On 09/15/2014 06:13 PM, John Ferlan wrote:
Well I guess it happens to everyone - hopefully only once though...
Yep, the brown bag of shame lets you learn from mistakes rather quickly :)
The IOThreads code really messed up a few things - git bisection for the build is broken as of 5f6ad32c733a3bd158938aecabb0508a434ece95, but that's resolved by 938fb12fad6d15c9fdb73f998c4e0ec1e278721f which adds the 'niothreadspin && iothreadspin' definitions.
Secondarily it seems things got worse, because guests weren't able to be started because of a very bad logic error (<= 0 vs. < 0 and specific check that 0 is "OK" (meaning no IOThreads).
How that got by own self testing I'm still not quite sure, but it did and it's a mea culpa for that.
Anyway, patch 1 fixes the running issue... Patch 2 fixes some spacing issues that Eric pointed out privately (prefer "i + 1" vs. "i+1"). Patch 3 just makes sure to use the right cgroup setup/init parameters since there is no "iothread0"
John Ferlan (3): qemu: Fix iothreads issue qemu_cgroup: Adjust spacing around incrementor qemu: Fix call in qemuDomainSetNumaParamsLive for virCgroupNewIOThread
Thanks for the fast fixes. I confirm that patch 1 fixes the regression in starting my dummy transient domain. ACK series. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org