Re: [Qemu-devel] [Qemu-block] RFC: Operation Blockers in QEMU Block Nodes

[Alberto Garcia <berto@igalia.com>]

Hi,

I'm late to the party but I wanted to say that I generally like the idea
of expressing the notion of op blockers in terms of the specific things
that they allow and require in each one of the affected nodes.

I don't know if all current operations can be expressed easily with
these semantics, though.

I also noticed this:

On Sat 19 Dec 2015 06:42:00 AM CET, Jeff Cody <jcody@redhat.com> wrote:

 [block-streaming from B to C]
>      A <- B <- C <--- virtio-blk
>
>
>     1. Data is read from (B) and written into (C)
>
>          - This is not a "Write Visible Data" operation, because a read
>            from (C) will return the same results as it did prior to the
>            operation.
>
>     2. The backing file in (C) is modified to reflect (A) as the new
>        backing file
>
>          - This is what would require the "Write Metadata".  Not
>            because it doesn't require "Write Visible Data", but
>            because we change the backing file of (C).
>
>     3. Node (B) is dropped from the internal BDS graph, and (C)
>        becomes the direct overlay of (A).
>
>           - This would require the "Graph Reconfiguration"
>
>
> Here is how I would see the full set of Require / Allow flags for each
> node involved in your hypothetical block job:
>
>                      Nodes involved: (A), (B), (C).
>
>                 ----------------------------------------------
>                 |   Node A     |   Node B     |   Node C     |
>                 | Req  : Allow | Req  : Allow | Req  : Allow |
> =============================================================|
> Modify Visible  |   0  :  1    |   0  :  0    |   0  :  1    |
> Modify Metadata |   0  :  1    |   0  :  1    |   1  :  0    |
> Graph Reconfig  |   1  :  0    |   0  :  1    |   1  :  0    |
> Read Visible    |   0  :  1    |   1  :  1    |   0  :  1    |
> Read Metadata   |   1  :  1    |   1  :  1    |   1  :  1    |

Why does this require "graph reconfig" in (A) but not in (B)? The result
of the operation is that (B) is dropped and (C) now points to (A), but
(A) itself doesn't change.

Berto