Re: [PATCH v14 4/4] iotests: 287: add qcow2 compression type test

[Eric Blake]

On 3/31/20 1:43 PM, Eric Blake wrote:

> Is it also worth trying to generate an image with (pseudo-)random 
> contents, and trying qemu-img convert to prove that uncompressable 
> clusters are handled sanely?  Ideally, there would be a way to use a 
> fixed PRNG and seed that produces a deterministic sequence that cannot 
> be compressed, but even if we can't meet the ideal, having a test that 
> non-deterministically is likely to generate an uncompressable cluster in 
> most runs is better than nothing (such as 1M of data copied from 
> /dev/urandom, then qemu-img convert on that data).

For an example,

nbdkit -U - random 1M seed=1234 --run 'qemu-img convert $nbd file'

produces 'file' containing deterministic pseudo-random data.  And 'zstd 
file' proves that the data was non-compressible in bulk:
$ zstd file
file                  :100.00%   (1048576 => 1048613 bytes, file.zst)

In fact, inspecting od output, I can see that the first few bytes of the 
original file are replayed verbatim after a 12-byte header in the .zst file.

But we don't have any other uses of nbdkit in our iotests, so I'm not 
sure if making this the first test to depend on that is ideal.

--
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org