Re: [Condor-users] Can't find transition out of state "START" for event "CHILD_EXIT"

[Matthew Farrellee <matt@xxxxxxxxxx>]

On 01/11/2011 03:15 AM, Carsten Aulbert wrote:
> Hi all,
>
> anyone with an idea what this message wants to tell me?
>
> 10.12.0.20 us the local IP of the execute node where this was found in
> StarterLog.slot3
>
> 01/11 08:33:47 ********** STARTER starting up ***********
> 01/11 08:33:47 ** $CondorVersion: 7.4.4 Oct 13 2010 BuildID: 279383 $
> 01/11 08:33:47 ** $CondorPlatform: X86_64-LINUX_DEBIAN50 $
> 01/11 08:33:47 ******************************************
> 01/11 08:33:47 Submitting machine is "atlas2.atlas.local"
> 01/11 08:33:47 EventHandler {
> 01/11 08:33:47  func = 0x4fb7ca
> 01/11 08:33:47  mask = SIGALRM SIGHUP SIGINT SIGUSR1 SIGUSR2 SIGCHLD SIGTSTP
> 01/11 08:33:47 }
> 01/11 08:33:47 condor_write(): Socket closed when trying to write 25 bytes to
> <10.12.0.20:0>, fd is 17
> 01/11 08:33:47 Buf::write(): condor_write() failed
> Stack dump for process 4844 at timestamp 1294731227 (13 frames)
> condor_starter(dprintf_dump_stack+0xb7)[0x4f6eb5]
> condor_starter[0x4f7122]
> /lib/libpthread.so.0[0x7f860d925a80]
> /lib/libc.so.6(gsignal+0x35)[0x7f860ce4fed5]
> /lib/libc.so.6(abort+0x183)[0x7f860ce513f3]
> /lib/libc.so.6(__assert_fail+0xe9)[0x7f860ce48dc9]
> condor_starter(REMOTE_CONDOR_register_fs_domain+0x195)[0x4d76d4]
> condor_starter(_Z21init_environment_infov+0x26)[0x4c0ee7]
> condor_starter(_Z4initv+0xe)[0x4c2176]
> condor_starter(_ZN12StateMachine7executeEv+0x20e)[0x4fbad6]
> condor_starter(main+0x11d)[0x4c231d]
> /lib/libc.so.6(__libc_start_main+0xe6)[0x7f860ce3c1a6]
> condor_starter(_ZNSt8ios_base4InitD1Ev+0x59)[0x4c0c09]
> 01/11 08:33:47 ERROR "Can't find transition out of state "START" for event
> "CHILD_EXIT"" at line 326 in file state_machine_driver.cpp
>
> Cheers
>
> Carsten

I can tell you that it is unique to the standard universe starter. The 
starter.std uses a general state machine structure to define its 
operation. That error comes from when an event (CHILD_EXIT) is not 
defined as a possibly transition for a state (START).

If I had to guess, which is the case, I'd say the message doesn't matter 
much because the starter probably lost its connection to the shadow 
during startup. You weren't going to make much progress anyway. You 
should check the submit side to see if I'm right.

It's probably also a bug, where a transition for a failed shadow should 
be added to the state machine, but there isn't much the starter could do 
to to compensate except maybe exit quietly.

Best,


matt