fix: harden task recovery and DR contract handling

This commit is contained in:
2026-09-20 13:36:48 +08:00
parent 492676344a
commit f9d8feb6f0
49 changed files with 2083 additions and 165 deletions
+71
View File
@@ -0,0 +1,71 @@
#!/usr/bin/env python3
"""Measure fresh-connection task admission bursts against an isolated local API.
The planner is deliberately idle: receipts must not wait for planning. Latency
is reported as evidence, never asserted as a machine-independent unit-test SLA.
"""
import argparse
from concurrent.futures import ThreadPoolExecutor
import http.client
import json
import math
from pathlib import Path
import sys
import tempfile
import threading
import time
ROOT=Path(__file__).resolve().parents[1]
sys.path.insert(0,str(ROOT/'coordinator'))
from robot_bt_coordinator.backends import ManualBackend
from robot_bt_coordinator.http_api import make_server
from robot_bt_coordinator.service import Coordinator
def probe(clients=10,rounds=3):
if not 1<=clients<=128 or not 1<=rounds<=100:
raise ValueError('clients must be 1..128 and rounds 1..100')
with tempfile.TemporaryDirectory(prefix='bt-http-probe-') as folder:
coordinator=Coordinator(str(Path(folder)/'tasks.db'),ManualBackend(),{'probe_robot'})
server=make_server(coordinator,'127.0.0.1',0,'local-probe-token')
thread=threading.Thread(target=server.serve_forever,daemon=True)
thread.start()
samples=[]
try:
for iteration in range(rounds):
gate=threading.Barrier(clients)
def submit(index):
connection=http.client.HTTPConnection('127.0.0.1',server.server_port,timeout=10)
request=dict(client_request_id=f'probe-{iteration}-{index}',robot_id='probe_robot',instruction='local admission probe',known_info={})
gate.wait(timeout=15)
started=time.perf_counter()
try:
connection.request('POST','/v1/tasks',json.dumps(request),{'Authorization':'Bearer local-probe-token','Content-Type':'application/json'})
response=connection.getresponse()
body=json.loads(response.read())
return dict(elapsed_ms=(time.perf_counter()-started)*1000,http_status=response.status,task_id=body.get('task_id'),status=body.get('status'))
finally:connection.close()
with ThreadPoolExecutor(max_workers=clients) as pool:
samples.extend(pool.map(submit,range(clients)))
successful=all(s['http_status']==202 and s['status']=='QUEUED' for s in samples)
successful=successful and len({s['task_id'] for s in samples})==clients*rounds
durations=sorted(s['elapsed_ms'] for s in samples)
return dict(clients=clients,rounds=rounds,requests=len(samples),all_accepted=successful,
p50_ms=durations[math.ceil(len(durations)*.5)-1],
p95_ms=durations[math.ceil(len(durations)*.95)-1],max_ms=max(durations),
listen_backlog=server.request_queue_size,planner_started=bool(coordinator.backend.plans),samples=samples)
finally:
server.shutdown();server.server_close();thread.join();coordinator.close()
def main():
parser=argparse.ArgumentParser()
parser.add_argument('--clients',type=int,default=10)
parser.add_argument('--rounds',type=int,default=3)
args=parser.parse_args()
result=probe(args.clients,args.rounds)
print(json.dumps(result,indent=2))
if not result['all_accepted'] or result['planner_started']:raise SystemExit(1)
if __name__=='__main__':main()