Re: [VOTE] Release 2.2.0, release candidate #3

2017-11-12 Thread Jean-Baptiste Onofré 

Hi Reuven,

+1 for RC4, and don't worry: it's part of the process. I prefer to have a long 
release process than a crappy a release ;) That's exactly the purpose of review 
& vote.


I definitely think that having releases more often will reduce such kind of 
issue.

Regards
JB

On 11/12/2017 09:04 AM, Reuven Lax wrote:

I definitely appreciate the frustration about how long this release is
taking. It's verging on the point of ridiculous at this point, and we need
to fix some of the things that caused us to get to this state (for one
thing our infrastructure was so busted at one point, that Valentyn spent 2
weeks trying to get on PR merged into the release branch).

At this point, let's try and fix this Monday. Unfortunately this is not the
sole issue requiring RC4. Python verification failed as well, and we need
an RC4 regardless to merge those PRs. I'm hoping that RC4 is our final RC,
and we can finish voting next week.

Reuven

On Sat, Nov 11, 2017 at 6:24 AM, Romain Manni-Bucau 
wrote:


Le 11 nov. 2017 09:52, "Jean-Baptiste Onofré"  a écrit :

If the purpose is to release 2.2.1 in one week, why not just to a RC4 ?

It's not a regression because WriteFiles is new and extend the previous
FileSource. So it could consider as a severe bug, especially on WriteFiles
which is important.


Fair enough.


The core issue is the time we spent already on this release: roughly 1
month !!! It's clearly too long due to different causes.
When I did the previous releases, it took 3 or 4 days. It's clearly the
target as, as said, I would like to have a release pace of a release every
6 weeks.



Agree and this is why 2.2.0 must be out now IMHO. If you are confident next
week is sufficient just go ahead and ignore my comment but my point was the
same: it shouldnt last so long if there is no regression :(.



Regards
JB


On 11/11/2017 08:41 AM, Romain Manni-Bucau wrote:


You can see it differently: is there a critical bug? Yes! Is there a
regression? No!

So no need to wait another week (keep in mind 2 days + 3 days of vote
makes easily 1 working week). This vote could be closed already and next
week 2.2.1 could fix this bug, no? Overall idea is to not hold the
community more than needed if there is no regression compared to last few
releases.

Le 11 nov. 2017 07:46, "Jean-Baptiste Onofré"  a écrit

:


-1 (binding)


I agree with Eugene, data loss is severe.

As Eugene seems confident to fix that quickly, I think it's worth to

cut a

RC4.

However, I would introduce a deadline. As I would like to propose a
release cycle of a release every 6 weeks (whatever it contains, but it
really important to keep  a regular pace in releases), a release should

be

cut in couple of days. So, maybe we can give us 2 business days to fix
that
and propose a RC4. Basically, if this issue is not fix on Tuesday night,
then, we move forward anyway.

Regards
JB

On 11/10/2017 07:42 PM, Eugene Kirpichov wrote:

Unfortunately I think I found a data loss bug - it was there since 2.0.0

but I think it's serious enough that delaying a fix until the next
release
would be irresponsible.
See https://issues.apache.org/jira/browse/BEAM-3169

On Thu, Nov 9, 2017 at 3:57 PM Robert Bradshaw

wrote:

Our release notes look like nothing more than a query for the closed


jira issues. Do we have a top-level summary to highlight the big
ticket items in the release? And in particular somewhere to mention
that this is likely the last release to support Java 7 that'll get
widely read?

On Thu, Nov 9, 2017 at 3:39 PM, Reuven Lax 
wrote:

Thanks,


This RC is currently failing on a number of validation steps, so we
need

to


cut at least one more RC. Fingers crossed that it will be the last

one.


Reuven

On Thu, Nov 9, 2017 at 3:36 PM, Konstantinos Katsiapis <
katsia...@google.com.invalid> wrote:

Just a remark: Release of Tensorflow Transform


 0.4.0 depends on release

of

Apache Beam 2.2.0 so upvoting for a release (the sooner the better).

On Thu, Nov 9, 2017 at 3:33 PM, Reuven Lax  wrote:

https://github.com/apache/beam/pull/4109 is out to address both



findings I


reported earlier.


On Thu, Nov 9, 2017 at 8:54 AM, Etienne Chauchot <

echauc...@gmail.com>





wrote:





Just as a remark, I compared (on my laptop though) queries



execution





times





on my previous run of 2.2.0-RC3 with release 2.1.0 and I did not


see





any





performance regression.




Best

Etienne


Le 09/11/2017 à 03:13, Valentyn Tymofieiev a écrit :

I looked at Python

Re: [VOTE] Release 2.2.0, release candidate #3

2017-11-12 Thread Reuven Lax 
I definitely appreciate the frustration about how long this release is
taking. It's verging on the point of ridiculous at this point, and we need
to fix some of the things that caused us to get to this state (for one
thing our infrastructure was so busted at one point, that Valentyn spent 2
weeks trying to get on PR merged into the release branch).

At this point, let's try and fix this Monday. Unfortunately this is not the
sole issue requiring RC4. Python verification failed as well, and we need
an RC4 regardless to merge those PRs. I'm hoping that RC4 is our final RC,
and we can finish voting next week.

Reuven

On Sat, Nov 11, 2017 at 6:24 AM, Romain Manni-Bucau 
wrote:

> Le 11 nov. 2017 09:52, "Jean-Baptiste Onofré"  a écrit :
>
> If the purpose is to release 2.2.1 in one week, why not just to a RC4 ?
>
> It's not a regression because WriteFiles is new and extend the previous
> FileSource. So it could consider as a severe bug, especially on WriteFiles
> which is important.
>
>
> Fair enough.
>
>
> The core issue is the time we spent already on this release: roughly 1
> month !!! It's clearly too long due to different causes.
> When I did the previous releases, it took 3 or 4 days. It's clearly the
> target as, as said, I would like to have a release pace of a release every
> 6 weeks.
>
>
>
> Agree and this is why 2.2.0 must be out now IMHO. If you are confident next
> week is sufficient just go ahead and ignore my comment but my point was the
> same: it shouldnt last so long if there is no regression :(.
>
>
>
> Regards
> JB
>
>
> On 11/11/2017 08:41 AM, Romain Manni-Bucau wrote:
>
> > You can see it differently: is there a critical bug? Yes! Is there a
> > regression? No!
> >
> > So no need to wait another week (keep in mind 2 days + 3 days of vote
> > makes easily 1 working week). This vote could be closed already and next
> > week 2.2.1 could fix this bug, no? Overall idea is to not hold the
> > community more than needed if there is no regression compared to last few
> > releases.
> >
> > Le 11 nov. 2017 07:46, "Jean-Baptiste Onofré"  a écrit
> :
> >
> > -1 (binding)
> >>
> >> I agree with Eugene, data loss is severe.
> >>
> >> As Eugene seems confident to fix that quickly, I think it's worth to
> cut a
> >> RC4.
> >>
> >> However, I would introduce a deadline. As I would like to propose a
> >> release cycle of a release every 6 weeks (whatever it contains, but it
> >> really important to keep  a regular pace in releases), a release should
> be
> >> cut in couple of days. So, maybe we can give us 2 business days to fix
> >> that
> >> and propose a RC4. Basically, if this issue is not fix on Tuesday night,
> >> then, we move forward anyway.
> >>
> >> Regards
> >> JB
> >>
> >> On 11/10/2017 07:42 PM, Eugene Kirpichov wrote:
> >>
> >> Unfortunately I think I found a data loss bug - it was there since 2.0.0
> >>> but I think it's serious enough that delaying a fix until the next
> >>> release
> >>> would be irresponsible.
> >>> See https://issues.apache.org/jira/browse/BEAM-3169
> >>>
> >>> On Thu, Nov 9, 2017 at 3:57 PM Robert Bradshaw
> >>> 
> >>> wrote:
> >>>
> >>> Our release notes look like nothing more than a query for the closed
> >>>
>  jira issues. Do we have a top-level summary to highlight the big
>  ticket items in the release? And in particular somewhere to mention
>  that this is likely the last release to support Java 7 that'll get
>  widely read?
> 
>  On Thu, Nov 9, 2017 at 3:39 PM, Reuven Lax 
>  wrote:
> 
>  Thanks,
> >
> > This RC is currently failing on a number of validation steps, so we
> > need
> >
> > to
> 
>  cut at least one more RC. Fingers crossed that it will be the last
> one.
> >
> > Reuven
> >
> > On Thu, Nov 9, 2017 at 3:36 PM, Konstantinos Katsiapis <
> > katsia...@google.com.invalid> wrote:
> >
> > Just a remark: Release of Tensorflow Transform
> >
> >>  0.4.0 depends on release
> of
> >> Apache Beam 2.2.0 so upvoting for a release (the sooner the better).
> >>
> >> On Thu, Nov 9, 2017 at 3:33 PM, Reuven Lax  >
> >> wrote:
> >>
> >> Are we waiting for any more validation of this candidate? If people
> >>
> >>>
> >>> are
> >>
> >
>  still running tests I'll hold off on RC4 (to reduce the chance of an
> >
> >>
> >>> RC5),
> >>
> >> otherwise I'll cut RC4 once Valentyn's PR is merged.
> >>>
> >>> Reuven
> >>>
> >>> On Thu, Nov 9, 2017 at 2:26 PM, Valentyn Tymofieiev <
> >>> valen...@google.com.invalid> wrote:
> >>>
> >>> https://github.com/apache/beam/pull/4109 is out to address both
> >>>
> 
>  findings I
> >>>
> >>> reported earlier.
> 
>  On Thu,