Deadlock on Shutdown under MacOSX in 1.1

[fceller]

(gdb) thread 1
[Switching to thread 1 (process 81365), "com.apple.main-thread"]
0x00007fff85e41386 in __semwait_signal ()
(gdb) where
#0 0x00007fff85e41386 in __semwait_signal ()
#1 0x00007fff8e30dcbd in pthread_join ()
#2 0x0000000108e5fa73 in TRI_ShutdownLogging () at logging.c:1781
#3 0x0000000108e5b80d in TRI_ShutdownC () at init.c:75
#4 0x0000000108d8ccef in main (argc=4, argv=0x0) at arango.cpp:63

(gdb) thread 2
[Switching to thread 2 (process 81365)]
0x00007fff8e2cbfcd in __findenv ()
(gdb) where
#0 0x00007fff8e2cbfcd in __findenv ()
#1 0x00007fff8e2cc020 in getenv ()
#2 0x00007fff8e2b17e0 in _st_tzset_basic ()
#3 0x00007fff8e2b3091 in localtime_r ()
#4 0x00007fff8e2b31ea in ctime_r ()
#5 0x00007fff8e2973d9 in _asl_time_string ()
#6 0x00007fff8e29730e in asl_msg_to_string_raw ()
#7 0x00007fff8e28fc77 in _asl_send_message ()
#8 0x00007fff8e28f412 in _asl_lib_vlog ()
#9 0x00007fff8e27c839 in vsyslog ()
#10 0x00007fff8e27c7b6 in syslog ()
#11 0x0000000108e600ab in LogAppenderSyslog_Log (appender=0x7fff8e31cfce, level=37, severity=0, msg=0x7fff8e31cfce "TZ", length=4484685360) at logging.c:1485
#12 0x0000000108e5fdf0 in MessageQueueWorker (data=0x7fff8e31cfce) at logging.c:632
#13 0x0000000108e6730b in ThreadStarter (data=0x7f8e11e00000) at threads-posix.c:93
#14 0x00007fff8e283742 in _pthread_start ()
#15 0x00007fff8e270181 in thread_start ()

(gdb) thread 3
[Switching to thread 3 (process 81365)]
0x00007fff85e41386 in __semwait_signal ()
(gdb) where
#0 0x00007fff85e41386 in __semwait_signal ()
#1 0x00007fff8e30d800 in nanosleep ()
#2 0x00007fff8e30d717 in usleep ()
#3 0x0000000108e866e2 in StatisticsLoop () at statistics.cpp:497
#4 0x0000000108e6730b in ThreadStarter (data=0x7f8e11e043b0) at threads-posix.c:93
#5 0x00007fff8e283742 in _pthread_start ()
#6 0x00007fff8e270181 in thread_start ()

[fceller]

Note: I cannot reproduce this problem

[fceller]

Logging should be in GMT

[jsteemann]

Actually we're not calling localtime ourselves but syslog does it internally.
The problem may be caused by syslog being non-reentrant (this is a guess only but is reported occasionally on Google).
I have put a mutex around the syslog calls in 1.1 which hopefully makes it reentrant. I have also changed the behavior of 1.1 to only use the syslog if the server is started with the --logger.syslog option. Previously, the syslog was always used, even if the server was not started with that option.

[jsteemann]

Please reopen if it occurs again.