I'm working on an embedded system whoose details are explained at the end of this post. Tools are ruby on linux.
I put my ruby code on a remote NAS device on network and mount it to the embedded system I'm working on using NFS v3. My project scenario is like this: there is an all-in-one auto starter which 1.mounts the nfs share, 2. starts the webrick ruby code on mounted dir, 3. starts some other biz-logic related ruby code again on mounted dir, 4. starts a ruby code on local file system which does some basic logging and stuff... I put each step in a loop that will retry when a start operation fails (start operations are made by system 'ruby XXXX' or popen3 calls). Here's what happens when I run this starter: step1 goes ok, step 2 fails, step 3 fails, step 4 goes ok, then step2 and 3 are retried and both starts ok too allways at second trial(I want to add that started codes are independent so; it doesnt matter which one starts first and changing the start order didnt make any difference).
I have inspected this problem and found out it also happens when I start things manually from shell and filtered it down to this: My webrick code loads .rb files that are at the same directory as itself and has some FileHandlers using the subdirs in NFS mount and my biz-logic also loads .rb files from nfs mount. And this, I believe, is the source of my problem (I have tried to load those files using both absolute paths like '/mnt/myMount/myLib.rb' and File.Join which both didnt make any difference).
Also, the webrick filehandlers too act pretty strange in this scenario, they always fail at first request of a file and then succeeds at the second request, no matter the file size.
To sum up, as the title says, when a Ruby code on NFS mount uses another file on same mount, ruby gives an EIO (Input Output Error). I was able to reproduce this problem using simple .rb files (that just print itselves filename) loading each other as explained in the above scenario on my environment (though I havent tried this on a regular PC). How can I solve this, is it about the paths given to load/require methods or what?
My working envrionment: The device I'm working with is an intelligent RFID reader, with an embedded linux arm based computer on-board. The only api/language manufacturer allows and provides (w/o getting out of license terms) is ruby, and they give no-support for installing extra-stuff and they have customized the kernel in device to make installing/compiling anything as hard as possible. I can use extra ruby stuff if they're just .rb lib files that is I can easily just include them in my code, but anything that requires compiling is a no-no. Also, this system does not have gem.