[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] condor termsig = 11 on checkpoint restart

Hmmm, perhaps my description is too vague to attract a response.

Let's start with this. How can I manually test if checkpointing is working? I'd like to test checkpointing a job on a FC3 node and restarting it on a RH7.3 node. Again, I am using condor 6.7.7 installed from rpms.


On Jun 2, 2005, at 6:38 PM, Naveen Neelakantam wrote:


First let my start off by saying that I have been a very happy condor user for over a year now. However, a few of our machines got upgraded to Fedora Core 3 and all hell has broken loose on our condor install. Now, I am a frustrated condor user who is desperately trying to be happy again. :-)

I am having problems with condor version 6.7.7 and standard universe jobs exiting with signal 11 after restarting from a checkpoint. Every time the problem occurs I see something like the following in the shadow log:

6/2 17:26:37 (10.5) (312):Read: Done restoring file state
6/2 17:26:37 (10.5) (312):Read: About to restore signal state
6/2 17:26:37 (10.5) (312):Read: About to return to user code
6/2 17:26:37 (10.5) (312):Shadow: Job 10.5 exited, termsig = 11, coredump = 0, retcode = 0
6/2 17:26:37 (10.5) (312):Shadow: was killed by signal 11.

Specifically the problem only occurs when a job running on a Fedora Core 3 node gets checkpointed and restarted on one of our Redhat Linux 7.3 nodes (and possibly vice versa).

Some info about our setup:
-Negotiator/Collector is running Redhat Enterprise Linux AS release 3 and is not a member of the pool (not running condor_startd)
-compute nodes are either running Redhat Linux 7.3 or Fedora Core (1 and 3)
-all nodes were installed using the 6.7.7 dynamic lib rpm's (glibc22 or glibc23 where appropriate)

I had previously been using condor version 6.6.9 with the USER_JOB_WRAPPER variable set as mentioned in order to workaround the 2.6 kernel exec_shield problem. However, I would still see the same symptoms (termsig = 11 on checkpoint restart).

In order to test if checkpointing works from the command line I tried the following:
fedora_node> setarch i386 ./myapp
fedora_node> kill -TSTP <myapp_pid>
RHL_7_3_node> ./myapp -_condor_restart myapp.ckpt
And the app segfaulted.

Any help would be greatly appreciated!

Condor-users mailing list