cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rajani Karuturi <>
Subject Re: [VOTE] Apache Cloudstack RC3
Date Wed, 28 Jun 2017 06:58:53 GMT
I don't think creating a branch will help in releasing faster. It
will only make it worse in my opinion.

If we can release faster, features will stay in the PR branch for
a short while and can be merged quickly.

~ Rajani

On June 28, 2017 at 12:17 PM, Daan Hoogland
( wrote:

I'm with Mike on this. fixes go into the rc branch, features
don't and
that's a clearer line then we have now. or we could just keep
untill one passes and keep working on stablising whichever
branch we
choose for that allowing both features and fixes.

On Wed, Jun 28, 2017 at 8:40 AM, Tutkowski, Mike
<> wrote:


I personally still like the idea of a new branch being created
right around the time we cut our first RC.

Even if people want to commit changes to the new branch, they
should understand that that code won't be formally released until
the pending RC is validated and released.

That being the case, I would think those who choose to commit to
the new branch will have a vested interest in the RC going out,
as well.

In any event, in addition to the current automated regression
tests that are run, we still have a lot of tests that are not
hooked into the build that are being run ad hoc (managed storage
automated tests are an example). Additionally, we seem to have a
lot of manual tests being run.

Until we can deliver a framework in which we have a very high
percentage of the system covered by automated tests, there is
really no way we should consider monthly releases.

I think we are still shooting for releases every four months,
which seems fair given our current system.

If we enact some deadlines like a code freeze going forward,
that should help. With only blocker PRs going into subsequent
RCs, we should be able to avoid a lot of unnecessary spin.

I definitely want to point out that I appreciate everyone's time
and effort. In particular, I want to be clear that it is not my
intent to be critical of anyone who's been working in release
management. My only goal with this chain of e-mails is to see if
we can continue to improve the process.

Thanks, everyone!

On Jun 27, 2017, at 11:14 PM, Rajani Karuturi <>

We can do a release every month as long as we have enough people
actively participating in the release process.

We have people who wants to have their code/features checked in.
We, very clearly do not have enough people working on
releases/blockers. How many of us are testing/voting on releases
or PRs? We have blockers in jira, with no one to fix. We have
open for release blockers for more than a month with no one to

I would ask everyone to start testing releases/PRs and voting on
them actively.

We need people who can do the work. We already know what needs
be done as outlined in the release principles wiki after long
discussions on this list.

Whether we create a branch off RC or continue on master wont
change the current situation.

We, as community should commit to testing and releasing code.
principles and theory wont help.


~ Rajani

On June 27, 2017 at 9:43 PM, Rafael Weingärtner
( wrote:

+1 to what Paul said.
IMHO, as soon as we start a release candidate to close a
version, all
merges should stop (period); the only exceptions should be PRs
that address
specific problems in the RC.
I always thought that we had a protocol for that [1]; maybe for
version, we have not followed it?


On Tue, Jun 27, 2017 at 1:32 AM, Paul Angus

Hi All,

>From my view point 'we' have been the architects of our own
downfall. Once
a code freeze is in place NO new features, NO enhancements
should be going
in. Once we're at an RC stage, NO new bug fixes other that for
the blockers
should be going in. that way the release gets out, and the next
one can get
going. If 4.10 had gone out in a timely fashion, then we'd
probably be on
4.11 if not 4.12 by now, with all the new features AND all the
new fixes in.

People sliding new changes/bug fixes/enhancements in are not
making the
product better, they're stopping progress. As we can clearly see

Kind regards,

Paul Angus ( ) 
( )
53 Chandos Place, Covent Garden, London WC2N 4HSUK

-----Original Message-----
From: Tutkowski, Mike []
Sent: 27 June 2017 01:25
Cc: Wido den Hollander <>
Subject: Re: [VOTE] Apache Cloudstack RC3

I tend to agree with you here, Daan. I know the downside we’ve
in the past is that overall community participation in the RC
process has
dropped off when such a new branch is created (since the
community as a
whole tends to focus more on the new branch rather than on
testing the RC
and releasing it).

I believe we should do the following: As we approach the first
RC, we need
to limit the number of PRs going into the branch (in order to
it). If we had a super duper array of automated regression tests
that ran
against the code, then we might be able to avoid this, but our
test suite is not extensive enough for us to do so.

As we approach the first RC, only blockers and trivial (ex. text
PRs should be permitted in. Once we cut the first RC, create a
new branch
for ongoing dev work. In between RCs, we can only allow in code
related to
blocker PRs (or trivial text changes, as discussed before).

What do people think?

On 6/13/17, 4:56 AM, "Daan Hoogland" <>

this is why i say we should branch on first RC, fix in release
only and merge forward

On Tue, Jun 13, 2017 at 12:41 PM, Will Stevens <> wrote:

I know it is hard to justify not merging PRs that seem ready but


blockers in an RC, but it is a vicious circle which ultimately

results in a

longer RC process.

It is something i struggled with as a release manager as well.

On Jun 13, 2017 1:56 AM, "Rajani Karuturi" <>


Thanks Mike,

Will hold off next RC until we hear an update from you.

Regarding merging non-blockers, unfortunately, its a side-effect
of taking more than three months in the RC phase :(


~ Rajani

On June 13, 2017 at 10:10 AM, Tutkowski, Mike
( wrote:

Hi everyone,

I had a little time this evening and re-ran some VMware-related
tests around managed storage. I noticed a problem that I’d like
to investigate before we spin up the next RC. Let’s hold off on
the next RC until I can find out more (I should know more within
24 hours).


On 6/12/17, 2:40 AM, "Wido den Hollander" <>

Op 10 juni 2017 om 21:18 schreef "Tutkowski, Mike"



I opened a PR against the most recent RC:

I ran all managed-storage regression tests against it and they

pass (as noted in detail in the PR).

If someone wants to take this code and create a new RC from

it, I’m +1 on the new RC as long as this is the only commit
addedto it since the current RC.

Thanks Mike!

If this PR is good we should probably merge it asap and go for

4.10 should really be released by now.



On 6/9/17, 7:43 PM, "Tutkowski, Mike"

<> wrote:

Hi everyone,

I found a critical issue that was introduced into this RC

since the most recent RC, so I am -1 on this RC.

The fix for this ticket breaks the support for storing volume

snapshots on primary storage (which is a feature managed
storagecan support):

Here is the SHA: 336df84f1787de962a67d0a34551f9027303040e

At a high level, what it does is remove a row from the

cloud.snapshot_store_ref table when a volume is deleted that
hasone or more volume snapshots.

This is fine for non-managed (traditional) storage; however,

managed storage can store volume snapshots on primary storage,
soremoving this row breaks that functionality.

I can fix the problem that this commit introduced by looking

at the primary storage that supports the volume snapshot
andchecking the following: 1) Is this managed storage? 2) If
the snapshot in question stored on that primary storage?

The problem is I will be out of the office for a couple weeks

and will not be able to address this until I return.

We could revert the commit, but I still will not have time to

run the managed-storage regression test suite until I return.

On a side note, it looks like this commit was introduced since

the most recent RC. I would argue that it was not a blocker
andshould not have been placed into the new RC. We (as a
tend to have a lot of code go in between RCs and that just
increases the chances of introducing critical issues and thus
delaying the release. We’ve gotten better at this over the
but we should focus more on only allowing the entry of new code
into a follow-on RC that is critical (or so trivial as to not at
all be likely to introduce any problems…like fixing an error

Thanks for your efforts on this, everyone!

On 6/9/17, 8:52 AM, "Tutkowski, Mike"

<> wrote:

Hi Rajani,

I will see if I can get all of my managed-storage testing

(both automated and manual) done today. If not, we’ll need to
seeif someone else can complete it before we OK this RC as I
be back in the office for a couple weeks. I’ll report back later


On 6/9/17, 2:34 AM, "Rajani Karuturi" <>


Yup. thats right. I dont know how it happened but, it created
from the previous RC commit. The script is supposed to do a


pull. I didn't notice any failures. Not sure what went wrong.

Thanks for finding it mike. I am creating RC4 now and



~ Rajani

On June 9, 2017 at 12:07 PM, Tutkowski, Mike
( wrote:

Hi Rajani,

I don’t see the following PR in this RC:

I ran all of my managed-storage regression tests. They all
passed with the exception of the one that led to PR 2098.

As I examine the RC in a bit more detail, it sits on top of
ed2f573, but I think it should sit on top of ed376fc.

As a result, I am -1 on the RC.

It takes me about a day to run all of the managed-storage
regression tests and I am out of the office for the next


weeks, so I’d really like to avoid another RC until I’m back


able to test the next RC.


On 6/7/17, 4:36 AM, "Rajani Karuturi" <>


Hi All,

I've created release with the following artifacts up
for a vote:

Git Branch and Commit SH:;a=commit;h=a55738a31d0073f2925c6fb84bf7a6bb32f4ca27


Source release (checksums and signatures are available at the

SystemVm Templates:

PGP release keys (signed using CBB44821):

Vote will be open for 72 hours.

For sanity in tallying the vote, can PMC members please be


to indicate
"(binding)" with their vote?

[ ] +1 approve
[ ] +0 no opinion
[ ] -1 disapprove (and reason why)



Rafael Weingärtner

  • Unnamed multipart/alternative (inline, 7-Bit, 0 bytes)
View raw message