Re: [PATCH] iotests/MRCE: Write data to source

[Hanna Reitz <hreitz@redhat.com>]

On 23.12.21 17:53, Hanna Reitz wrote:
> This test assumes that mirror flushes the source when entering the READY
> state, and that the format level will pass that flush on to the protocol
> level (where we intercept it with blkdebug).
>
> However, apparently that does not happen when using a VMDK image with
> zeroed_grain=on, which actually is the default set by testenv.py.  Right
> now, Python tests ignore IMGOPTS, though, so this has no effect; but
> Vladimir has a series that will change this, so we need to fix this test
> before that series lands.
>
> We can fix it by writing data to the source before we start the mirror
> job; apparently that makes the (VMDK) format layer change its mind and
> pass on the pre-READY flush to the protocol level, so the test passes
> again.  (I presume, without any data written, mirror just does a 64M
> zero write on the target, which VMDK with zeroed_grain=on basically just
> ignores.)
>
> Without this, we do not get a flush, and so blkdebug only sees a single
> flush at the end of the job instead of two, and therefore does not
> inject an error, which makes the block job complete instead of raising
> an error.
>
> Signed-off-by: Hanna Reitz <hreitz@redhat.com>
> ---
> As hinted at above, I think this should be merged before Vladimir's
> "iotests: support zstd" series, or said test fails for me with VMDK.
> (At least on one system, not the other...  Would be too easy otherwise,
> obviously.)
> ---
>   tests/qemu-iotests/tests/mirror-ready-cancel-error | 7 ++++++-
>   1 file changed, 6 insertions(+), 1 deletion(-)

Applied to my block branch.

Hanna