You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hbase.apache.org by Andrew Purtell <an...@gmail.com> on 2019/04/11 16:10:15 UTC
Branch-1 releasing paused; HBase community: please help clean up your test code
I have tried to release four 1.5.0 candidates from head of branch-1.
Some veto votes were cast for compatibility issues. This was fine.
Others are for unit test results that do not reproduce locally for me. I do not have the ability or bandwidth to fix tests which do not fail reliably for me. So let me appeal to the community. If you find a repeatable test failure on branch-1 please file a JIRA and fix it.
As things stand now I am pausing any attempts to make more 1.5.0 release candidates until the community steps forward to clean up its code. As an alternative I may start aggressively disabling unit tests which do not fail for me but have been reported in on candidate vote vetoes. Otherwise it is impossible to make forward progress.
Thank you for your attention.
Re: Branch-1 releasing paused; HBase community: please help clean up
your test code
Posted by Andrew Purtell <ap...@apache.org>.
Pardon, that git ref I failed to include below is commit 677ed9ef89 (HEAD
-> branch-1)
On Thu, Apr 11, 2019 at 12:14 PM Andrew Purtell <ap...@apache.org> wrote:
> Thanks for clarifying that this is not a foundation wide policy. It is my
> personal policy to cancel a vote whenever there is a veto, if the reason is
> sound and technical and, presumably, repeatable, at least in the voter's
> environment(s).
>
> Circling back to the main topic, a unit test run of latest branch-1 (git
> ref) completed successfully on my dev host. Based on this result, I would
> release it. I'll include details on my environment below.
>
> It is my theory that the reason some environments observe flaky unit test
> failures and others don't is there are occasional unexpected interactions
> between various units when surefire executes them concurrently, and the
> order in which unit tests are run is dependent on how the underlying OS's
> readdir() call orders directory entries, and this will vary from host to
> host and even from checkout to checkout.
>
> My dev host details
>
> apurtell$ uname -a
> Darwin HOSTNAME 17.7.0 Darwin Kernel Version 17.7.0: Thu Dec 20 21:47:19
> PST 2018; root:xnu-4570.71.22~1/RELEASE_X86_64 x86_64
>
> apurtell$ mvn -v
> Apache Maven 3.6.0 (97c98ec64a1fdfee7767ce5ffb20918da4f719f3;
> 2018-10-24T11:41:47-07:00)
> Maven home: /usr/local/Cellar/maven/3.6.0/libexec
> Java version: 1.8.0_172, vendor: Azul Systems, Inc., runtime:
> /Users/apurtell/tools/Darwin/jdk/openjdk1.8.0_172_x64/jre
> Default locale: en_US, platform encoding: UTF-8
> OS name: "mac os x", version: "10.13.6", arch: "x86_64", family: "mac"
>
>
> On Thu, Apr 11, 2019 at 11:11 AM Sean Busbey <bu...@apache.org> wrote:
>
>> Just as a point of clarification, -1 votes on releases aren't vetos. ASF
>> policy requires release votes to be majority.
>>
>> As release manager it's your perogative to cancel a vote for whatever
>> reason, but I don't want future RMs thinking they have to fail the vote
>> once there's a -1.
>>
>> On Thu, Apr 11, 2019, 12:18 Andrew Purtell <an...@gmail.com>
>> wrote:
>>
>> > I put back "1.6.0" for TinyLFU backport to branch-1, which I don't think
>> > can happen in the near term because it depends on improvements to
>> precommit
>> > being in place first.
>> >
>> >
>> > On Thu, Apr 11, 2019 at 9:37 AM Andrew Purtell <
>> andrew.purtell@gmail.com>
>> > wrote:
>> >
>> > > A helpful view for tracking 1.5.0 issues is
>> > > https://issues.apache.org/jira/projects/HBASE/versions/12340316 .
>> > >
>> > > I deleted fixversions "1.5.1" and "1.6.0" and had JIRA rewrite them
>> back
>> > > to "1.5.0" so you won't be confused by JIRA thinking 1.5.0 is in
>> released
>> > > state. All open and pending branch-1 work has the "1.5.0" fix version
>> > > again, until we try again for a release at some future time.
>> > >
>> > >
>> > > On Thu, Apr 11, 2019 at 9:10 AM Andrew Purtell <
>> andrew.purtell@gmail.com
>> > >
>> > > wrote:
>> > >
>> > >> I have tried to release four 1.5.0 candidates from head of branch-1.
>> > >>
>> > >> Some veto votes were cast for compatibility issues. This was fine.
>> > >>
>> > >> Others are for unit test results that do not reproduce locally for
>> me. I
>> > >> do not have the ability or bandwidth to fix tests which do not fail
>> > >> reliably for me. So let me appeal to the community. If you find a
>> > >> repeatable test failure on branch-1 please file a JIRA and fix it.
>> > >>
>> > >> As things stand now I am pausing any attempts to make more 1.5.0
>> > release
>> > >> candidates until the community steps forward to clean up its code.
>> As an
>> > >> alternative I may start aggressively disabling unit tests which do
>> not
>> > fail
>> > >> for me but have been reported in on candidate vote vetoes. Otherwise
>> it
>> > is
>> > >> impossible to make forward progress.
>> > >>
>> > >> Thank you for your attention.
>>
>
Re: Branch-1 releasing paused; HBase community: please help clean up
your test code
Posted by Andrew Purtell <ap...@apache.org>.
Thanks for clarifying that this is not a foundation wide policy. It is my
personal policy to cancel a vote whenever there is a veto, if the reason is
sound and technical and, presumably, repeatable, at least in the voter's
environment(s).
Circling back to the main topic, a unit test run of latest branch-1 (git
ref) completed successfully on my dev host. Based on this result, I would
release it. I'll include details on my environment below.
It is my theory that the reason some environments observe flaky unit test
failures and others don't is there are occasional unexpected interactions
between various units when surefire executes them concurrently, and the
order in which unit tests are run is dependent on how the underlying OS's
readdir() call orders directory entries, and this will vary from host to
host and even from checkout to checkout.
My dev host details
apurtell$ uname -a
Darwin HOSTNAME 17.7.0 Darwin Kernel Version 17.7.0: Thu Dec 20 21:47:19
PST 2018; root:xnu-4570.71.22~1/RELEASE_X86_64 x86_64
apurtell$ mvn -v
Apache Maven 3.6.0 (97c98ec64a1fdfee7767ce5ffb20918da4f719f3;
2018-10-24T11:41:47-07:00)
Maven home: /usr/local/Cellar/maven/3.6.0/libexec
Java version: 1.8.0_172, vendor: Azul Systems, Inc., runtime:
/Users/apurtell/tools/Darwin/jdk/openjdk1.8.0_172_x64/jre
Default locale: en_US, platform encoding: UTF-8
OS name: "mac os x", version: "10.13.6", arch: "x86_64", family: "mac"
On Thu, Apr 11, 2019 at 11:11 AM Sean Busbey <bu...@apache.org> wrote:
> Just as a point of clarification, -1 votes on releases aren't vetos. ASF
> policy requires release votes to be majority.
>
> As release manager it's your perogative to cancel a vote for whatever
> reason, but I don't want future RMs thinking they have to fail the vote
> once there's a -1.
>
> On Thu, Apr 11, 2019, 12:18 Andrew Purtell <an...@gmail.com>
> wrote:
>
> > I put back "1.6.0" for TinyLFU backport to branch-1, which I don't think
> > can happen in the near term because it depends on improvements to
> precommit
> > being in place first.
> >
> >
> > On Thu, Apr 11, 2019 at 9:37 AM Andrew Purtell <andrew.purtell@gmail.com
> >
> > wrote:
> >
> > > A helpful view for tracking 1.5.0 issues is
> > > https://issues.apache.org/jira/projects/HBASE/versions/12340316 .
> > >
> > > I deleted fixversions "1.5.1" and "1.6.0" and had JIRA rewrite them
> back
> > > to "1.5.0" so you won't be confused by JIRA thinking 1.5.0 is in
> released
> > > state. All open and pending branch-1 work has the "1.5.0" fix version
> > > again, until we try again for a release at some future time.
> > >
> > >
> > > On Thu, Apr 11, 2019 at 9:10 AM Andrew Purtell <
> andrew.purtell@gmail.com
> > >
> > > wrote:
> > >
> > >> I have tried to release four 1.5.0 candidates from head of branch-1.
> > >>
> > >> Some veto votes were cast for compatibility issues. This was fine.
> > >>
> > >> Others are for unit test results that do not reproduce locally for
> me. I
> > >> do not have the ability or bandwidth to fix tests which do not fail
> > >> reliably for me. So let me appeal to the community. If you find a
> > >> repeatable test failure on branch-1 please file a JIRA and fix it.
> > >>
> > >> As things stand now I am pausing any attempts to make more 1.5.0
> > release
> > >> candidates until the community steps forward to clean up its code. As
> an
> > >> alternative I may start aggressively disabling unit tests which do not
> > fail
> > >> for me but have been reported in on candidate vote vetoes. Otherwise
> it
> > is
> > >> impossible to make forward progress.
> > >>
> > >> Thank you for your attention.
> > >>
> > >>
> > >
> > > --
> > > Best regards,
> > > Andrew
> > >
> > > Words like orphans lost among the crosstalk, meaning torn from truth's
> > > decrepit hands
> > > - A23, Crosstalk
> > >
> >
> >
> > --
> > Best regards,
> > Andrew
> >
> > Words like orphans lost among the crosstalk, meaning torn from truth's
> > decrepit hands
> > - A23, Crosstalk
> >
>
--
Best regards,
Andrew
Words like orphans lost among the crosstalk, meaning torn from truth's
decrepit hands
- A23, Crosstalk
Re: Branch-1 releasing paused; HBase community: please help clean up
your test code
Posted by Sean Busbey <bu...@apache.org>.
Just as a point of clarification, -1 votes on releases aren't vetos. ASF
policy requires release votes to be majority.
As release manager it's your perogative to cancel a vote for whatever
reason, but I don't want future RMs thinking they have to fail the vote
once there's a -1.
On Thu, Apr 11, 2019, 12:18 Andrew Purtell <an...@gmail.com> wrote:
> I put back "1.6.0" for TinyLFU backport to branch-1, which I don't think
> can happen in the near term because it depends on improvements to precommit
> being in place first.
>
>
> On Thu, Apr 11, 2019 at 9:37 AM Andrew Purtell <an...@gmail.com>
> wrote:
>
> > A helpful view for tracking 1.5.0 issues is
> > https://issues.apache.org/jira/projects/HBASE/versions/12340316 .
> >
> > I deleted fixversions "1.5.1" and "1.6.0" and had JIRA rewrite them back
> > to "1.5.0" so you won't be confused by JIRA thinking 1.5.0 is in released
> > state. All open and pending branch-1 work has the "1.5.0" fix version
> > again, until we try again for a release at some future time.
> >
> >
> > On Thu, Apr 11, 2019 at 9:10 AM Andrew Purtell <andrew.purtell@gmail.com
> >
> > wrote:
> >
> >> I have tried to release four 1.5.0 candidates from head of branch-1.
> >>
> >> Some veto votes were cast for compatibility issues. This was fine.
> >>
> >> Others are for unit test results that do not reproduce locally for me. I
> >> do not have the ability or bandwidth to fix tests which do not fail
> >> reliably for me. So let me appeal to the community. If you find a
> >> repeatable test failure on branch-1 please file a JIRA and fix it.
> >>
> >> As things stand now I am pausing any attempts to make more 1.5.0
> release
> >> candidates until the community steps forward to clean up its code. As an
> >> alternative I may start aggressively disabling unit tests which do not
> fail
> >> for me but have been reported in on candidate vote vetoes. Otherwise it
> is
> >> impossible to make forward progress.
> >>
> >> Thank you for your attention.
> >>
> >>
> >
> > --
> > Best regards,
> > Andrew
> >
> > Words like orphans lost among the crosstalk, meaning torn from truth's
> > decrepit hands
> > - A23, Crosstalk
> >
>
>
> --
> Best regards,
> Andrew
>
> Words like orphans lost among the crosstalk, meaning torn from truth's
> decrepit hands
> - A23, Crosstalk
>
Re: Branch-1 releasing paused; HBase community: please help clean up
your test code
Posted by Andrew Purtell <an...@gmail.com>.
I put back "1.6.0" for TinyLFU backport to branch-1, which I don't think
can happen in the near term because it depends on improvements to precommit
being in place first.
On Thu, Apr 11, 2019 at 9:37 AM Andrew Purtell <an...@gmail.com>
wrote:
> A helpful view for tracking 1.5.0 issues is
> https://issues.apache.org/jira/projects/HBASE/versions/12340316 .
>
> I deleted fixversions "1.5.1" and "1.6.0" and had JIRA rewrite them back
> to "1.5.0" so you won't be confused by JIRA thinking 1.5.0 is in released
> state. All open and pending branch-1 work has the "1.5.0" fix version
> again, until we try again for a release at some future time.
>
>
> On Thu, Apr 11, 2019 at 9:10 AM Andrew Purtell <an...@gmail.com>
> wrote:
>
>> I have tried to release four 1.5.0 candidates from head of branch-1.
>>
>> Some veto votes were cast for compatibility issues. This was fine.
>>
>> Others are for unit test results that do not reproduce locally for me. I
>> do not have the ability or bandwidth to fix tests which do not fail
>> reliably for me. So let me appeal to the community. If you find a
>> repeatable test failure on branch-1 please file a JIRA and fix it.
>>
>> As things stand now I am pausing any attempts to make more 1.5.0 release
>> candidates until the community steps forward to clean up its code. As an
>> alternative I may start aggressively disabling unit tests which do not fail
>> for me but have been reported in on candidate vote vetoes. Otherwise it is
>> impossible to make forward progress.
>>
>> Thank you for your attention.
>>
>>
>
> --
> Best regards,
> Andrew
>
> Words like orphans lost among the crosstalk, meaning torn from truth's
> decrepit hands
> - A23, Crosstalk
>
--
Best regards,
Andrew
Words like orphans lost among the crosstalk, meaning torn from truth's
decrepit hands
- A23, Crosstalk
Re: Branch-1 releasing paused; HBase community: please help clean up
your test code
Posted by Andrew Purtell <an...@gmail.com>.
A helpful view for tracking 1.5.0 issues is
https://issues.apache.org/jira/projects/HBASE/versions/12340316 .
I deleted fixversions "1.5.1" and "1.6.0" and had JIRA rewrite them back to
"1.5.0" so you won't be confused by JIRA thinking 1.5.0 is in released
state. All open and pending branch-1 work has the "1.5.0" fix version
again, until we try again for a release at some future time.
On Thu, Apr 11, 2019 at 9:10 AM Andrew Purtell <an...@gmail.com>
wrote:
> I have tried to release four 1.5.0 candidates from head of branch-1.
>
> Some veto votes were cast for compatibility issues. This was fine.
>
> Others are for unit test results that do not reproduce locally for me. I
> do not have the ability or bandwidth to fix tests which do not fail
> reliably for me. So let me appeal to the community. If you find a
> repeatable test failure on branch-1 please file a JIRA and fix it.
>
> As things stand now I am pausing any attempts to make more 1.5.0 release
> candidates until the community steps forward to clean up its code. As an
> alternative I may start aggressively disabling unit tests which do not fail
> for me but have been reported in on candidate vote vetoes. Otherwise it is
> impossible to make forward progress.
>
> Thank you for your attention.
>
>
--
Best regards,
Andrew
Words like orphans lost among the crosstalk, meaning torn from truth's
decrepit hands
- A23, Crosstalk