cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sebastien Goasguen <run...@gmail.com>
Subject Re: VR refactoring, concerns and a way out ?
Date Thu, 24 Sep 2015 20:38:00 GMT
Folks, 

Let’s take a deep breath here, everyone is aiming for a good release.

With 4.6 we are trying a new way of creating the release, it may not be the best, but I think
we need to stick with the current process and release.
We can then have a post-mortem and see what worked and what did not work.

David and co have been working on hardware setup to finally get a CI running under ASF. This
will help.

For now, we need to focus on the release. Reverting a 6 months old feature is in my view a
no go, and I don’t have the feeling that it is the actual problem.

I suggest we all jump on a hangout tomorrow and try to figure this out.

* Wilder and co have done great work to test pending PR
* We have a set of blockers
* We have two great RMs.

The one thing that is very troubling to me are the statements that “this used to work 3
weeks ago”. I don’t understand how things can be broken now if only the RMs can merge
to master. I will have a look tomorrow.

So please, let’s take a deep breath, only RMs can touch master, and let’s work it out
tomorrow.

Cheers,

-sebastien

> On Sep 24, 2015, at 10:27 PM, Wilder Rodrigues <WRodrigues@schubergphilis.com>
wrote:
> 
> Thinking about being disrespectful when one doesn’t read the emails, or does but filters
parts of the message, and keeps storming about unclear things.
> 
> Yes, time to move on. We have to get a cloud running.
> 
> Cheers,
> Wilder
> 
>> On 24 Sep 2015, at 20:29, Raja Pullela <raja.pullela@citrix.com> wrote:
>> 
>> this is very disrespectful... Sorry to say that you don't understand the complexity
and impact of this..  Let's not discuss this over an email and agree to disagree with each
other... move on! 
>> 
>>> On Sep 24, 2015, at 10:20 PM, Wilder Rodrigues <WRodrigues@schubergphilis.com>
wrote:
>>> 
>>> Raja,
>>> 
>>> Do you actually know the amount of blockers we have and how many are VR related?
Because I have seen emails from Rajani around concerning the blockers and I don’t see many.
So, yes, I really do think your approach is non-sense.
>>> 
>>> I mentioned it before, about 1 week ago, but I think you just ignored the content
of the email. We have 7 blockers, from which 4 are VR related but probably only 2 are related
to the refactor of the router side (python code). You created 2 of the blockers. So, I think
would be better to focus on fixing them other than making a storm out of it.
>>> 
>>> You can see the list here: https://issues.apache.org/jira/secure/Dashboard.jspa?selectPageId=12326765
>>> 
>>> The java part of the router refactor was released on 4.5, quite some time ago.
So, please have a look at git log before mentioned the refactor as a whole.
>>> 
>>> Another thing is: master is unstable - not because VR changes - and nobody could
tests the PRs that should fix the VR issues.  When we suggested to stabilise Master, people
kept pushing features through PRs thinking that it would help - even after we said only BLOCKER
issues would be merged.
>>> 
>>> So, please stop this storm around the VR because we are trying to work.
>>> 
>>> Cheers,
>>> Wilder
>>> 
>>> 
>>> On 24 Sep 2015, at 17:21, Raja Pullela <raja.pullela@citrix.com<mailto:raja.pullela@citrix.com>>
wrote:
>>> 
>>> @wilder, Not sure why you would think it as a nonsense approach? sure, you realize
amount of code churn and blockers we are dealing with when 4.6 is ready to go out.
>>> 
>>> Agreed, the refactoring happened several months ago and we could have taken a
closer look then-   the recent blockers filed have uncovered areas where in the implementation
didn't exist.  I will post the bug details around this.
>>> 
>>> Obviously, the refactoring changes missed multiple critical test scenarios and
will take substantial time to test/stabilize.
>>> 
>>> The BVTs are coming good for basic zone and Adv zone there are still a number
of failures and it will take us good time to get those fixed.
>>> 
>>> Besides the BVTs, regression tests are in a very bad shape.  Hope to get to these
starting next week.
>>> 
>>> Please see my latest bvt report, I will post in another 2 hrs, waiting for a
new run to complete.
>>> 
>>> On Sep 24, 2015, at 7:00 PM, sebgoa <runseb@gmail.com<mailto:runseb@gmail.com>>
wrote:
>>> 
>>> 
>>> On Sep 24, 2015, at 3:17 PM, Remi Bergsma <RBergsma@schubergphilis.com<mailto:RBergsma@schubergphilis.com>>
wrote:
>>> 
>>> Are you serious? You consider to revert a PR that was merged over 6 months ago?
And expect it to become more stable?
>>> 
>>> I have not followed all the latest development, but if we are talking about the
VR refactoring, indeed it happened several months back. Reverting it now does not seem like
a good idea.
>>> 
>>> I am probably missed a beat here, but the latest BVT results sent by Raja showed
XS tests almost at 100%, there were only some issues with KVM.
>>> 
>>> The problem, in MHO, is not that we find bugs that we consider blockers. The
problem is we are unable to resolve them effectively because master is unstable. There currently
isn’t a single PR that solves it, hence there is no way to test PRs. This is because we
have many PRs open and they were all branched off of a master that doesn’t work. I simply
can't test proposed PRs.
>>> 
>>> This problem occurred about 3 weeks ago, because before that master worked and
we could solve issues and merge PRs. I’m not saying it was bug-free, but at least we could
work on stabilising it. Most likely, we accepted a “fix” that made things worse. Probably
even multiple of them.
>>> 
>>> Master seemed stable and PR where being merged towards 4.6 with success (it seemed),
so indeed if we have issues now of stability, we should identify what caused it
>>> 
>>> To get out of this, I think we need to combine a few PRs that make it stable.
I’ll have a look today with Wilder and Funs to see if what fixes we need to combine to make
it work again. O
>>> nce we merge it and master actually works again, we can rebase any open PR with
current master and work from there.
>>> 
>>> Potentially, if you identify the commit or commits that brought the instability
you could revert to that point and play forward PRs that did not render master unstable.
>>> 
>>> Thanks for looking into it.
>>> 
>>> -seb
>>> 
>>> Regards,
>>> Remi
>>> 
>>> 
>>> 
>>> 
>>> On 24/09/15 14:00, "Ramanath Katru" <ramanath.katru@citrix.com<mailto:ramanath.katru@citrix.com>>
wrote:
>>> 
>>> My vote is for the approach no.1 - to backout completely. Most of VR functionalities
are broken and are in a mess to say the least. It definitely will take some time and effort
from several folks to get it to a stable state.
>>> 
>>> Ram Katru
>>> 
>>> 
>>> -----Original Message-----
>>> From: Raja Pullela [mailto:raja.pullela@citrix.com]
>>> Sent: Thursday, September 24, 2015 2:06 PM
>>> To: dev@cloudstack.apache.org<mailto:dev@cloudstack.apache.org>
>>> Subject: VR refactoring, concerns and a way out ?
>>> 
>>> Hi,
>>> 
>>> 
>>> 
>>> I understand a concern on the VR changes was raised earlier.  My apologies to
restart this thread again.
>>> 
>>> 
>>> 
>>> However, my last conversation with Jayapal, who has fixed/have been fixing lot
of VR issues, about the VR issues and he is pretty concerned about the refactoring that has
happened.  I have had the same concern for sometime now  (VR issues have been on the list
of issues to be looked into for at least 4+ weeks) and wanted to see a good solution for this-
with VR being very fundamental to the system.
>>> 
>>> 
>>> 
>>> Couple of solutions/proposals –
>>> 
>>> 1)      Back out the VR changes – Pros: VR has been stable for some time and
it is working well.
>>> 
>>> 2)      Continue to fix/stability VR changes -   Concerns: is the unknowns, what
we will find out and how long this will take to stabilize the VR functionality.
>>> 
>>> 
>>> 
>>> Please chime in if you have any thoughts or concerns around this,
>>> 
>>> 
>>> 
>>> best,
>>> 
>>> Raja
>>> 
>>> 
> 


Mime
View raw message