On Fri, Nov 13, 2020 at 09:53:28 +0300, Nikolay Shirokovskiy wrote:
This is successor to [1] but I changed the subject as in the review the patch 'qemu: sync backing chain update in virDomainGetBlockJobInfo' was not considered good one from design POV. However I think the basic patch is helpful to address similar issues. Look at [*] for example, there it allows to sync backing chains during query block stats.
In discussion of [1] I stated that first patch will also allow to get rid of stale block job events on reconnection to qemu. But this will require additional work actually so with this patch series stale events are still present. And in the discussion it was also mentioned by Peter that stale events are not harmful for legacy blockjobs code. I also removed previous attempts to eliminate some of stale events.
I still keep patch for virDomainGetBlockJobInfo. May be in comparsion to [*] the approach the patch takes will look not so bad :)
[1] First version of the patch series https://www.redhat.com/archives/libvir-list/2020-October/msg01133.html
I had a look at this series and it's just doing too much just to "fix" apps which insist on polling especially with pre-blockdev qemu. I simply it's not worth adding the complexity you are proposing. NACK series. Instead I propose we add a workaround which will report fake unfinished job: https://www.redhat.com/archives/libvir-list/2020-December/msg00352.html The scope of this is way more limited and it doesn't actually try to modify the job code handovers which is very dangerous. I had a look at this series and it's just doing too much just to "fix" apps which insist on polling especially with pre-blockdev qemu. Additionally it adds documentation stating that apps should not poll virDomainGetBlockJobInfo.