test_rpc_tracker.py 3.17 KB
Newer Older
1 2 3 4 5 6 7 8 9 10
import tvm
import logging
import numpy as np
import time
import multiprocessing
from tvm.contrib import rpc

def check_server_drop():
    """test when server drops"""
    try:
11
        from tvm.contrib.rpc import tracker, proxy, base
12 13 14 15 16 17 18 19 20 21 22
        from tvm.contrib.rpc.base import TrackerCode

        @tvm.register_func("rpc.test2.addone")
        def addone(x):
            return x + 1

        def _put(tclient, value):
            base.sendjson(tclient._sock, value)
            base.recvjson(tclient._sock)

        tserver = tracker.Tracker("localhost", 8888)
23 24
        tproxy = proxy.Proxy("localhost", 8881,
                             tracker_addr=("localhost", tserver.port))
25
        tclient = rpc.connect_tracker("localhost", tserver.port)
26 27 28 29 30

        server0 = rpc.Server(
            "localhost", port=9099,
            tracker_addr=("localhost", tserver.port),
            key="abc")
31 32 33 34 35
        server1 = rpc.Server(
            "localhost", port=9099,
            tracker_addr=("localhost", tserver.port),
            key="xyz")
        server2 = rpc.Server(
36
            "localhost", tproxy.port, is_proxy=True,
37
            key="xyz")
38 39 40
        server3 = rpc.Server(
            "localhost", tproxy.port, is_proxy=True,
            key="xyz1")
41

42 43 44 45
        # Fault tolerence to un-handled requested value
        _put(tclient, [TrackerCode.REQUEST, "abc", "", 1])
        _put(tclient, [TrackerCode.REQUEST, "xyz1", "", 1])

46 47 48
        # Fault tolerence to stale worker value
        _put(tclient, [TrackerCode.PUT, "xyz", (server1.port, "abc")])
        _put(tclient, [TrackerCode.PUT, "xyz", (server1.port, "abcxxx")])
49
        _put(tclient, [TrackerCode.PUT, "xyz", (tproxy.port, "abcxxx11")])
50 51 52

        # Fault tolerence server timeout
        def check_timeout(timeout, sleeptime):
53 54 55 56 57 58 59 60 61
            def myfunc(remote):
                time.sleep(sleeptime)
                f1 = remote.get_function("rpc.test2.addone")
                assert f1(10) == 11
            try:
                tclient.request_and_run("xyz", myfunc, session_timeout=timeout)
            except RuntimeError:
                pass
            print(tclient.text_summary())
62 63 64 65 66 67 68 69
            try:
                remote = tclient.request("xyz", priority=0, session_timeout=timeout)
                remote2 = tclient.request("xyz", session_timeout=timeout)
                time.sleep(sleeptime)
                f1 = remote.get_function("rpc.test2.addone")
                assert f1(10) == 11
                f1 = remote2.get_function("rpc.test2.addone")
                assert f1(10) == 11
70

71 72
            except tvm.TVMError as e:
                pass
73 74 75 76 77
            remote3 = tclient.request("abc")
            f1 = remote3.get_function("rpc.test2.addone")
            remote3 = tclient.request("xyz1")
            f1 = remote3.get_function("rpc.test2.addone")
            assert f1(10) == 11
78

79 80
        check_timeout(0.01, 0.1)
        check_timeout(2, 0)
81
        tserver.terminate()
82
        server0.terminate()
83
        server1.terminate()
84
        server2.terminate()
85 86
        server3.terminate()
        tproxy.terminate()
87 88 89 90 91 92 93
    except ImportError:
        print("Skip because tornado is not available")


if __name__ == "__main__":
    logging.basicConfig(level=logging.INFO)
    check_server_drop()