[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH v12 2/3] quorum: implement bdrv_add_child() and

From: Max Reitz
Subject: Re: [Qemu-devel] [PATCH v12 2/3] quorum: implement bdrv_add_child() and bdrv_del_child()
Date: Wed, 30 Mar 2016 17:07:15 +0200
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.7.1

On 30.03.2016 13:39, Alberto Garcia wrote:
> On Tue 29 Mar 2016 05:51:22 PM CEST, Max Reitz wrote:
>>> It sounds like the argument here, and in Max's thread on
>>> query-block-node-tree, is that we DO have cases where order matters, and
>>> so we need a way for the hot-add operation to explicitly specify where
>>> in the list a child is inserted (whether it is being inserted as the new
>>> primary image, or explicitly as the last resort, or somewhere in the
>>> middle).  An optional parameter, that defaults to appending, may be ok,
>>> but we definitely need to consider how the order of children is affected
>>> by hot-add.
>> However, the order should be queriable after the fact, and there are
>> three ways I see to accomplish this:
>> (1) Make this information queriable as driver-specific BDS information.
>>     I personally don't like it very much, but it would be fine.
>> (2) Implement query-block-node-tree, make the order of child nodes
>>     significant and thus represent the FIFO order there. I don't like
>>     this because it would mean returning two orders through that child
>>     node list: One is the numeric order (children.0, children.1, ...)
>>     and another is the FIFO order, which are not necessarily equal.
>> (3) Fix FIFO order to the child name (its role). I'm very much in favor
>>     of this.
>> While I don't have good arguments against (1), I think I have good
>> arguments for (3) instead: It just doesn't make sense to have a numeric
>> order of children if this order doesn't mean anything; especially if you
>> suddenly do need the list of child nodes to be ordered. To me, it
>> doesn't make any sense to introduce a new hidden order which takes
>> precedence over this obvious user-visible order.
> I'm not sure if I understand correctly what you mean in (3). The
> user-visible FIFO order is the one specified when the Quorum is created:
> children.0.file.filename=hd0.qcow2,
> children.1.file.filename=hd1.qcow2,
> ...
> Would you then call those BDS children.0, children.1, etc

They are already called that way; it's not their node name but their
"child role" name.

>                                                           and make those
> names be the ones that actually define how they are ordered internally?

Yes, that's what I meant.

> I also have another (not directly related) question: why not simply use
> the node name when removing children? I understood that the idea was
> that it's possible to have the same node attached twice to the same
> Quorum, but can you actually do that? And what's the use case?

What I like about using the child role name is that it automatically
prevents you from specifying a node that is not a child of the given parent.

Which makes me notice that it might be a good idea to require the user
to specify the child's role when adding a new child. In this version of
this series (where only quorum is supported), the children are just
inserted in numerical order (first free slot is taken first), but maybe
the user wants to insert them in a different order.

For quorum, this is basically irrelevant if the order doesn't mean
anything (which I don't like), but it may be relevant for other block

And the "filling up quorum's children" topic then makes me notice that
(x-)blockdev-change should probably be transaction-able (so you can
restructure the whole BDS graph in a single transaction), but that's
something we can care about later on.


Attachment: signature.asc
Description: OpenPGP digital signature

reply via email to

[Prev in Thread] Current Thread [Next in Thread]