[libvirt] [RFC PATCH] libvirt support to force convergence of live guest migration

[Jiri Denemark]

On Thu, Nov 21, 2013 at 17:47:24 -0800, Chegu Vinod wrote:
> Busy enterprise workloads hosted on large sized VM's tend to dirty
> memory faster than the transfer rate achieved via live guest migration.
> Despite some good recent improvements (& using dedicated 10Gig NICs
> between hosts) the live migration may NOT converge.
> 
> Recently support was added in qemu (version 1.6) to allow a user to
> choose if they wish to force convergence of their migration via a
> new migration capability : "auto-converge". This feature allows for qemu
> to auto-detect lack of convergence and trigger a throttle-down of the
> VCPUs. 
> 
> This RFC patch includes the libvirt support needed to trigger this
> feature. (Testing is still in progress)
> 
> Signed-off-by:  Chegu Vinod <chegu_vinod at hp.com>
> ---
>  include/libvirt/libvirt.h.in |    1 +
>  src/qemu/qemu_migration.c    |   44 ++++++++++++++++++++++++++++++++++++++++++
>  src/qemu/qemu_migration.h    |    1 +
>  src/qemu/qemu_monitor.c      |    2 +-
>  src/qemu/qemu_monitor.h      |    1 +
>  tools/virsh-domain.c         |    7 ++++++
>  6 files changed, 55 insertions(+), 1 deletions(-)
> 
> diff --git a/include/libvirt/libvirt.h.in b/include/libvirt/libvirt.h.in
> index 146a59b..13b0bfc 100644
> --- a/include/libvirt/libvirt.h.in
> +++ b/include/libvirt/libvirt.h.in
> @@ -1192,6 +1192,7 @@ typedef enum {
>      VIR_MIGRATE_OFFLINE           = (1 << 10), /* offline migrate */
>      VIR_MIGRATE_COMPRESSED        = (1 << 11), /* compress data during migration */
>      VIR_MIGRATE_ABORT_ON_ERROR    = (1 << 12), /* abort migration on I/O errors happened during migration */
> +    VIR_MIGRATE_AUTO_CONVERGE     = (1 << 13), /* force auto-convergence during during migration */
>  } virDomainMigrateFlags;

I feel like there must be a better name we could use for this flag but
I'm not able to come up with one... :-)

> diff --git a/src/qemu/qemu_migration.c b/src/qemu/qemu_migration.c
> index e87ea85..8cc0c56 100644
> --- a/src/qemu/qemu_migration.c
> +++ b/src/qemu/qemu_migration.c
> @@ -1565,6 +1565,40 @@ cleanup:
>  }
>  
>  static int
> +qemuMigrationSetAutoConverge(virQEMUDriverPtr driver,
> +                            virDomainObjPtr vm,
> +                            enum qemuDomainAsyncJob job)

Nicely copied&pasted from qemuMigrationSetCompression but you forgot to
fix indentation :-)

> +{
> +    qemuDomainObjPrivatePtr priv = vm->privateData;
> +    int ret;
> +
> +    if (qemuDomainObjEnterMonitorAsync(driver, vm, job) < 0)
> +        return -1;
> +
> +    ret = qemuMonitorGetMigrationCapability(
> +                priv->mon,
> +                QEMU_MONITOR_MIGRATION_CAPS_AUTO_CONVERGE);
> +
> +    if (ret < 0) {
> +        goto cleanup;
> +    } else if (ret == 0) {
> +            virReportError(VIR_ERR_ARGUMENT_UNSUPPORTED, "%s",
> +                           _("Auto-Converge migration is not supported by "
> +                             "QEMU binary"));

Reduce indentation by one level. Also I think "migration" is not really
needed in the error message, I'd just change it to
"Auto-converge is not supported by QEMU binary".

> +        ret = -1;
> +        goto cleanup;
> +    }
> +
> +    ret = qemuMonitorSetMigrationCapability(
> +                priv->mon,
> +                QEMU_MONITOR_MIGRATION_CAPS_AUTO_CONVERGE);
> +
> +cleanup:
> +    qemuDomainObjExitMonitor(driver, vm);
> +    return ret;
> +}
> +
> +static int
>  qemuMigrationWaitForSpice(virQEMUDriverPtr driver,
>                            virDomainObjPtr vm)
>  {
> @@ -2389,6 +2423,11 @@ qemuMigrationPrepareAny(virQEMUDriverPtr driver,
>                                      QEMU_ASYNC_JOB_MIGRATION_IN) < 0)
>          goto stop;
>  
> +    if (flags & VIR_MIGRATE_AUTO_CONVERGE &&
> +        qemuMigrationSetAutoConverge(driver, vm,
> +                                    QEMU_ASYNC_JOB_MIGRATION_IN) < 0)
> +        goto stop;
> +

Hmm, I know you are just following what I did with
VIR_MIGRATE_COMPRESSED, but setting auto-converge on destination doesn't
make any sense. And it doesn't even make a lot of sense to set
compression on destination (other than checking the destination supports
compression) so I'm wondering why I did so.

>      if (mig->lockState) {
>          VIR_DEBUG("Received lockstate %s", mig->lockState);
>          VIR_FREE(priv->lockState);
> @@ -3181,6 +3220,11 @@ qemuMigrationRun(virQEMUDriverPtr driver,
>                                      QEMU_ASYNC_JOB_MIGRATION_OUT) < 0)
>          goto cleanup;
>  
> +    if (flags & VIR_MIGRATE_AUTO_CONVERGE &&
> +        qemuMigrationSetAutoConverge(driver, vm,
> +                                    QEMU_ASYNC_JOB_MIGRATION_OUT) < 0)

Indentation is off by one space.

> +        goto cleanup;
> +
>      if (qemuDomainObjEnterMonitorAsync(driver, vm,
>                                         QEMU_ASYNC_JOB_MIGRATION_OUT) < 0)
>          goto cleanup;
...

So except for the small issues, the patch looks good to me. However, do
I remember correctly that this feature can be turned on dynamically for
an already running migration? If so, I think we want a second patch
adding a new API for setting this auto-converge feature.

Jirka