04.11.2020 20:30, Max Reitz wrote:
On 22.10.20 23:10, Vladimir Sementsov-Ogievskiy wrote:
23.07.2020 11:03, Max Reitz wrote:
On 01.06.20 20:11, Vladimir Sementsov-Ogievskiy wrote:
After introducing parallel async copy requests instead of plain
cluster-by-cluster copying loop, backup job may finish earlier than
final assertion in do_test_stop. Let's require slow backup explicitly
by specifying speed parameter.
Isn’t the problem really that block_set_io_throttle does absolutely
nothing? (Which is a long-standing problem with 129. I personally just
never run it, honestly.)
Hmm.. is it better to drop test_drive_backup() from here ?
I think the best would be to revisit this:
https://lists.nongnu.org/archive/html/qemu-block/2019-06/msg00499.html
Max
I've checked, no, that doesn't help. I suppose that throttling has same defect
as
original block-job speed: it calculates throttling after request, and if we
issue
simultaneously many requests they all will be handled (and after it we'll have
long
throttling pause). New solution for backup in these series works better with
parallel
requests (see patch [PATCH v2 07/20] block/block-copy: add ratelimit to
block-copy).
So, I'd keep this patch for now.
--
Best regards,
Vladimir
