Files
ssw-mpc/ray/armv7/decode_gcs.py
T
rong 00e4fc322b test(ray): 1.0.1 与 2.59 双向不兼容;改做 x86+armv7 混合 1.0.1 集群并跑通
兼容性(双向都测了):
- 1.0.1 raylet → 2.59 head:6379 报 Protocol error, got "\x00"
  (2.59 的 6379 是 gRPC/HTTP2,1.0.1 说 RESP;Ray 2.0 移除了 Redis GCS 后端)
- 2.59 ray start --address=<1.0.1 gcs:6402> 报 RpcError
- 结论:没有中间路线,只能整集群同版本

混合架构集群(x86_64 原生 + armv7 qemu,同为 1.0.1):
- x86 raylet  ID f85f933f…(cpu=4 / nm:6403)
- armv7 raylet ID 9f3e5ae1…(cpu=2 / nm:6413)
- 两侧节点通知是同一对 ID,均 IsAlive=1;armv7 收到 x86 心跳
- GCS 分片表 4 条:2×NODE_RESOURCE + 2×CLIENT,
  CLIENT socket 分别 raylet_a.sock(armv7) / raylet_x.sock(x86)

新增 ray/armv7/{mixed_cluster.sh,compat_probe.sh,decode_gcs.py};
docs/RAY_ARMV7.md 增 §6 兼容性、§7 混合集群、§8 复现命令。
2026-10-30 12:14:26 +08:00

45 lines
1.3 KiB
Python

import socket, sys
def cmd(f, *args):
out = b"*%d\r\n" % len(args)
for a in args:
if isinstance(a, str):
a = a.encode()
out += b"$%d\r\n%s\r\n" % (len(a), a)
f.write(out); f.flush()
def read(f):
line = f.readline()
t, rest = line[:1], line[1:-2]
if t == b'+': return rest
if t == b'-': raise RuntimeError(rest)
if t == b':': return int(rest)
if t == b'$':
n = int(rest)
if n < 0: return None
d = f.read(n + 2)[:-2]
return d
if t == b'*':
n = int(rest)
return [read(f) for _ in range(n)]
raise RuntimeError('bad type ' + repr(t))
s = socket.create_connection(('127.0.0.1', 6401))
f = s.makefile('rwb')
cmd(f, 'DBSIZE'); print("shard 键总数:", read(f))
cmd(f, 'KEYS', '*')
keys = read(f)
print("条目:")
for k in keys:
cmd(f, 'GET', k)
v = read(f)
kind = k.split(b':')[0].decode('ascii', 'replace')
nid = k.split(b':', 1)[1] if b':' in k else b''
print(f" {kind:14s} node={nid.hex()[:24]}… value={len(v or b'')}B")
# 把 value 里的可读片段抽出来(资源表是 protobuf,字符串字段能直接看见)
printable = ''.join(chr(c) if 32 <= c < 127 else ' ' for c in (v or b''))
runs = [t for t in printable.split() if len(t) >= 3]
if runs:
print(f" 可读片段: {' | '.join(runs[:8])}")