RFC: Fault injection

RFC: Fault injection

[Michael Ablassmeier]

hi,

im searching for a way to be able to easily simulate faulty tape drives
for CI/Testing purposes. The VTL emulation in TGT seems to have all the
functionality required for my purpose and so i went on and implemented a
proof of concept.

The POC uses the VTL Tapes MAM header to store information about which
errnous behavior should be simulated in case the tape is written.
Different tapes can as such simulate different fault behaviors.

See following description for more details:

 https://github.com/abbbi/tgt/blob/fault_injection/doc/README.ssc-fault-injection

Is there any interest in getting such a feature into TGT's VTL
emulation? Full diff here:

 https://github.com/fujita/tgt/compare/master...abbbi:fault_injection

As the MAM header is probably not the best way for storing the
information, would it be a good solution to store a special file
alongside to the VTL file, including more meta information related to
this feature?

bye,
    - michael

Re: RFC: Fault injection

[Mark Harvey]

The other option would be to use (hijack) the SCSI ‘SEND DIAGNOSTIC’ / ‘RECEIVE DIAGNOSTIC RESULTS’ op codes..

Code up a list of useful fault-injection conditions within DIAGNOSTIC OP code.. Pass any metadata as part of the payload - such as block number(s), after x reads/writes, only fault on a block size of x, only error on barcode xyz, fail first n attempts on reading block.

The RECEIVE DIAGNOSTIC could return any conditions already set..

Just a thought.

Cheers
Mark




> On 9 Jan 2021, at 01:12, Michael Ablassmeier <abi@grinser.de> wrote:
> 
> hi,
> 
> im searching for a way to be able to easily simulate faulty tape drives
> for CI/Testing purposes. The VTL emulation in TGT seems to have all the
> functionality required for my purpose and so i went on and implemented a
> proof of concept.
> 
> The POC uses the VTL Tapes MAM header to store information about which
> errnous behavior should be simulated in case the tape is written.
> Different tapes can as such simulate different fault behaviors.
> 
> See following description for more details:
> 
> https://github.com/abbbi/tgt/blob/fault_injection/doc/README.ssc-fault-injection
> 
> Is there any interest in getting such a feature into TGT's VTL
> emulation? Full diff here:
> 
> https://github.com/fujita/tgt/compare/master...abbbi:fault_injection
> 
> As the MAM header is probably not the best way for storing the
> information, would it be a good solution to store a special file
> alongside to the VTL file, including more meta information related to
> this feature?
> 
> bye,
>    - michael