Re: slapd deadlocked in test008 (hdb) (ITS#2580)

[h.b.furuseth@usit.uio.no]

Someone kick me, please...

I didn't even notice the two latest threads, doing select().

'truss -p' shows that slapd is looping, doing the system calls
   poll(0xFE001108, 20, 0)                         = 0
   yield()                                         = 0
   time()                                          = 1054845428
A select() loop with no timeout and no intervening sleep().

Here is another GDB session with the same process.  All but the two
bottom thread were the same.  These two were doing yield().

Thread 2 (Thread 4 (LWP 1)):
#0  0xfef9c3a0 in yield () from /usr/lib/libc.so.1
#1  0x00080c64 in ldap_pvt_thread_yield () at thr_posix.c:183
#2  0x0002e670 in slapd_daemon_task (ptr=0x3edfacd7) at daemon.c:1892

Thread 1 (LWP    1        ):
#0  0xfef9c3a0 in yield () from /usr/lib/libc.so.1
#1  0x00080c64 in ldap_pvt_thread_yield () at thr_posix.c:183
#2  0x0002e670 in slapd_daemon_task (ptr=0x3edfacd7) at daemon.c:1892
#0  0xfef9c3a0 in yield () from /usr/lib/libc.so.1
(gdb) n
Single stepping until exit from function yield,
which has no line number information.
ldap_pvt_thread_yield () at thr_posix.c:199
199    }
(gdb) n
slapd_daemon_task (ptr=0x3edfacd7) at daemon.c:1258
1258            now = slap_get_time();
(gdb)
1260            if( ( global_idletimeout > 0 ) &&
(gdb)
1266            tv = idle;
(gdb) p idle
$1 = {tv_sec = 0, tv_usec = 0}
(gdb) n
1269            if( slapd_gentle_shutdown ) {
(gdb) n
1289            FD_ZERO( &writefds );
(gdb) n
1290            FD_ZERO( &readfds );
(gdb)
1292            at = 0;
(gdb)
1294            ldap_pvt_thread_mutex_lock( &slap_daemon.sd_mutex );
(gdb)
1306            AC_MEMCPY( &readfds, &slap_daemon.sd_readers, sizeof(fd_set) );
(gdb)
1307            AC_MEMCPY( &writefds, &slap_daemon.sd_writers, sizeof(fd_set) );
(gdb)
1310            FD_SET( wake_sds[0], &readfds );
(gdb) p readfds
$2 = {fds_bits = {268434496, 0 <repeats 31 times>}}
(gdb) set radix 16
Input and output radices now set to decimal 16, hex 10, octal 20.
(gdb) p readfds
$3 = {fds_bits = {0xffffc40, 0x0 <repeats 31 times>}}
(gdb) p writefds
$4 = {fds_bits = {0x0 <repeats 32 times>}}
(gdb) n
1312            for ( l = 0; slap_listeners[l] != NULL; l++ ) {
(gdb) p wake_sds[0]
$5 = 0x4
(gdb) p readfds
$6 = {fds_bits = {0xffffc50, 0x0 <repeats 31 times>}}
(gdb) n
1313                if ( slap_listeners[l]->sl_sd == AC_SOCKET_INVALID )
(gdb)
1315                if ( slap_listeners[l]->sl_is_mute )
(gdb)
1318                if (!FD_ISSET(slap_listeners[l]->sl_sd, &readfds))
(gdb)
1323            nfds = slap_daemon.sd_nfds;
(gdb)
1327            if ( global_idletimeout && slap_daemon.sd_nactives )
(gdb)
1330            ldap_pvt_thread_mutex_unlock( &slap_daemon.sd_mutex );
(gdb)
1332            if ( !at ) {
(gdb)
1333                at = ldap_pvt_thread_pool_backload(&connection_pool) -
(gdb)
1337            tvp = at ? &tv : NULL;
(gdb) p at
$7 = 0x12
(gdb) p tv
$8 = {tv_sec = 0x0, tv_usec = 0x0}
(gdb) n
1340            ldap_pvt_thread_mutex_lock( &syncrepl_rq.rq_mutex );
(gdb) n
1341            rtask = ldap_pvt_runqueue_next_sched( &syncrepl_rq, &cat );
(gdb) n
1342            while ( cat && cat->tv_sec && cat->tv_sec <= now ) {
(gdb)
1355            ldap_pvt_thread_mutex_unlock( &syncrepl_rq.rq_mutex );
(gdb)
1357            if ( cat != NULL ) {
(gdb)
1394                NULL, tvp ))
(gdb) n
1892            ldap_pvt_thread_yield();
(gdb) detach

-- 
Hallvard