Re: [Xen-devel] Xen Project Community Call June 27th (instead of July 4th): @15:00 UTC Call for agenda items

[Lars Kurth <lars.kurth@xxxxxxxxxx>]

On 26/06/2019, 18:44, "Stefano Stabellini" <sstabellini@xxxxxxxxxx> wrote:

    On Wed, 26 Jun 2019, Rich Persaud wrote:
    > > On Jun 26, 2019, at 06:45, Lars Kurth <lars.kurth@xxxxxxxxxx> wrote:
    > > 
    > > 
    > > 
    > > On 25/06/2019, 21:27, "Rich Persaud" <persaur@xxxxxxxxx> wrote:
    > > 
    > >> On Jun 25, 2019, at 16:17, Julien Grall <julien.grall@xxxxxxx> wrote:
    > >> 
    > >> Hi Rich,
    > >> 
    > >> On 6/25/19 8:38 PM, Rich Persaud wrote:
    > >>>> On Jun 25, 2019, at 12:36, Lars Kurth <lars.kurth@xxxxxxxxxx> wrote:
    > >>>> 
    > >>>> Hi all:
    > >>>> please add your agenda items. I had only ONE series which was 
highlighted as needing attention from others. Is this seriously the only one?
    > > 
    > > We had quite a few additions to the agenda. One problem I have is that 
cryptpad history does not tell me who added an agenda item. We will have to 
manage this appropriately in the meeting.
    > > 
    > >>> Proposed agenda item: in the absence of Jira tickets, would it be 
useful to have a list (e.g. generated by a script) with the lifecycle status of 
all outstanding patch series, e.g.
    > >>> METADATA
    > >>> - bug fix / improvement / refactor / cleanup / new feature
    > >>> - impacted Xen subsystems/components/features
    > >>> - targeted version of Xen
    > >>> - contributing person/org
    > >>> - relevance of patch series to the goals set by RM for the next Xen 
release
    > >>> - related patch series (with below status info)
    > >>> STATUS:
    > >>> - patch series version
    > >>> - date of patch series v1
    > >>> - no responses received + ping count + days since submission + days 
since ping
    > >>> - reviewed with objections
    > >>> - reviewed without objections, awaiting ack
    > >>> - acked, awaiting merge
    > >>> From such a summary, patch series could be prioritized for 
review/triage in the community call, based on uniform criteria and project-wide 
context.
    > >> 
    > >> While I think raising awareness of the stuck series is a good idea. I 
still have some concern regarding the prioritization. Who is going to consume 
the result of the discussion? Is it the maintainers?
    > > 
    > >   Anyone who typically answers the question raised by Lars in this 
thread, presumably a subset of call attendees.
    > > 
    > > This would only work if there was consensus on the priority amongst the 
key stake-holders. I believe that some limited prioritization has happened in 
the past, e.g. for some Arm related features for Xen 4.12 where, if I recall 
correctly you, Stefano and EPAM did this. 
    > > 
    > > Maybe we can trial this type of approach for a small number of series 
first. At the end of the day this is about queue management. Right now, when a 
new series comes in it ends up in one big queue (xen-devel@). Most complex 
series have to go through a series of gates (aka code reviews from different 
people) before they get applied and when a new version comes out the series 
ends up in the queue again: each reviewer today prioritizes their own review 
queues, where no-one else sees the prioritisation of other reviewers. Unless 
there is lot of spare review capacity (which there isn't) we essentially end up 
in grid-lock, with an ever-growing queue. We seem to have specific additional 
complexity in Xen because most recent series, typically involve a large number 
of reviewers.
    > > 
    > > In theory, there are several ways to address this:
    > > * Queue management either by a set of agreed criteria which all 
reviewers follow or through some agreement about which series we actively try 
and shepherd through the gates
    > > * We have an additional issue that in many areas we have multiple 
reviewers/committers reviewing the same area of code, which also frequently 
leads to slow-downs, because the multiple reviewers/committers can disagree. We 
could look at a model where the reviewers/committers agree that one takes the 
lead on reviewing a specific series. We could try and streamline the ownership 
structure to create a clearer mapping.
    > > * The queues of each reviewer are somehow made public (assuming this is 
possible) and we hope that the system self-regulates. Not sure this will 
actually work
    > > 
    > > The big problem I have is that mailing lists really don't lend 
themselves well to highlight what is going on. We have been grappling with this 
for years and things are getting worse, not better.
    > >
    > > In past community calls when we tried to do this with specific series, 
in practice we ended up discovering obstacles that were concerning a specific 
series, such as unexposed dependencies, lack of HW, specs against which to 
review being too complex, ...
    > > 
    > > In any case, given that quite a few series were added to the agenda, 
maybe we shouldn't talk about process in the meeting, but see whether we can 
unblock those series. I am going to annotate some of the agenda items to 
highlight WHO needs to take action on items added
    > > 
    > > We could have a wider discussion about the process at the summit, as 
everyone who would need to be involved is at the summit.
    > 
    > This has likely been raised before, but ... could the Xen community use 
Github/Gitlab PRs to reduce the overhead of managing a review queue?  PR-based 
workflows have helped open-source projects to better utilize teams for review 
queue management.
    > 
    > PRs could be used in parallel with the existing mailing list patch 
process.  To link the two workflows, PR review comments could be mirrored to 
the mailing list, and a link to the PR included in the first patch of the 
series revisions.
    > 
    > If PRs are used, Jira can automatically link PRs that are associated with 
a XEN-nnnn ticket number.  With  PR comment mirroring, the mailing list would 
remain the definitive archive of review comments.  There may also be 
opportunities for integration with Xen's Gitlab CI, e.g. series testing on 
multiple architectures.
    
    Yes, this has been brought up in the past, and the majority (me
    included) preferred doing doing reviews on a mailing list. However,
    patchworks has been getting better and better and should now be able to
    give you a github-like web interface to patch series submissions and
    reviews while retaining the mailing list based workflow most still
    prefer. I know patchworks has also been used to trigger testing by some.
    I don't know the state of the patchworks instance for Xen.

As far as I recall Wei has looked at both patchwork 
(https://patchwork.kernel.org/project/xen-devel/list/) and :patchew 
(https://patchew.org/Xen) for the purpose of looking at triggers. I recall he 
sent a summary mail, but I can't find it. As far as I recall :patchew was 
deemed nicer, as it retains the cover letter of patches in most cases.

We will need to decide for one or the other at the summit, as we need this to 
trigger GitLabCi build and smoke tests (QEMU based) from patches posted to the 
list. 

Regards
Lars

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel