feat(pool): grant deterministic worker leases
This commit is contained in:
parent
d4cb582e45
commit
9c31e7f90c
1 changed files with 11 additions and 2 deletions
|
|
@ -1,7 +1,7 @@
|
|||
#!/usr/bin/env python3
|
||||
"""Loopback-first compute-pool entry with authenticated worker heartbeats."""
|
||||
from __future__ import annotations
|
||||
import json, os, secrets, ssl, threading
|
||||
import json, os, secrets, ssl, threading, time, uuid
|
||||
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
|
||||
from pathlib import Path
|
||||
|
||||
|
|
@ -50,7 +50,16 @@ class Handler(BaseHTTPRequestHandler):
|
|||
if node != peer: self.send_json(403,{'error':'WORKER_CERT_ID_MISMATCH'}); return
|
||||
state['workers'][node]={**payload,'last_seen':payload.get('observed_at'),'authority_granted':False}; state['registered_workers']=len(state['workers']); write_state(state); self.send_json(200,{'outcome':'HEARTBEAT_ACCEPTED','node_id':node,'authority_granted':False}); return
|
||||
if self.path=='/v1/lease/request':
|
||||
self.send_json(200,{'outcome':'LEASE_QUEUED','state':'PENDING_DETERMINISTIC_SELECTION','authority_granted':False}); return
|
||||
min_cpu=int(payload.get('min_cpu_cores',1)); min_mem=int(payload.get('min_memory_mb',256)); candidates=[]
|
||||
for worker in state['workers'].values():
|
||||
metrics=worker.get('metrics') or {}
|
||||
if worker.get('state')!='ACTIVE_WORKER': continue
|
||||
if int(metrics.get('cpu_cores') or 0)<min_cpu: continue
|
||||
if int(metrics.get('memory_available_bytes') or 0)<min_mem*1024*1024: continue
|
||||
candidates.append(worker)
|
||||
if not candidates: self.send_json(409,{'outcome':'NO_ELIGIBLE_WORKER','state':'LEASE_NOT_GRANTED','authority_granted':False}); return
|
||||
candidates.sort(key=lambda worker:(int((worker.get('metrics') or {}).get('cpu_cores') or 0),int((worker.get('metrics') or {}).get('memory_available_bytes') or 0)))
|
||||
node=candidates[0]['node_id']; lease_id=f'LEASE-{uuid.uuid4()}'; lease={'lease_id':lease_id,'node_id':node,'task_id':payload.get('task_id','UNNAMED_TASK'),'created_at':time.time(),'expires_at':time.time()+900,'state':'GRANTED','authority_granted':False}; state['leases'][lease_id]=lease; write_state(state); self.send_json(200,{'outcome':'LEASE_GRANTED','state':'GRANTED','lease':lease,'authority_granted':False}); return
|
||||
self.send_json(404,{'error':'NOT_FOUND'})
|
||||
def log_message(self,*args): pass
|
||||
|
||||
|
|
|
|||
Loading…
Reference in a new issue