Let me first off admit that I have not gotten checkpoint/restart to work with the ckpt-v23-rc1-pids tree. But hopefully this will help make it easier to debug that. Ideally (and eventually), the lxc toolsuite will be used for checkpoint/restart. However, c/r is too unflexible and lxc too flexible, so for now I've just coded some scripts to create little containers. The start_job.sh script takes two arguments, a name for the job, and the program to run. It creates a container with private netns, running an ssh server and a running your job in a screen session. You can checkpoint it with checkpoint.sh, and restart it with restart.sh. kill_job.sh kills the job and cleans up the cgroup and other state. You can get this from https://launchpad.net/~appcr/+archive/ppa, or just put the attached scripts into /usr/bin. They assume nsexec, appcheckpoint, and apprestart exist (yes, bc 'restart' is hijacked by upstart :( ). The next time I can spend some time on c/r, I'll be using these scripts to try and figure out why kernel isn't treating me right. I may go back to the rc1 (not '-pids') version. -serge PS - I did start on creating a 'lxc-cr'-like container template to create checkpointable containers with lxc, but never quite got that to work. Perfecting that seemed inappropriate when sys_checkpoint isn't working anyway.
Attachment:
checkpoint.sh
Description: Bourne shell script
Attachment:
jobinit.sh
Description: Bourne shell script
Attachment:
kill_job.sh
Description: Bourne shell script
Attachment:
restart_real.sh
Description: Bourne shell script
Attachment:
restart.sh
Description: Bourne shell script
Attachment:
start_job.sh
Description: Bourne shell script
_______________________________________________ Containers mailing list Containers@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linux-foundation.org/mailman/listinfo/containers
