[libvirt] [PATCH v2] qemu: Honour <on_reboot/>

[Michal Privoznik]

On 08/29/2017 11:08 AM, Martin Kletzander wrote:
> On Wed, Aug 16, 2017 at 04:38:09PM +0200, Michal Privoznik wrote:
>> https://bugzilla.redhat.com/show_bug.cgi?id=1476866
>>
>> For some reason, we completely ignore <on_reboot/> setting for
>> domains. The implementation is simply not there. It never was.
>>
>> Signed-off-by: Michal Privoznik <mprivozn at redhat.com>
>> ---
>>
>> diff to v1:
>> - dropped the spoofed logic
>> - Switch from qemuProcessShutdownOrReboot() to qemuPrcoessStop()
>> because that's
>>  what <on_crash/> impl does too.
>>
>> src/qemu/qemu_process.c | 27 ++++++++++++++++++++++++---
>> 1 file changed, 24 insertions(+), 3 deletions(-)
>>
>> diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c
>> index fed2bc588..3df6c320e 100644
>> --- a/src/qemu/qemu_process.c
>> +++ b/src/qemu/qemu_process.c
>> @@ -484,6 +484,7 @@ qemuProcessHandleReset(qemuMonitorPtr mon
>> ATTRIBUTE_UNUSED,
>>     virObjectEventPtr event;
>>     qemuDomainObjPrivatePtr priv;
>>     virQEMUDriverConfigPtr cfg = virQEMUDriverGetConfig(driver);
>> +    int ret = -1;
>>
>>     virObjectLock(vm);
>>
>> @@ -495,12 +496,32 @@ qemuProcessHandleReset(qemuMonitorPtr mon
>> ATTRIBUTE_UNUSED,
>>     if (virDomainSaveStatus(driver->xmlopt, cfg->stateDir, vm,
>> driver->caps) < 0)
>>         VIR_WARN("Failed to save status on vm %s", vm->def->name);
>>
>> +    if (vm->def->onReboot == VIR_DOMAIN_LIFECYCLE_DESTROY ||
>> +        vm->def->onReboot == VIR_DOMAIN_LIFECYCLE_PRESERVE) {
>> +
>> +        if (qemuDomainObjBeginJob(driver, vm, QEMU_JOB_MODIFY) < 0)
>> +            goto cleanup;
>> +
>> +        if (!virDomainObjIsActive(vm)) {
>> +            VIR_DEBUG("Ignoring RESET event from inactive domain %s",
>> +                      vm->def->name);
>> +            goto endjob;
>> +        }
>> +
>> +        qemuProcessStop(driver, vm, VIR_DOMAIN_SHUTOFF_DESTROYED,
>> +                        QEMU_ASYNC_JOB_NONE, 0);
>> +        virDomainAuditStop(vm, "destroyed");
> 
> Queuing another event here that the domain is being destroyed seems both
> appropriate and weird to me.  So I'll leave it up to you.  It's not like
> anyone ever used this functionality... ever.  ACK either way.

I think we want to queue the event here. Imagine that there's a mgmt app
that acts like HA daemon. Whenever a domain is destroyed it has to start
it up again. Well, with the current code it has to listen to RESET event
and depending on libvirt version it has to either ignore the event or
start it up again. However, if we queue the event all that the app has
to do is to listen to DESTROY event. Therefore I'm inclined to leave it
here.

Pushed, thanks.

Michal