[RESOLVED] Restart problem (old server isn't killed)

[Grzegorz Derebecki]

i will test it on ruby becous my lsapi for php is older

[Grzegorz Derebecki]

it won't help

server exits and leave lsapi process (ID: 16273) and

in4max@core2:~% ps aux |grep RAILS
in4max 16506 25.4  1.8 728968 152608  ??  S     8:07PM   0:16.78 ruby: RAILS: fdb.pl (production ID:16496) (ruby)
in4max 16507 22.7  1.8 728968 151572  ??  S     8:07PM   0:30.46 ruby: RAILS: fdb.pl (production ID:16496) (ruby)
in4max 16505 15.0  1.8 727944 149984  ??  S     8:07PM   0:18.65 ruby: RAILS: fdb.pl (production ID:16496) (ruby)
in4max 16273  0.0  1.8 721800 145036  ??  Ss    8:06PM   0:04.22 ruby: RAILS: fdb.pl (production ID:16273) (ruby)
in4max 16281  0.0  1.8 727944 149996  ??  S     8:06PM   0:06.69 ruby: RAILS: fdb.pl (production ID:16273) (ruby)
in4max 16282  0.0  1.8 727944 150796  ??  S     8:06PM   0:07.13 ruby: RAILS: fdb.pl (production ID:16273) (ruby)
in4max 16283  0.0  1.8 728968 152244  ??  S     8:06PM   0:14.94 ruby: RAILS: fdb.pl (production ID:16273) (ruby)
in4max 16496  0.0  1.8 721800 145040  ??  Ss    8:07PM   0:04.32 ruby: RAILS: fdb.pl (production ID:16496) (ruby)
in4max 16508  0.0  1.8 727944 149956  ??  S     8:07PM   0:07.46 ruby: RAILS: fdb.pl (production ID:16496) (ruby)

when i truss them are killed:

root@core2:/var/db/mysql# truss -p 16281
clock_gettime(4,{4022.010929191 })		 = 0 (0x0)
select(1,{0},0x0,0x0,{0.774455 })		 = 0 (0x0)
getppid(0x1,0x7fffffffe0e0,0x0,0x0,0x7fffffffdff0,0x0) = 17682 (0x4512)
sigaction(SIGINT,{ SIG_DFL SA_RESTART ss_t },{ 0x45ee24 SA_RESTART ss_t }) = 0 (0x0)
close(3)					 = 0 (0x0)
close(6)					 = 0 (0x0)
close(6)					 ERR#9 'Bad file descriptor'
write(5,"\^A\0\0\0\^A",5)			 = 5 (0x5)
shutdown(5,SHUT_RDWR)				 = 0 (0x0)
close(5)					 = 0 (0x0)
sigprocmask(SIG_BLOCK,SIGHUP|SIGINT|SIGQUIT|SIGKILL|SIGPIPE|SIGALRM|SIGTERM|SIGURG|SIGSTOP|SIGTSTP|SIGCONT|SIGCHLD|SIGTTIN|SIGTTOU|SIGIO|SIGXCPU|SIGXFSZ|SIGVTALRM|SIGPROF|SIGWINCH|SIGINFO|SIGUSR1|SIGUSR2,0x0) = 0 (0x0)
sigprocmask(SIG_SETMASK,0x0,0x0)		 = 0 (0x0)
...

[mistwang]

There should be one ruby process with parent PID of 1, that's the one started by LSWS, the other worker ruby processes are forked by that process.
So, check that ruby process with ktrace/kdump, see if it is doing getppid(), and the result.

The configuration change only affect newly started ruby processes, so you may need to restart LSWS the second time to check the effect.

[Grzegorz Derebecki]

still sometimes all lsapi process arent turned off. Why server exit before it checkc that all sapi process are killed?

(i meen root process not www with not have rights to kill sapi process)

This is lived lsapi process:

root@core2:/opt/lsws/fcgi-bin# truss -p 68635
clock_gettime(13,{1265367432.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367432.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367432.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367432.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367432.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367432.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367432.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367432.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367432.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)
nanosleep({0.100000000 })			 = 0 (0x0)
clock_gettime(13,{1265367433.000000000 })	 = 0 (0x0)

if i do kill 68635 it property kills all childs. Why litespeed don't killed them?

[mistwang]

Are you sure you removed "LSAPI_PPID_NO_CHECK" completely from env, instead of set it to "0"?
As long as "LSAPI_PPID_NO_CHECK" is present, regardless value, LSAPI will not check PPID.

lshttpd process that starts the LSAPI app in suEXEC mode cannot kill it due to permission issue, only the parent lshttpd process running as "root" can, but it has exit already at this stage.

[Grzegorz Derebecki]

Originally Posted by mistwang

Are you sure you removed "LSAPI_PPID_NO_CHECK" completely from env, instead of set it to "0"?
As long as "LSAPI_PPID_NO_CHECK" is present, regardless value, LSAPI will not check PPID.

Hmm right i set it to 0 strange

lshttpd process that starts the LSAPI app in suEXEC mode cannot kill it due to permission issue, only the parent lshttpd process running as "root" can, but it has exit already at this stage.

of corse parent should to it i think. But i will test it with completly removed LSAPI_PPID_NO_CHECK

[Grzegorz Derebecki]

Ok after few days tests all working ok