A while back I posted a tool I called gnat. I can repost if you can't search the archives.
        Basically it writes a random collection of char data to a file based on a random seed. Then you can supply the known seed and read the file back and compare to make sure things are as they should be.
        This all works of course because of the psuedo random nature of the random number generator.
        One thing we've done here is use this tool as part of a larger set of scripts. It will use the tool to create a bunch of files of various sizes w/ keys that are known for the recovery test.
        Then the plug's pulled in whatever way you want and after the fs has a chance to recover the tool is run against the resultant filesystem to make sure that things are complete.
        One solution.
        I don't htink LTP's used for this type of solution since to my knowledge none of the standard LTP tests shutdown or reboot the system. Could it be used for this? I'm sure. You might have the best luck simply pulling out appropriate filesystem thrashers and wrapping some stuff around them.
        Either way I know we'd like to hear your solution.
        j


Jay Huie
 wjhuie@us.ibm.com
 zSeries Linux System Test
 Phone: 845-435-8164

Sent by:        ltp-list-admin@lists.sourceforge.net

To:        "'ltp-list@lists.sourceforge.net'" <ltp-list@lists.sourceforge.net>
cc:        
Subject:        [LTP] Filesystem Resilience?





Hello.  I'm looking for a good way to test that a filesystem comes back clean after a dirty shutdown.  I'm in the process of testing suitability of network raid with MD and ENBD (http://www.it.uc3m.es/~ptb/nbd/).  The filesystem I'm using is JFS, and the raw devices come from LVM.  I'm filesystem agnostic.  JFS seems fine so I'll use it unless there's a good reason to switch.

 

Anyway, what I'd like to do is test how well my ENBD-exported device fares in the event of a dirty shutdown on the primary.  Supposedly, the data should be identical on both sides.  From what I've tested so far using simple scripts that write and sync data it has appeared to work pretty well -- the network-accessible mirror comes up very nicely.  However, what I would like to do is put the system under the type of duress that would most likely trigger such things as a corrupted journal or something else which would make it difficult, if not impossible, to recover the filesystem after a dirty shutdown.

 

Is fsstress a good way to do this?  Does anybody on the list use LTP for this type of testing?


Thank you very much for any assistance!

Mark Travis

mtravis@cybersource.com