python-memcached並發調用get/set時出現RunTimeError：Second simultaneous read on fileno 8 detected.

出現異常如下 RuntimeError: Second simultaneous read on fileno 8 detected.  Unless you really know what you're doing, make sure that only one greenthread can read any particular socket.  Consider using a pools.Pool. If you do know what you're doing and want to
disable this error, call eventlet.debug.hub_multiple_reader_prevention(False)

2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake   File "/usr/lib/python2.7/dist-packages/memcache.py", line 862, in get2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake     return self._get('get', key)2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake   File "/usr/lib/python2.7/dist-packages/memcache.py", line 846, in _get2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake     return _unsafe_get()2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake   File "/usr/lib/python2.7/dist-packages/memcache.py", line 830, in _unsafe_get2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake     rkey, flags, rlen, = self._expectvalue(server)2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake   File "/usr/lib/python2.7/dist-packages/memcache.py", line 955, in _expectvalue2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake     line = server.readline()2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake   File "/usr/lib/python2.7/dist-packages/memcache.py", line 1125, in readline2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake     data = recv(4096)2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake   File "/usr/local/lib/python2.7/dist-packages/eventlet/greenio.py", line 249, in recv2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake     timeout_exc=socket.timeout("timed out"))2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake   File "/usr/local/lib/python2.7/dist-packages/eventlet/hubs/__init__.py", line 117, in trampoline2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake     listener = hub.add(hub.READ, fileno, current.switch)2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake   File "/usr/local/lib/python2.7/dist-packages/eventlet/hubs/poll.py", line 27, in add2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake     listener = super(Hub, self).add(evtype, fileno, cb)2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake   File "/usr/local/lib/python2.7/dist-packages/eventlet/hubs/hub.py", line 126, in add2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake     evtype, fileno, evtype))2013-03-08 18:27:27 17870 TRACE glance.store.chunkcache.fake RuntimeError: Second simultaneous read on fileno 8 detected.  Unless you really know what you're doing, make sure that only one greenthread can read any particular socket.  Consider using a pools.Pool. If you do know what you're doing and want to disable this error, call eventlet.debug.hub_multiple_reader_prevention(False)

除了這個之外再還有一堆的memcache.set返回0的情況。

出現問題時的大致模型如下

class CacheBackend(CacheBackendBase):    '''    Cache in memcached servers.    '''    def __init__(self):        value_length = CONF.chunkcache_memcached_value_length        servers = CONF.chunkcache_memcached_servers        self.memcache = memcache.Client(servers, debug=1,                                        server_max_value_length=value_length)    @staticmethod    def instance():        global CACHE_BACKEND        if CACHE_BACKEND is None:            CACHE_BACKEND = CacheBackend()        return CACHE_BACKEND    def get(self, key, default=None):        result = self.memcache.get(str(key))        return result or default    def exist(self, key):        return self.get(str(key)) != None    def set(self, key, value):        result = self.memcache.set(str(key), value)        if result == 0:            raise exception.MemcacheSetError()        return result    def delete(self, key):        if self.exist(key):            result = self.memcache.delete(str(key))            if result == 0:                raise exception.MemcacheDeleteError()        return True    def clear(self):        # No need for clearing memcached.        return True

經過同事交流，推測可能是因為fork之前有socket串連在或者是多個進程重用了一個socket導致，但是感覺都不太能講得通，對這塊確實不太瞭解，先記錄下來解決方案，有誰瞭解的求解釋~

下面是解決的模型，大致改動就是在每個set，get，delete處都new一個memcache client，

可以建立一個socket？不過看memcache模組，好像也是有socket緩衝的啊，會去緩衝中拿同個socket，

下面的disconnect_all加與不加都一樣，不會出現問題，而且在上面的模組中加上disconnect_all再重試也不起作用，

實在無法理解，但是確實解決了問題，多次測試都未複現上述問題。

class CacheBackend(CacheBackendBase):    '''    Cache in memcached servers.    '''    def __init__(self):        self.value_length = CONF.chunkcache_memcached_value_length        self.servers = CONF.chunkcache_memcached_servers    @staticmethod    def instance():        global CACHE_BACKEND        if CACHE_BACKEND is None:            CACHE_BACKEND = CacheBackend()        return CACHE_BACKEND    def get(self, key, default=None):        client = _connect(self.servers, self.value_length)        result = client.get(str(key))        client.disconnect_all()        return result or default    def exist(self, key):        return self.get(str(key)) != None    def set(self, key, value):        client = _connect(self.servers, self.value_length)        result = client.set(str(key), value)        client.disconnect_all()        if result == 0:            raise exception.MemcacheSetError()        return result    def delete(self, key):        client = _connect(self.servers, self.value_length)        if self.exist(key):            result = client.delete(str(key))            if result == 0:                raise exception.MemcacheDeleteError()        client.disconnect_all()        return True    def clear(self):        # No need for clearing memcached.        return Truedef _connect(servers, value_length=chunkstore.Store.CHUNKSIZE):        return memcache.Client(servers, debug=1,                               server_max_value_length=value_length)

另外想解釋下，並不是單例或者維護一個client對象導致，去掉單例，問題一樣，而且並發是多進程並發的，也不應該存在什麼單例之類的問題。