[libvirt] [PATCH v3] qemu: wait for SPICE to migrate

[Michal Privoznik]

On 25.09.2012 19:57, Daniel P. Berrange wrote:
> On Thu, Sep 20, 2012 at 01:29:21PM +0200, Michal Privoznik wrote:
>> Recently, there have been some improvements made to qemu so it
>> supports seamless migration or something very close to it.
>> However, it requires libvirt interaction. Once qemu is migrated,
>> the SPICE server needs to send its internal state to the destination.
>> Once it's done, it fires SPICE_MIGRATE_COMPLETED event and this
>> fact is advertised in 'query-spice' output as well.
>> We must not kill qemu until SPICE server finishes the transfer.
>> ---
>>  src/qemu/qemu_capabilities.c |    4 +++
>>  src/qemu/qemu_capabilities.h |    1 +
>>  src/qemu/qemu_command.c      |    8 ++++++
>>  src/qemu/qemu_migration.c    |   33 ++++++++++++++++++++++++--
>>  src/qemu/qemu_monitor.c      |   24 +++++++++++++++++++
>>  src/qemu/qemu_monitor.h      |    2 +
>>  src/qemu/qemu_monitor_json.c |   52 ++++++++++++++++++++++++++++++++++++++++++
>>  src/qemu/qemu_monitor_json.h |    3 ++
>>  8 files changed, 124 insertions(+), 3 deletions(-)
>>

>> @@ -940,6 +949,8 @@ qemuMigrationUpdateJobStatus(struct qemud_driver *driver,
>>  
>>      case QEMU_MONITOR_MIGRATION_STATUS_COMPLETED:
>>          priv->job.info.type = VIR_DOMAIN_JOB_COMPLETED;
>> +        if (spice_migrated)
>> +            priv->job.info.type = VIR_DOMAIN_JOB_COMPLETED;
> 
> This addition doesn't do anything, since the previous line
> has already set the same value.

Yeah, there should be '-' sign on the previous line...

> 
> 
> I don't find this very readable - it is better to do it as a
> regular if() IMHO. eg
> 
>   bool wait_for_spice = false;
> 
>   if (...)
>     wait_for_spice = true;
> 
> 
>>  
>>      switch (priv->job.asyncJob) {
>>      case QEMU_ASYNC_JOB_MIGRATION_OUT:
>> @@ -988,7 +1006,8 @@ qemuMigrationWaitForCompletion(struct qemud_driver *driver, virDomainObjPtr vm,
>>          /* Poll every 50ms for progress & to allow cancellation */
>>          struct timespec ts = { .tv_sec = 0, .tv_nsec = 50 * 1000 * 1000ull };
>>  
>> -        if (qemuMigrationUpdateJobStatus(driver, vm, job, asyncJob) < 0)
>> +        if (qemuMigrationUpdateJobStatus(driver, vm, job,
>> +                                         asyncJob, wait_for_spice) < 0)
>>              goto cleanup;
>>  
>>          if (dconn && virConnectIsAlive(dconn) <= 0) {
>> @@ -1840,6 +1859,7 @@ qemuMigrationRun(struct qemud_driver *driver,
>>      int fd = -1;
>>      unsigned long migrate_speed = resource ? resource : priv->migMaxBandwidth;
>>      virErrorPtr orig_err = NULL;
>> +    bool wait_for_spice;
>>  
>>      VIR_DEBUG("driver=%p, vm=%p, cookiein=%s, cookieinlen=%d, "
>>                "cookieout=%p, cookieoutlen=%p, flags=%lx, resource=%lu, "
>> @@ -1848,6 +1868,12 @@ qemuMigrationRun(struct qemud_driver *driver,
>>                cookieout, cookieoutlen, flags, resource,
>>                spec, spec->destType, spec->fwdType);
>>  
>> +    /* If guest uses SPICE and supports seamless migration we have to hold up
>> +     * migration finish until SPICE server transfers its data */
>> +    wait_for_spice = (vm->def->ngraphics == 1) &&
>> +        (vm->def->graphics[0]->type == VIR_DOMAIN_GRAPHICS_TYPE_SPICE) &&
>> +        qemuCapsGet(priv->caps, QEMU_CAPS_SEAMLESS_MIGRATION);
> 
> Same comment as before.
> 
> 
> Since both callers of qemuMigrateUpdateJobStatus() have to repeat
> this logic, why not push this down into that method itself and
> make 'wait_for_spice' be a local variable there instead of a param.

Well, I wanted to optimize here since qemuMigrateUpdateJobStatus is
called every 50 ms so I think accessing memory on each run would hurt
the performance (and this really do a lot of mem traffic since it access
vm->def->[n]graphics; qemuCaps). Or is this a premature optimization
which should be left to compiler and it's caching capabilities?

Michal