Re: [PATCH RFC 00/10] libxl: new hotplug calling convention

["Roger Pau Monné" <roger.pau@citrix.com>]

On 17/01/13 14:56, Ian Campbell wrote:
> On Fri, 2012-12-21 at 16:59 +0000, Roger Pau Monne wrote:
>> This series implements a new hoplug calling convention for libxl.
>>
>> The aim of this new convention is to reduce the blackout phase of 
>> migration when using complex hotplug scripts, like iSCSI or other kind 
>> of storage backends that might have a non trivial setup time.
>>
>> There are some issues that I would like to discuss, the first one is 
>> the fact that pdev_path field in libxl_device_disk is no longuer used 
>> to store a physical path, since diskspec "target" can now contain 
>> "random" information to connect a block device.
>>
>> To solve this I would like to introduce a new field in 
>> libxl_device_disk called "target", that will be used to store the 
>> diskspec target parameter. This can later be copied to pdev_path if 
>> using the old hotplug calling convention.
> 
> Perhaps we could arrange either via a union or via LIBXL_API_VERSION
> ifdefs that "pdev_path" and "target" are in the same place?

We still need pdev_path, I would say only internally because once the
"target" has been attached to the Dom0, pdev_path has the path to the
block device that will be used in blkback, but the caller doesn't need
to know about this.

Old callers can continue to pass pdev_path as currently doing, but
callers wanting to use the new hotplug interface should instead use
target, and set hotplug_version to 1.

What I'm doing in this series is overwriting pdev_path with the real
device path once the hotplug script has been executed, but that's clumsy.

>> The second issue is related to iSCSI and iscsiadm, specifically the 
>> way to set authentication parameters, which is done with command line 
>> parameters or editing files (which each distro seems to place in 
>> different locations). I will look into this to see if we can find a 
>> suitable solution.
>>
>> Thanks for the comments, Roger.
>>
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.xen.org
>> http://lists.xen.org/xen-devel
> 
>