Re: [RFC 0/4] POC: Generating realistic block errors

[Tony Asleson]

On 9/20/19 12:08 PM, Eric Blake wrote:
> I am worried, however, that making data transactions have to go through
> QMP to get an answer on how to handle a specific guest request will slow
> things down; QMP is not built to be an efficient dataplane interface.

IMHO we only need to make the code path efficient up to the point we
know we have an IO operation that contains sector(s) that we want to do
something with.  Once that happens I don't think it will be an issue for
those to be slower.  A real spinning disk drive takes longer to handle a
request when it encounters an error too.

> If you truly want isolation (where another process receives all guest
> transactions, and makes decisions on how to handle them)

It would only handle specific sector(s) of interest, not all.

-Tony