Re: [PATCH v5 1/9] migration: Add switchover ack capability

[Avihai Horon <avihaih@nvidia.com>]

On 15/06/2023 16:49, Cédric Le Goater wrote:
> External email: Use caution opening links or attachments
>
>
> On 6/15/23 14:38, YangHang Liu wrote:
>> Test in the following two scenarios:
>>
>> [1]  Test scenario:  Both source VM and target VM (in listening mode)
>> have enabled return-path and switchover-ack capability:
>>
>> Test result : The VFIO migration completed successfully
>>
>> [2]  Test scenario :  The source VM has enabled return-path and
>> switchover-ack capability while the target VM (in listening mode) not
>>
>> Test result : The VFIO migration fails
>>
>> The detailed error thrown by qemu-kvm when VFIO migration fails:
>>      Target VM:
>>            0000:17:00.2: Received INIT_DATA_SENT but switchover ack 
>> is not used
>>            error while loading state section id 
>> 81(0000:00:02.4:00.0/vfio)
>>            load of migration failed: Invalid argument
>>      Source VM:
>>             failed to save SaveStateEntry with id(name): 2(ram): -5
>>             Unable to write to socket: Connection reset by peer
>>             Unable to write to socket: Connection reset by peer
>>
>> Tested-by: YangHang Liu <yanghliu@redhat.com>
>
> Some more info,
>
> Tests were performed with a mainline Linux and a mainline QEMU including
> this series - patch8.
>
> The amount of precopy data for a CX-7 VF is not very large. Any idea how
> to generate some more initial state with such devices ?
>
> I suppose pre-copy will be more important with vGPUs.

In CX-7 the precopy data is not expected to be very large, because it's 
mainly used to pre-allocate resources in the destination.
However, precopy and switchover-ack are very important for CX-7, because 
they allow doing the resource pre-allocation in the destination when the 
source VM is running and reduce downtime significantly (see the example 
I gave in the cover letter).

Thanks.

>
>
>
> YangHang,
>
> Could you please reply with a Tested-by on the cover letter, so that the
> whole series is tagged and not only patch 1.
>
> Thanks,
>
> C.
>
>>
>>
>> On Wed, May 31, 2023 at 1:46 AM Avihai Horon <avihaih@nvidia.com> wrote:
>>>
>>> Migration downtime estimation is calculated based on bandwidth and
>>> remaining migration data. This assumes that loading of migration 
>>> data in
>>> the destination takes a negligible amount of time and that downtime
>>> depends only on network speed.
>>>
>>> While this may be true for RAM, it's not necessarily true for other
>>> migrated devices. For example, loading the data of a VFIO device in the
>>> destination might require from the device to allocate resources, 
>>> prepare
>>> internal data structures and so on. These operations can take a
>>> significant amount of time which can increase migration downtime.
>>>
>>> This patch adds a new capability "switchover ack" that prevents the
>>> source from stopping the VM and completing the migration until an ACK
>>> is received from the destination that it's OK to do so.
>>>
>>> This can be used by migrated devices in various ways to reduce 
>>> downtime.
>>> For example, a device can send initial precopy metadata to pre-allocate
>>> resources in the destination and use this capability to make sure that
>>> the pre-allocation is completed before the source VM is stopped, so it
>>> will have full effect.
>>>
>>> This new capability relies on the return path capability to communicate
>>> from the destination back to the source.
>>>
>>> The actual implementation of the capability will be added in the
>>> following patches.
>>>
>>> Signed-off-by: Avihai Horon <avihaih@nvidia.com>
>>> Reviewed-by: Peter Xu <peterx@redhat.com>
>>> Acked-by: Markus Armbruster <armbru@redhat.com>
>>> ---
>>>   qapi/migration.json | 12 +++++++++++-
>>>   migration/options.h |  1 +
>>>   migration/options.c | 21 +++++++++++++++++++++
>>>   3 files changed, 33 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/qapi/migration.json b/qapi/migration.json
>>> index 179af0c4d8..061ea512e0 100644
>>> --- a/qapi/migration.json
>>> +++ b/qapi/migration.json
>>> @@ -487,6 +487,16 @@
>>>   #     and should not affect the correctness of postcopy migration.
>>>   #     (since 7.1)
>>>   #
>>> +# @switchover-ack: If enabled, migration will not stop the source VM
>>> +#     and complete the migration until an ACK is received from the
>>> +#     destination that it's OK to do so.  Exactly when this ACK is
>>> +#     sent depends on the migrated devices that use this feature.
>>> +#     For example, a device can use it to make sure some of its data
>>> +#     is sent and loaded in the destination before doing switchover.
>>> +#     This can reduce downtime if devices that support this capability
>>> +#     are present.  'return-path' capability must be enabled to use
>>> +#     it.  (since 8.1)
>>> +#
>>>   # Features:
>>>   #
>>>   # @unstable: Members @x-colo and @x-ignore-shared are experimental.
>>> @@ -502,7 +512,7 @@
>>>              'dirty-bitmaps', 'postcopy-blocktime', 
>>> 'late-block-activate',
>>>              { 'name': 'x-ignore-shared', 'features': [ 'unstable' ] },
>>>              'validate-uuid', 'background-snapshot',
>>> -           'zero-copy-send', 'postcopy-preempt'] }
>>> +           'zero-copy-send', 'postcopy-preempt', 'switchover-ack'] }
>>>
>>>   ##
>>>   # @MigrationCapabilityStatus:
>>> diff --git a/migration/options.h b/migration/options.h
>>> index 45991af3c2..9aaf363322 100644
>>> --- a/migration/options.h
>>> +++ b/migration/options.h
>>> @@ -40,6 +40,7 @@ bool migrate_postcopy_ram(void);
>>>   bool migrate_rdma_pin_all(void);
>>>   bool migrate_release_ram(void);
>>>   bool migrate_return_path(void);
>>> +bool migrate_switchover_ack(void);
>>>   bool migrate_validate_uuid(void);
>>>   bool migrate_xbzrle(void);
>>>   bool migrate_zero_blocks(void);
>>> diff --git a/migration/options.c b/migration/options.c
>>> index b62ab30cd5..16007afca6 100644
>>> --- a/migration/options.c
>>> +++ b/migration/options.c
>>> @@ -185,6 +185,8 @@ Property migration_properties[] = {
>>>       DEFINE_PROP_MIG_CAP("x-zero-copy-send",
>>>               MIGRATION_CAPABILITY_ZERO_COPY_SEND),
>>>   #endif
>>> +    DEFINE_PROP_MIG_CAP("x-switchover-ack",
>>> +                        MIGRATION_CAPABILITY_SWITCHOVER_ACK),
>>>
>>>       DEFINE_PROP_END_OF_LIST(),
>>>   };
>>> @@ -308,6 +310,13 @@ bool migrate_return_path(void)
>>>       return s->capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
>>>   }
>>>
>>> +bool migrate_switchover_ack(void)
>>> +{
>>> +    MigrationState *s = migrate_get_current();
>>> +
>>> +    return s->capabilities[MIGRATION_CAPABILITY_SWITCHOVER_ACK];
>>> +}
>>> +
>>>   bool migrate_validate_uuid(void)
>>>   {
>>>       MigrationState *s = migrate_get_current();
>>> @@ -547,6 +556,18 @@ bool migrate_caps_check(bool *old_caps, bool 
>>> *new_caps, Error **errp)
>>>           }
>>>       }
>>>
>>> +    if (new_caps[MIGRATION_CAPABILITY_SWITCHOVER_ACK]) {
>>> +        if (!new_caps[MIGRATION_CAPABILITY_RETURN_PATH]) {
>>> +            error_setg(errp, "Capability 'switchover-ack' requires 
>>> capability "
>>> +                             "'return-path'");
>>> +            return false;
>>> +        }
>>> +
>>> +        /* Disable this capability until it's implemented */
>>> +        error_setg(errp, "'switchover-ack' is not implemented yet");
>>> +        return false;
>>> +    }
>>> +
>>>       return true;
>>>   }
>>>
>>> -- 
>>> 2.26.3
>>>
>>>
>>
>