I think there is a bug in windows port (are you on windows?) that doesn't set recursive attribute for the to_send mutex. Please open a jira:
https://issues.apache.org/jira/browse/ZOOKEEPER Thanks! --Michi On Fri, Jun 8, 2012 at 1:00 AM, 乱麻的魅力 <[email protected]> wrote: > hi dev： > I now try to use the zookeeper cli （c code version）to connect the > zookeeper server, but i find only can connect to ZK,but cann't send any cmd > to ZK, like "ls /". if i send cmd ,then zk-cli goto deadlock at this line > lock_buffer_list(list) {//LINE 00945 dequeue_buffer() function of > zookeeper.c}; then i try to locate this case. > > i download the zk cli (ver 3.4.3) from > http://labs.renren.com/apache-mirror/zookeeper/ , buid the project again, > find bug locate the line 00945 in > zookeeper-3.4.3.tar.gz\zookeeper-3.4.3\src\c\src\zookeeper.c too. now i > describe this case below: > > 1 if client send cmd to ZKserver, client need call some function to send the > cmd ,like zoo_awget,send_ping,zoo_aget,etc.., all this function need call > adaptor_send_queue(zh, 0); then below... > > 2 adaptor_send_queue(zh, 0) call flush_send_queue(zh, timeout); > > int flush_send_queue(zhandle_t*zh, int timeout) > { > int rc= ZOK; > struct timeval started; > #ifdef WIN32 > fd_set pollSet; > struct timeval wait; > #endif > gettimeofday(&started,0); > // we can't use dequeue_buffer() here because if (non-blocking) > send_buffer() > // returns EWOULDBLOCK we'd have to put the buffer back on the queue. > // we use a recursive lock instead and only dequeue the buffer if a send > was > // successful > lock_buffer_list(&zh->to_send); /*first time lock the buffer, wfs > 20120608 */ > while (zh->to_send.head != 0&& zh->state == ZOO_CONNECTED_STATE) { > if(timeout!=0){ > int elapsed; > struct timeval now; > gettimeofday(&now,0); > elapsed=calculate_interval(&started,&now); > if (elapsed>timeout) { > rc = ZOPERATIONTIMEOUT; > break; > } > #ifdef WIN32 > wait = get_timeval(timeout-elapsed); > FD_ZERO(&pollSet); > FD_SET(zh->fd, &pollSet); > // Poll the socket > rc = select((int)(zh->fd)+1, NULL, &pollSet, NULL, &wait); > #else > struct pollfd fds; > fds.fd = zh->fd; > fds.events = POLLOUT; > fds.revents = 0; > rc = poll(&fds, 1, timeout-elapsed); > #endif > if (rc<=0) { > /* timed out or an error or POLLERR */ > rc = rc==0 ? ZOPERATIONTIMEOUT : ZSYSTEMERROR; > break; > } > } > rc = send_buffer(zh->fd, zh->to_send.head); > if(rc==0 && timeout==0){ > /* send_buffer would block while sending this buffer */ > rc = ZOK; > break; > } > if (rc < 0) { > rc = ZCONNECTIONLOSS; > break; > } > // if the buffer has been sent successfully, remove it from the queue > if (rc > 0) > remove_buffer(&zh->to_send); /*this function will second time lock > the buffer with lock under locked status, wfs 20120608 */ > > gettimeofday(&zh->last_send, 0); > rc = ZOK; > } > unlock_buffer_list(&zh->to_send); > return rc; > } > > static int remove_buffer(buffer_head_t *list) > { > buffer_list_t *b = dequeue_buffer(list); > if (!b) { > return 0; > } > free_buffer(b); > return 1; > } > > static buffer_list_t *dequeue_buffer(buffer_head_t *list) > { > buffer_list_t *b; > lock_buffer_list(list); /*this function second time lock the buffer with > lock under locked status 20120608 , then will lead the function to deadlock > at this line; > > if i re-write a new function like *dequeue_buffer(buffer_head_t *list) > and remove_buffer function without lock and unlock to be callback by > flush_send_queue, then zk-cli can send cmd to the zookkeeper server, clie > don't deadlock*/ > > b = list->head; > if (b) { > list->head = b->next; > if (!list->head) { > assert(b == list->last); > list->last = 0; > } > } > unlock_buffer_list(list); > return b; > } > > i don't known whether I detailly describe this case, and i find old version > 3.3.3 have this bug too,i think this c source-code maybe never be tested or i > use wrong way, can you help me clear this case。 > > thanks！ > wfs fr china 20120608
