From: Fredine, J. <joh...@in...> - 2012-08-28 21:26:33
|
Thanks for the quick replay. I appreciate how responsive you and Gene have been. I'll probably do a hack fix in my application for now, but I am interested in the API you mention and will take a look. JF From: Kapil Arya [mailto:ka...@cc...] Sent: Tuesday, August 28, 2012 2:07 PM To: Fredine, John Cc: dmt...@li... Subject: Re: [Dmtcp-forum] dmtcpaware API shows different save directories in pre/post hook functions Hi John, I am working on making my application dmtcp aware (using dmtcp 1.2.6) because of some troublesome things I won't detail here. The idea is that I use the preCheckpoint hook to save some information about my application and then destroy troublesome items. During the postCheckpoint/postRestart hook functions I then recreate the troublesome items and restore them from the saved state. The obvious (at least to me) place to save the state is in the checkpoint directory. The problem I'm running into is that the checkpoint directory stored in the local status structure returned by dmtcpGetLocalStatus() is different between the preCheckpoint and postCheckpoint hook functions. It appears the checkpoint count is incremented between the preCheckpoint and postCheckpoint hooks. Is this by design? I can work around this, but it's tedious and makes me wonder if there is a reason for doing things this way or if I may be doing something in an undesirable way. First of all, I would encourage you to use the dmtcp-plugin api instead of dmtcpaware for your task. This is a more powerful interface and will have more support in future. At this point, we do support dmtcpaware interface, but we may deprecate it in future. Second, about the problem that you are seeing, this is certainly a bug in DMTCP. I will look into it and write back once I have a fix. Thanks, Kapil |