[ovirt-users] stalls during live Merge Centos 7 / qemu 2.3

[Pavel Gashev]

Markus,

So all CPU threads are blocked by the main loop. The main loop is busy draining IO requests from all drives while there are other block commit jobs active... It looks like currently QEMU just can't live merge more than one drive simultaneously without blocking VCPUs.

As workaround you can delete disk snapshots one-by-one, and then delete disk-less VM snapshot. The Storage tab has Disk Snapshots sub tab where you can remove a specific snapshot of a specific disk.





On 12/04/16 21:39, "Markus Stockhausen" <stockhausen at collogia.de> wrote:

>> Von: Pavel Gashev [Pax at acronis.com]
>> Gesendet: Dienstag, 12. April 2016 16:15
>> An: Markus Stockhausen; users
>> Betreff: Re: [ovirt-users] stalls during live Merge Centos 7 / qemu 2.3
>> 
>> Markus,
>> 
>> I saw similar issues. Looks like it's related to multidisk VMs under high disk load. No relations with guest OS. No relations with IO threads.
>> 
>> 
>> I didn't debug it. Can you please show a backtrace of a thread in __lll_lock_wait ?
>
>Hi Pavel. 
>
>thanks for your response. I dont even need high disk load to 
>reproduce the issue. Just a simple script that writes 8k blocks
>every 0.5 secs.  I added the required info to the BZ.
>
>https://bugzilla.redhat.com/show_bug.cgi?id=1319400
>
>Markus
>
>> 
>> 
>> On 11/04/16 18:13, "users-bounces at ovirt.org on behalf of Markus Stockhausen" <users-bounces at ovirt.org on behalf of stockhausen at collogia.de> wrote:
>> 
>> >Hi there,
>> >
>> >I'm getting slowly mad about our new Centos 7 cluster. Whenever
>> >I start a live merge the machine completely freezes. It seems to be
>> >independent of the Guest OS (tried SLES 11 SP3, SLES11 SP4 and
>> >SLES12). I already opened BZ1319400 because I'm clueless.
>> >
>> >Doing the same in our Fedora 20 cluster (qemu 2.1.3) we only see
>> >small hiccups of 1-3 seconds at the start of the live merge but no
>> >complete stall. Its the same VM!
>> >
>> >Just to be sure that it is not a specific problem of our topology
>> >or SLES I would like to receive some feedback of the list if anyone
>> >observes similar lockups in their environment.
>> >
>> >Thanks in advance.
>> >
>> >Markus
>> >