Re: crashblk: a block device crash emulator.

[Takashi HOSHINO]

On 2016/01/16 5:09, Josef Bacik wrote:
> On 01/15/2016 01:45 AM, Takashi HOSHINO wrote:
> > Hi Andreas,
> >
> > On 2016/01/15 4:44, Andreas Dilger wrote:
> > > On Jan 13, 2016, at 6:50 PM, Takashi HOSHINO
> > > <hoshino@xxxxxxxxxxxxxxxxx> wrote:
> > > > Hi all,
> > > >
> > > > I'm developing 'crashblk'.
> > > > It is a memory block device driver to emulate crash, IO error,
> > > > and various response time.
> > >
> > > Why not use/improve dm-flakey for this, rather than making yet another
> > > device that does something similar?
> >
> > I did not know dm-flakey until your reply. Thank you.
> >
> > I read its documents and code a bit.
> > The big purpose of two projects are the same, which is to test software
> > on block devices with errors, but their functions are almost different.
> >
> > It may be nice idea to implement alternative implementation of crashblk
> > on device-mapper version by someone.
> > Currently I'm not familiar with device-mapper development, and I do not
> > have any motivation with it.
>
> There's also dm-log-writes which will log writes to an external journal
> to allow you to replay writes and verify the fs is consistent at all
> steps of operation.  Thanks,

My colleague let me know about dm-log-writes several months ago and I 
read its code. Its very small and nice! (<1000 lines)
I felt it is useful to DEBUG file systems or wrapper devices but not 
useful to TEST them because it will tell us too much information.

BTW, our another software WalB, which is our primary test target of 
crashblk, also log-writes to an journal device.
Its primary usage is to backup devices incrementally or to replicate 
them, while it may also useful for debug purpose.
But WalB uses journals not only to extract logs for replay but also to 
keep consistency and control durability of data like some filesystems, 
so it requires more code.

Thanks,

Takashi