Re: Xen 4.3 released!

[George Dunlap <george.dunlap@eu.citrix.com>]

On 11/07/13 10:41, Ben Guthro wrote:
> On Tue, Jul 9, 2013 at 9:01 AM, George Dunlap
> <George.Dunlap@eu.citrix.com> wrote:
>> The Xen Project team is pleased to announce the release of Xen 4.3.
>>
>> The result of nearly 10 of development, new features include:
>>   * Early support for ARM 32- and 64-bit architectures
>>   * qemu-upstream is now the default for VMs not using stub domains.
>>   * openvswitch hot-plug script support.
>>   * NUMA affinity for the scheduler
>>   * xl can now accept several USB devices, rather than only one.
>>   * XSM improvements.  XSM can now override all IS_PRIV checks in the hypervisor.
>>   * As always, a number of stability, performance, and security
>> enhancements "under the hood".
>>
>> Detailed release notes, including a more extensive feature list:
>>    http://wiki.xenproject.org/wiki/Xen_4.3_Release_Notes
>>
>> To download tarballs:
>>    http://www.xenproject.org/downloads/xen-archives/supported-xen-43-series/xen-430.html
>> Or the git source repository (tag 'RELEASE-4.3.0'):
>>    http://xenbits.xen.org/gitweb/?p=xen.git
>>
> Hi George,
>
> Thanks for all the effort in coordinating this release.
>
> Generally, I prefer to wait until things clear the necessary gates to
> get out of the staging-4.3 (or whatever staging tree) before rebasing
> our XenClient patches on top of it, as it usually means it passed some
> level of automated testing through osstest.
>
> In this case, the staging-4.3 tree is where the RELEASE-4.3.0 tag
> exists, but the stable-4.3 tag seems to be a handful of changesets
> behind, despite being released a couple days ago.
>
> Could you, or others comment on this discrepancy?
> Is the fact that the stable-4.3 tree is lagging behind the staging-4.3
> branch an oversight, or has it not passed some push gate?

The last few changes have not passed the push gate.

Unfortunately, osstest died over the weekend, at the same time that Ian 
Jackson, who is in charge of it, fell dreadfully ill.  By the time it 
was noticed, the announcement had already been given to the press.  We 
had the choice of going through with the release with the last few 
changes not quite tested, or having an apology on the download page; and 
we had only a few hours or so to actually make a decision.

Looking at the changesets, we decided that on the whole the best thing 
in the circumstances would be to do our own testing and then go through 
with the release.

The tester ran on 4.3 for the first time this morning, and there was 
some issue with the QEMU tag.  That will probably be sorted shortly.

This has obviously highlighted a few deficiencies in the process that 
we'll have to fix in future releases:

* We should remove debug=y in the middle of RCs, to make sure we get 
good testing without that change made
* If we're going to do a big announcement, Linux Foundation-style, we 
need to have things branched, tagged and tested a week before the 
announcement.

Any other feedback is also welcome.

  -George