[ZOOKEEPER-591] The C Client cannot exit properly in some situation

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 3.2.1
• Fix Version/s: 3.3.0
• Component/s: c client

Environment

Linux db-passport-test05.vm 2.6.9_5-4-0-5 #1 SMP Tue Apr 14 15:56:24 CST 2009 x86_64 x86_64 x86_64 GNU/Linux

• Hadoop Flags: Reviewed
• Tags: Zookeeper C Client
• Assignee: Mahadev Konar
• Reporter: Qian Ye
• Created: 23/Nov/09 09:35
• Updated: 26/Mar/10 17:25
• Resolved: 18/Mar/10 01:45

Description

The following code produce a situation, where the C Client can not exit properly,

#include "include/zookeeper.h"

void default_zoo_watcher(zhandle_t zzh, int type, int state, const char *path, void context){
int zrc = 0;
struct String_vector str_vec =

{0, NULL};
printf("in the default_zoo_watcher\n");
zrc = zoo_wget_children(zzh, "/mytest", default_zoo_watcher, NULL, &str_vec);
printf("zoo_wget_children, error: %d\n", zrc);

return;
}

int main()
{
int zrc = 0;
int buff_len = 10;
char buff[10] = "hello";
char path[512];
struct Stat stat;
struct String_vector str_vec = {0, NULL}

;

zhandle_t *zh = zookeeper_init("10.81.20.62:2181", NULL, 30000, 0, 0, 0);
zrc = zoo_create(zh, "/mytest", buff, 10, &ZOO_OPEN_ACL_UNSAFE, 0, path, 512);
printf("zoo_create, error: %d\n", zrc);

zrc = zoo_wget_children(zh, "/mytest", default_zoo_watcher, NULL, &str_vec);
printf("zoo_wget_children, error: %d\n", zrc);

zrc = zoo_create(zh, "/mytest/test1", buff, 10, &ZOO_OPEN_ACL_UNSAFE, 0, path, 512);
printf("zoo_create, error: %d\n", zrc);

zrc = zoo_wget_children(zh, "/mytest", default_zoo_watcher, NULL, &str_vec);
printf("zoo_wget_children, error: %d\n", zrc);

zrc = zoo_delete(zh, "/mytest/test1", -1);

printf("zoo_delete, error: %d\n", zrc);
zookeeper_close(zh);
return 0;
}

running this code can cause the program hang at zookeeper_close(zh);(line 38). using gdb to attach the process, I found that the main thread is waiting for do_completion thread to finish,
(gdb) bt
#0 0x000000302b806ffb in pthread_join () from /lib64/tls/libpthread.so.0
#1 0x000000000040de3b in adaptor_finish (zh=0x515b60) at src/mt_adaptor.c:219
#2 0x00000000004060ba in zookeeper_close (zh=0x515b60) at src/zookeeper.c:2100
#3 0x000000000040220b in main ()

and the thread which handle the zoo_wget_children(in the default_zoo_watcher) is waiting for sc->cond.
(gdb) thread 2
[Switching to thread 2 (Thread 1094719840 (LWP 25093))]#0 0x000000302b8089aa in pthread_cond_wait@@GLIBC_2.3.2 ()
from /lib64/tls/libpthread.so.0
(gdb) bt
#0 0x000000302b8089aa in pthread_cond_wait@@GLIBC_2.3.2 () from /lib64/tls/libpthread.so.0
#1 0x000000000040d88b in wait_sync_completion (sc=0x5167f0) at src/mt_adaptor.c:82
#2 0x00000000004082c9 in zoo_wget_children (zh=0x515b60, path=0x40ebc0 "/mytest", watcher=0x401fd8 <default_zoo_watcher>, watcherCtx=Variable "watcherCtx" is not available.)
at src/zookeeper.c:2884
#3 0x0000000000402037 in default_zoo_watcher ()
#4 0x000000000040d664 in deliverWatchers (zh=0x515b60, type=4, state=3, path=0x515100 "/mytest", list=0x5177d8) at src/zk_hashtable.c:274
#5 0x0000000000403861 in process_completions (zh=0x515b60) at src/zookeeper.c:1631
#6 0x000000000040e1b5 in do_completion (v=Variable "v" is not available.) at src/mt_adaptor.c:333
#7 0x000000302b80610a in start_thread () from /lib64/tls/libpthread.so.0
#8 0x000000302afc6003 in clone () from /lib64/tls/libc.so.6
#9 0x0000000000000000 in ?? ()

here, a deadlock presents.