On 06/14/2013 12:39 PM, Wenchao Xia wrote: > This interface can generate snapshot name automatically if it is not > specified, since it is a single opertion.
s/opertion/operation/
>
> Snapshot ID can't be specified in this interface.
>
> Signed-off-by: Wenchao Xia <xiaw...@linux.vnet.ibm.com>
> ---
> blockdev.c | 25 +++++++++++++++++++++++++
> qapi-schema.json | 23 +++++++++++++++++++++++
> qmp-commands.hx | 31 +++++++++++++++++++++++++++++++
> 3 files changed, 79 insertions(+), 0 deletions(-)
>
> +++ b/qapi-schema.json
> @@ -1689,6 +1689,29 @@
> '*mode': 'NewImageMode'} }
>
> ##
> +# @blockdev-snapshot-internal-sync
> +#
> +# Synchronously take an internal snapshot of a block device, when the format
> +# of the image used supports it.
> +#
> +# @device: the name of the device to generate the snapshot from
> +#
> +# @name: #optional the new snapshot name. If not specified, a name will be
> +# generated according to time by qemu
So why is name optional here but mandatory within a transaction? If
qemu is able to generate names, then it should be able to generate names
in both cases. Otherwise, make the name mandatory in both places.
Should this patch be folded in to 7/12? Compare with Stefan's series on
adding block-snapshot as a transaction (for that matter, the two series
have a [trivial] merge conflict since both add a transaction), and make
sure you are using the same approach between the two series at
introducing things.
> +#
> +# Returns: nothing on success
> +# If @device is not a valid block device, DeviceNotFound
> +# If any snapshot matching @name exists, or the name is a numeric
> +# which may mess up with snapshot ID, generic error will be returned
> +# If the format of the image used does not support it,
> +# BlockFormatFeatureNotSupported
> +#
> +# Since 1.6
> +##
> +{ 'command': 'blockdev-snapshot-internal-sync',
> + 'data': { 'device': 'str', '*name': 'str'} }
Ah, so this answers my question in 7/12 about a witness that libvirt can
use for knowing when transaction supports the new action, without
needing introspection.
> +
> +SQMP
> +blockdev-snapshot-internal-sync
> +-------------------------------
> +
> +Synchronously take an internal snapshot of a block device when the format of
> +image used supports it. If name is not specified, it will be automatically
> +generated by qemu according to host time. If the name is a numeric string
> +which may mess up with ID, such as "19", the operation will fail.
Wait a second. If we DON'T pass a name, then the generated name is all
numeric. But if we DO pass an all-numeric name, it gets rejected.
That's awkward to explain. Maybe you want to instead have a cutoff,
where a number < 64k (is that the right threshold? I just picked a
number out of the air) is rejected, but a number >= 64k is treated as
valid because it might represent a timestamp. Or allow all possible
numbers, and only reject the creation of a name that collides with an
existing id. It may be too hard to predict if a name will collide with
a future id of a later snapshot operation.
> +Example:
> +
> +-> { "execute": "blockdev-snapshot-internal-sync",
> + "arguments": { "device": "ide-hd0",
> + "name": "snapshot0" }
> + }
> +<- { "return": {} }
Evil. If I don't pass a name, then I NEED to know what name got
generated on my behalf. So that argues you need to return something,
rather than nothing. I can see why you can't return a string via
'transaction', but maybe this is an argument that 'name' should be
mandatory in this QMP command (and any generation of a timestamp id must
be higher up in the stack, at the HMP level, so that HMP can still treat
name as optional). But then you are back to solving the problem of
allowing an all-numeric generated timestamp as a valid name.
--
Eric Blake eblake redhat com +1-919-301-3266
Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
