不存在")
return run
async def refresh_status(db, run):
await db.flush()
states = list(await db.scalars(select(SimulationAttempt.state).where(SimulationAttempt.run_id == run.id)))
if any(s in ("needs_review", "collection_failed") for s in states):
run.status = "needs_review"
elif all(s in ("completed", "failed", "skipped") for s in states):
run.status = (
"stopped"
if run.control == "stopped"
else "completed_with_errors"
if "failed" in states
else "completed"
)
elif run.control == "paused":
run.status = "paused"
elif run.control == "stopped":
run.status = "stopping"
elif any(s in ("submitting", "submitted", "collecting") for s in states):
run.status = "running"
else:
run.status = "queued"
class Backtests:
def __init__(self, db, ai_context=None):
self.db = db
self.ai_context = ai_context or {}
async def config(self):
row = await self.db.get(BacktestConfig, 1)
return jsonable_encoder(
{
k: getattr(row, k)
for k in ("concurrency", "batch_size", "version", "blocked_reason", "blocked_until")
}
)
async def configure(self, body):
result = await self.db.execute(
update(BacktestConfig)
.where(BacktestConfig.id == 1, BacktestConfig.version == body.version)
.values(
concurrency=body.concurrency,
batch_size=body.batch_size,
version=BacktestConfig.version + 1,
)
)
if result.rowcount != 1:
raise HTTPException(409, "调度配置已变化,请刷新后重试")
return await self.config()
async def capabilities(self):
return {
"alpha_types": ["REGULAR"],
"languages": ["FASTEXPR"],
"instrument_types": ["EQUITY"],
"settings_schema": Candidate.model_json_schema(),
"scheduler": await self.config(),
"max_candidates": 10000,
"remote_cancel": False,
"automatic_history_reuse": False,
"confirmation": "每个固定运行确认一次;启动后返回 ID,不循环等待",
"mapping": "完整输入匹配;证据不足待核对,不按 children 顺序匹配",
"limits": "并发和批大小为本地配置,并非平台可用额度;账户外提交不在预算内",
}
async def save_draft(self, body, draft_id=None):
data = body.model_dump(mode="json", exclude={"version"})
if draft_id:
changed = await self.db.execute(
update(BacktestDraft)
.where(BacktestDraft.id == draft_id, BacktestDraft.version == body.version)
.values(
**data,
version=BacktestDraft.version + 1,
updated_at=now(),
)
)
if changed.rowcount != 1:
raise HTTPException(409, "草稿已变化或不存在;保留当前编辑并重新载入")
else:
draft_id = uid()
self.db.add(BacktestDraft(id=draft_id, **data))
await self.db.flush()
return await self.draft(draft_id)
async def drafts(self, limit=25, offset=0):
rows = (
await self.db.scalars(
select(BacktestDraft)
.order_by(BacktestDraft.updated_at.desc(), BacktestDraft.id)
.limit(limit)
.offset(offset)
)
).all()
return {
"items": [
jsonable_encoder(
{
"id": r.id,
"version": r.version,
"name": r.name,
"total": len(r.candidates),
"updated_at": r.updated_at,
}
)
for r in rows
],
"total": await self.db.scalar(select(func.count()).select_from(BacktestDraft)),
"limit": limit,
"offset": offset,
}
async def draft(self, draft_id):
row = await self.db.get(BacktestDraft, draft_id)
if not row:
raise HTTPException(404, "候选草稿不存在")
return jsonable_encoder(
{k: getattr(row, k) for k in ("id", "version", "name", "source", "candidates", "updated_at")}
)
async def preview(self, body):
if body.inline:
data = body.inline.model_dump(mode="json")
else:
draft = await self.db.scalar(
select(BacktestDraft).where(BacktestDraft.id == body.draft_id).with_for_update()
)
if not draft or draft.version != body.draft_version:
raise HTTPException(409, "候选草稿已变化,请重新准备预览")
candidates = draft.candidates
if body.selection is not None:
selection = set(body.selection)
candidates = [c for c in candidates if c["client_item_id"] in selection]
if len(candidates) != len(selection):
raise HTTPException(422, "选择包含不属于当前草稿的候选")
data = {"name": draft.name, "source": draft.source, "candidates": candidates}
candidates = DraftInput.model_validate(data).model_dump(mode="json")["candidates"]
config = await self.db.get(BacktestConfig, 1)
groups = defaultdict(list)
hashes = []
for i, c in enumerate(candidates):
groups[group_key(c)].append(i)
hashes.append(fingerprint(Candidate.model_validate(c).platform_input()))
# Query hashes in bounded chunks, including SQLite's bind-parameter limit.
existing = set()
for index in range(0, len(hashes), 400):
existing.update(
await self.db.scalars(
select(BacktestItem.fingerprint)
.where(BacktestItem.fingerprint.in_(hashes[index : index + 400]))
.distinct()
)
)
seen, duplicates = set(), []
for c, h in zip(candidates, hashes):
if h in seen or h in existing:
duplicates.append(
{
"client_item_id": c["client_item_id"],
"historical": h in existing,
"within_preview": h in seen,
}
)
seen.add(h)
batches = []
for indices in groups.values():
local_batches = []
for index in indices:
batch = next(
(
b
for b in local_batches
if len(b) < config.batch_size and all(hashes[i] != hashes[index] for i in b)
),
None,
)
if batch is None:
batch = []
local_batches.append(batch)
batch.append(index)
batches.extend(local_batches)
row = BacktestPreview(
id=uid(),
name=data["name"],
source=data["source"],
candidates=candidates,
batches=batches,
batch_size=config.batch_size,
digest=fingerprint({"candidates": candidates, "source": data["source"]}),
duplicates=duplicates,
ai_context=self.ai_context,
)
self.db.add(row)
await self.db.flush()
return await self.get_preview(row.id)
async def get_preview(self, preview_id, limit=25, offset=0):
row = await self.db.get(BacktestPreview, preview_id)
if not row:
raise HTTPException(404, "回测预览不存在")
return jsonable_encoder(
{
"preview_id": row.id,
"version": row.version,
"name": row.name,
"source": row.source,
"digest": row.digest,
"total": len(row.candidates),
"batch_count": len(row.batches),
"batch_size": row.batch_size,
"duplicate_count": len(row.duplicates),
"duplicates": row.duplicates[offset : offset + limit],
"items": row.candidates[offset : offset + limit],
"limit": limit,
"offset": offset,
"has_more": offset + limit < len(row.candidates),
"created_at": row.created_at,
}
)
async def start(self, body):
# One account row serializes all starts; unique keys remain the final DB invariant.
account = await self.db.scalar(select(Account).where(Account.id == 1).with_for_update())
previous = await self.db.scalar(
select(BacktestRun).where(BacktestRun.idempotency_key == body.idempotency_key)
)
if previous:
if previous.preview_id != body.preview_id or body.version != 1:
raise HTTPException(409, "幂等键已用于另一份预览")
return await self.run(previous.id)
preview = await self.db.get(BacktestPreview, body.preview_id)
if not preview or preview.version != body.version:
raise HTTPException(409, "预览不存在或版本不匹配")
previous = await self.db.scalar(select(BacktestRun).where(BacktestRun.preview_id == preview.id))
if previous:
return await self.run(previous.id)
if not account or not account.wq_user_id or account.connection_status != "connected":
raise HTTPException(409, "请先连接并确认 WorldQuant 账户身份")
run = BacktestRun(
id=uid(),
preview_id=preview.id,
idempotency_key=body.idempotency_key,
name=preview.name,
source=preview.source,
total=len(preview.candidates),
batch_size=preview.batch_size,
ai_context=self.ai_context or preview.ai_context,
event_seq=0,
)
self.db.add(run)
await self.db.flush()
for n, indices in enumerate(preview.batches):
candidates = [Candidate.model_validate(preview.candidates[i]) for i in indices]
attempt = SimulationAttempt(
id=uid(), run_id=run.id, ordinal=n, payload=[c.platform_input() for c in candidates]
)
self.db.add(attempt)
await self.db.flush()
for i, c in zip(indices, candidates):
self.db.add(
BacktestItem(
id=uid(),
run_id=run.id,
attempt_id=attempt.id,
ordinal=i,
client_item_id=c.client_item_id,
expression=c.expression,
settings=c.settings.model_dump(),
fingerprint=fingerprint(c.platform_input()),
)
)
await event(self.db, run, "created", {"total": run.total, "batch_count": len(preview.batches)})
await self.db.flush()
return await self.run(run.id)
async def runs(self, limit=25, offset=0, source=None):
query = select(BacktestRun)
if source:
query = query.where(BacktestRun.source["kind"].as_string() == source)
total = await self.db.scalar(select(func.count()).select_from(query.subquery()))
rows = (
await self.db.scalars(
query.order_by(BacktestRun.created_at.desc(), BacktestRun.id).limit(limit).offset(offset)
)
).all()
return {
"items": [await self.run(r.id) for r in rows],
"total": total,
"limit": limit,
"offset": offset,
}
async def run(self, run_id):
row = await self.db.get(BacktestRun, run_id)
if not row:
raise HTTPException(404, "回测运行不存在")
groups = (
await self.db.execute(
select(
BacktestItem.platform_status,
BacktestItem.collection_status,
BacktestItem.persistence_status,
func.count(),
)
.where(BacktestItem.run_id == run_id)
.group_by(
BacktestItem.platform_status,
BacktestItem.collection_status,
BacktestItem.persistence_status,
)
)
).all()
counts = {"platform": Counter(), "collection": Counter(), "persistence": Counter()}
for p, c, s, n in groups:
for key, value in (("platform", p), ("collection", c), ("persistence", s)):
counts[key][value] += n
return jsonable_encoder(
{
"backtest_run_id": row.id,
**{
k: getattr(row, k)
for k in (
"preview_id",
"name",
"source",
"ai_context",
"control",
"status",
"version",
"total",
"batch_size",
"created_at",
"updated_at",
)
},
"counts": counts,
"cursor": row.event_seq,
"scheduler": await self.config(),
}
)
async def results(self, run_id, limit=25, offset=0):
run = await self.run(run_id)
rows = (
await self.db.execute(
select(BacktestItem, BacktestResult)
.outerjoin(BacktestResult, BacktestResult.item_id == BacktestItem.id)
.where(BacktestItem.run_id == run_id)
.order_by(BacktestItem.ordinal)
.limit(limit)
.offset(offset)
)
).all()
return jsonable_encoder(
{
"backtest_run_id": run_id,
"total": run["total"],
"limit": limit,
"offset": offset,
"items": [
{
**{
k: getattr(i, k)
for k in (
"id",
"client_item_id",
"expression",
"settings",
"attempt_id",
"platform_status",
"collection_status",
"persistence_status",
"simulation_id",
"alpha_id",
"error",
)
},
"result": {
"snapshot": r.snapshot,
"observed_at": r.observed_at,
"complete": r.complete,
}
if r
else None,
}
for i, r in rows
],
}
)
async def events(self, run_id, after=0, limit=100):
await self.run(run_id)
rows = (
await self.db.scalars(
select(BacktestEvent)
.where(BacktestEvent.run_id == run_id, BacktestEvent.seq > after)
.order_by(BacktestEvent.seq)
.limit(limit + 1)
)
).all()
return jsonable_encoder(
{
"items": [
{"seq": r.seq, "kind": r.kind, "payload": r.payload, "created_at": r.created_at}
for r in rows[:limit]
],
"next_cursor": rows[min(len(rows), limit) - 1].seq if rows else after,
"has_more": len(rows) > limit,
}
)
async def attempts(self, run_id):
await self.run(run_id)
rows = (
await self.db.scalars(
select(SimulationAttempt)
.where(SimulationAttempt.run_id == run_id)
.order_by(SimulationAttempt.ordinal)
)
).all()
return jsonable_encoder(
[
{
k: getattr(a, k)
for k in (
"id",
"state",
"ordinal",
"progress_url",
"remote_complete",
"children",
"error",
"error_code",
"poll_count",
"submit_count",
"next_poll_at",
)
}
for a in rows
]
)
async def control(self, run_id, body):
run = await locked_run(self.db, run_id)
if run.version != body.version:
raise HTTPException(409, "运行控制已变化,请重新确认")
attempts = (
await self.db.scalars(select(SimulationAttempt).where(SimulationAttempt.run_id == run_id))
).all()
if body.action == "recover":
for a in attempts:
if a.state in ("needs_review", "collection_failed") and a.progress_url:
if len(a.children) != len(a.payload):
# Re-enumerate missing children while retaining collected receipts/results.
a.children = []
a.state, a.poll_count, a.next_poll_at, a.error, a.error_code = (
"submitted",
0,
None,
None,
None,
)
# Recovery never clears uncertain submissions or creates a new POST.
elif body.action == "resume":
if run.control == "stopped":
raise HTTPException(409, "已停止的剩余项不能恢复,请生成重跑预览")
run.control = "active"
config = await self.db.scalar(
select(BacktestConfig).where(BacktestConfig.id == 1).with_for_update()
)
# An explicit resume may clear an indefinite quota block, never a Retry-After deadline.
if config.blocked_until is None:
config.blocked_reason = None
elif body.action == "pause":
if run.control == "stopped":
raise HTTPException(409, "该运行已经停止")
run.control = "paused"
else:
run.control = "stopped"
for a in attempts:
if a.state == "queued":
a.state = "skipped"
await self.db.execute(
update(BacktestItem)
.where(BacktestItem.attempt_id == a.id)
.values(
platform_status="skipped",
collection_status="not_required",
persistence_status="not_required",
)
)
run.version += 1
await refresh_status(self.db, run)
await event(self.db, run, "control", {"action": body.action, "control": run.control})
await self.db.flush()
return await self.run(run_id)
async def rerun(self, run_id, body):
run = await locked_run(self.db, run_id)
rows = (
await self.db.scalars(
select(BacktestItem).where(BacktestItem.run_id == run_id).order_by(BacktestItem.ordinal)
)
).all()
selected = [r for r in rows if r.id in set(body.item_ids)]
if len(selected) != len(set(body.item_ids)):
raise HTTPException(422, "重跑项不属于指定运行")
if any(r.platform_status not in ("completed", "failed", "skipped") for r in selected):
raise HTTPException(409, "仍在执行或结果未知的项须先核对,不能直接重跑")
return await self.preview(
PreviewInput(
inline=DraftInput(
name=f"{run.name[:190]} · 重跑",
source=Source.model_validate({**run.source, "parent_run_id": run.id}),
candidates=[
Candidate(
client_item_id=r.client_item_id, expression=r.expression, settings=r.settings
)
for r in selected
],
)
)
)
async def attach_reference(self, attempt_id, body):
"""Record a human-supplied original simulation; collection still verifies its input."""
a = await self.db.get(SimulationAttempt, attempt_id)
if not a:
raise HTTPException(404, "执行尝试不存在")
run = await locked_run(self.db, a.run_id)
if run.version != body.version or a.state != "needs_review" or a.progress_url:
raise HTTPException(409, "执行状态已变化或已有平台引用,请重新读取")
duplicate = await self.db.scalar(
select(SimulationAttempt.id).where(SimulationAttempt.progress_url == body.progress_url)
)
if duplicate:
raise HTTPException(409, "此模拟引用已经关联其他执行尝试")
a.progress_url, a.state, a.error, a.error_code = body.progress_url, "submitted", None, None
a.next_poll_at, a.poll_count = None, 0
run.version += 1
await self.db.execute(
update(BacktestItem)
.where(BacktestItem.attempt_id == a.id)
.values(platform_status="submitted", error=None)
)
await refresh_status(self.db, run)
await event(
self.db, run, "reference_attached", {"attempt_id": a.id, "progress_url": body.progress_url}
)
return await self.run(run.id)
async def subset(self, preview_id, body):
parent = await self.db.get(BacktestPreview, preview_id)
if not parent:
raise HTTPException(404, "预览不存在")
excluded = set(body.exclude_ids)
if not excluded.issubset({c["client_item_id"] for c in parent.candidates}):
raise HTTPException(422, "排除集合包含未知候选")
candidates = [c for c in parent.candidates if c["client_item_id"] not in excluded]
if not candidates:
raise HTTPException(422, "至少保留一条候选")
return await self.preview(
PreviewInput(inline=DraftInput(name=parent.name, source=parent.source, candidates=candidates))
)