Thanks a lot, Duncan!|
Hope they have a fix for that and will let us know about it. I have found the way how to get rid of this, but it depends on availability of the /tmp partition for the process. When a checkpointed process migrates and restarts on another host, the wrapper process has to write to /tmp the new working directory path. The restarting procress should read this path and perform chdir(2) to it before using getcwd() (which could be redundant in that case).
duncan brown wrote:
Anton Selikhov wrote:I'm trying to use Condor Stand-Alone Checkpointing to migrate tasks between hosts running Linux 2.4.9. Checkpointing is performed by implicit call to ckpt_and_exit(). When restarted, the process executes system function getcwd(), which actually returns the local path of the code before restart, while it restarts in different directory of another host. How to make getcwd() returning the right current working directory?Hi Anton, I have tried the same thing and run into the same problem. The condor team are aware of this limitation with the checkpointing mechanism and I believe that they have a fix for it. I'm not sure when it will be in a release, though. Cheers, Duncan.