You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@zookeeper.apache.org by Andor Molnar <an...@cloudera.com> on 2018/01/03 11:07:03 UTC
Bug in Zookeeper 3.4.11 release
Hi all,
Looks like we've a pretty serious bug in 3.4.11:
https://issues.apache.org/jira/browse/ZOOKEEPER-2960
In a nutshell: ZK uses dataDir and dataLogDir parameters in the wrong
order, meaning it tries to access logs in the snapshot folder and vica
versa.
Not sure what's the process in such case, but I believe we've to deprecate
3.4.11 and release a maintenance/hotfix/etc release asap.
Please advise.
In the meantime I'm creating a patch to address the issue on branch-3.4.
(other branches are not affected)
Thanks,
Andor
Re: Bug in Zookeeper 3.4.11 release
Posted by Andor Molnar <an...@cloudera.com>.
I created the jira and started to work on it.
https://issues.apache.org/jira/browse/ZOOKEEPER-2967
Andor
On Thu, Jan 18, 2018 at 6:15 AM, Andor Molnar <an...@cloudera.com> wrote:
> Perhaps we should introduce a check of some kind? If datalogdir is
> different that datadir and snapshots exist in datalogdir we throw an
> exception and quit. What do you think?
>
> That definitely makes sense.
>
> Andor
>
>
> On Thu, Jan 18, 2018 at 12:58 AM, Patrick Hunt <ph...@apache.org> wrote:
>
>> Abe mentioned the following on the jira "Perhaps we should introduce a
>> check of some kind?"
>>
>> I agree with Abe on this - if you're new to ZK in 3.4.11 and you split the
>> directories this commit is going to result in unexpected, and likely very
>> bad, situation.
>>
>> Abe - is there a jira to add this check? I'd like to resolve this and the
>> crc issue and push out a 3.4.12 ASAP.
>>
>> Patrick
>>
>>
>> On Mon, Jan 8, 2018 at 1:44 AM, Andor Molnar <an...@cloudera.com> wrote:
>>
>> > Thanks Camille. I closed the pull request.
>> >
>> > Andor
>> >
>> >
>> > On Mon, Jan 8, 2018 at 2:32 AM, Camille Fournier <ca...@apache.org>
>> > wrote:
>> >
>> > > Yeah, this is bad. I reviewed and merged the PR to get this process
>> > > started.
>> > >
>> > > On Fri, Jan 5, 2018 at 5:25 PM, Abraham Fine <af...@apache.org>
>> wrote:
>> > >
>> > > > I left a comment on the JIRA and I'm not sure what the conventional
>> way
>> > > to
>> > > > proceed would be either.
>> > > >
>> > > > I think we should work to get a release out as quickly as possible
>> with
>> > > > this fix. In addition, I think we may want to add a check on startup
>> > that
>> > > > forces a failure if snapshots/logs are where they shouldn't be. That
>> > way
>> > > we
>> > > > prevent people that are running 3.4.11 from experiencing data loss.
>> > > >
>> > > > Abe
>> > > >
>> > > > On Wed, Jan 3, 2018, at 03:07, Andor Molnar wrote:
>> > > > > Hi all,
>> > > > >
>> > > > > Looks like we've a pretty serious bug in 3.4.11:
>> > > > > https://issues.apache.org/jira/browse/ZOOKEEPER-2960
>> > > > >
>> > > > > In a nutshell: ZK uses dataDir and dataLogDir parameters in the
>> wrong
>> > > > > order, meaning it tries to access logs in the snapshot folder and
>> > vica
>> > > > > versa.
>> > > > >
>> > > > > Not sure what's the process in such case, but I believe we've to
>> > > > deprecate
>> > > > > 3.4.11 and release a maintenance/hotfix/etc release asap.
>> > > > >
>> > > > > Please advise.
>> > > > >
>> > > > > In the meantime I'm creating a patch to address the issue on
>> > > branch-3.4.
>> > > > > (other branches are not affected)
>> > > > >
>> > > > > Thanks,
>> > > > > Andor
>> > > >
>> > >
>> >
>>
>
>
Re: Bug in Zookeeper 3.4.11 release
Posted by Andor Molnar <an...@cloudera.com>.
Perhaps we should introduce a check of some kind? If datalogdir is
different that datadir and snapshots exist in datalogdir we throw an
exception and quit. What do you think?
That definitely makes sense.
Andor
On Thu, Jan 18, 2018 at 12:58 AM, Patrick Hunt <ph...@apache.org> wrote:
> Abe mentioned the following on the jira "Perhaps we should introduce a
> check of some kind?"
>
> I agree with Abe on this - if you're new to ZK in 3.4.11 and you split the
> directories this commit is going to result in unexpected, and likely very
> bad, situation.
>
> Abe - is there a jira to add this check? I'd like to resolve this and the
> crc issue and push out a 3.4.12 ASAP.
>
> Patrick
>
>
> On Mon, Jan 8, 2018 at 1:44 AM, Andor Molnar <an...@cloudera.com> wrote:
>
> > Thanks Camille. I closed the pull request.
> >
> > Andor
> >
> >
> > On Mon, Jan 8, 2018 at 2:32 AM, Camille Fournier <ca...@apache.org>
> > wrote:
> >
> > > Yeah, this is bad. I reviewed and merged the PR to get this process
> > > started.
> > >
> > > On Fri, Jan 5, 2018 at 5:25 PM, Abraham Fine <af...@apache.org> wrote:
> > >
> > > > I left a comment on the JIRA and I'm not sure what the conventional
> way
> > > to
> > > > proceed would be either.
> > > >
> > > > I think we should work to get a release out as quickly as possible
> with
> > > > this fix. In addition, I think we may want to add a check on startup
> > that
> > > > forces a failure if snapshots/logs are where they shouldn't be. That
> > way
> > > we
> > > > prevent people that are running 3.4.11 from experiencing data loss.
> > > >
> > > > Abe
> > > >
> > > > On Wed, Jan 3, 2018, at 03:07, Andor Molnar wrote:
> > > > > Hi all,
> > > > >
> > > > > Looks like we've a pretty serious bug in 3.4.11:
> > > > > https://issues.apache.org/jira/browse/ZOOKEEPER-2960
> > > > >
> > > > > In a nutshell: ZK uses dataDir and dataLogDir parameters in the
> wrong
> > > > > order, meaning it tries to access logs in the snapshot folder and
> > vica
> > > > > versa.
> > > > >
> > > > > Not sure what's the process in such case, but I believe we've to
> > > > deprecate
> > > > > 3.4.11 and release a maintenance/hotfix/etc release asap.
> > > > >
> > > > > Please advise.
> > > > >
> > > > > In the meantime I'm creating a patch to address the issue on
> > > branch-3.4.
> > > > > (other branches are not affected)
> > > > >
> > > > > Thanks,
> > > > > Andor
> > > >
> > >
> >
>
Re: Bug in Zookeeper 3.4.11 release
Posted by Patrick Hunt <ph...@apache.org>.
Abe mentioned the following on the jira "Perhaps we should introduce a
check of some kind?"
I agree with Abe on this - if you're new to ZK in 3.4.11 and you split the
directories this commit is going to result in unexpected, and likely very
bad, situation.
Abe - is there a jira to add this check? I'd like to resolve this and the
crc issue and push out a 3.4.12 ASAP.
Patrick
On Mon, Jan 8, 2018 at 1:44 AM, Andor Molnar <an...@cloudera.com> wrote:
> Thanks Camille. I closed the pull request.
>
> Andor
>
>
> On Mon, Jan 8, 2018 at 2:32 AM, Camille Fournier <ca...@apache.org>
> wrote:
>
> > Yeah, this is bad. I reviewed and merged the PR to get this process
> > started.
> >
> > On Fri, Jan 5, 2018 at 5:25 PM, Abraham Fine <af...@apache.org> wrote:
> >
> > > I left a comment on the JIRA and I'm not sure what the conventional way
> > to
> > > proceed would be either.
> > >
> > > I think we should work to get a release out as quickly as possible with
> > > this fix. In addition, I think we may want to add a check on startup
> that
> > > forces a failure if snapshots/logs are where they shouldn't be. That
> way
> > we
> > > prevent people that are running 3.4.11 from experiencing data loss.
> > >
> > > Abe
> > >
> > > On Wed, Jan 3, 2018, at 03:07, Andor Molnar wrote:
> > > > Hi all,
> > > >
> > > > Looks like we've a pretty serious bug in 3.4.11:
> > > > https://issues.apache.org/jira/browse/ZOOKEEPER-2960
> > > >
> > > > In a nutshell: ZK uses dataDir and dataLogDir parameters in the wrong
> > > > order, meaning it tries to access logs in the snapshot folder and
> vica
> > > > versa.
> > > >
> > > > Not sure what's the process in such case, but I believe we've to
> > > deprecate
> > > > 3.4.11 and release a maintenance/hotfix/etc release asap.
> > > >
> > > > Please advise.
> > > >
> > > > In the meantime I'm creating a patch to address the issue on
> > branch-3.4.
> > > > (other branches are not affected)
> > > >
> > > > Thanks,
> > > > Andor
> > >
> >
>
Re: Bug in Zookeeper 3.4.11 release
Posted by Andor Molnar <an...@cloudera.com>.
Thanks Camille. I closed the pull request.
Andor
On Mon, Jan 8, 2018 at 2:32 AM, Camille Fournier <ca...@apache.org> wrote:
> Yeah, this is bad. I reviewed and merged the PR to get this process
> started.
>
> On Fri, Jan 5, 2018 at 5:25 PM, Abraham Fine <af...@apache.org> wrote:
>
> > I left a comment on the JIRA and I'm not sure what the conventional way
> to
> > proceed would be either.
> >
> > I think we should work to get a release out as quickly as possible with
> > this fix. In addition, I think we may want to add a check on startup that
> > forces a failure if snapshots/logs are where they shouldn't be. That way
> we
> > prevent people that are running 3.4.11 from experiencing data loss.
> >
> > Abe
> >
> > On Wed, Jan 3, 2018, at 03:07, Andor Molnar wrote:
> > > Hi all,
> > >
> > > Looks like we've a pretty serious bug in 3.4.11:
> > > https://issues.apache.org/jira/browse/ZOOKEEPER-2960
> > >
> > > In a nutshell: ZK uses dataDir and dataLogDir parameters in the wrong
> > > order, meaning it tries to access logs in the snapshot folder and vica
> > > versa.
> > >
> > > Not sure what's the process in such case, but I believe we've to
> > deprecate
> > > 3.4.11 and release a maintenance/hotfix/etc release asap.
> > >
> > > Please advise.
> > >
> > > In the meantime I'm creating a patch to address the issue on
> branch-3.4.
> > > (other branches are not affected)
> > >
> > > Thanks,
> > > Andor
> >
>
Re: Bug in Zookeeper 3.4.11 release
Posted by Camille Fournier <ca...@apache.org>.
Yeah, this is bad. I reviewed and merged the PR to get this process started.
On Fri, Jan 5, 2018 at 5:25 PM, Abraham Fine <af...@apache.org> wrote:
> I left a comment on the JIRA and I'm not sure what the conventional way to
> proceed would be either.
>
> I think we should work to get a release out as quickly as possible with
> this fix. In addition, I think we may want to add a check on startup that
> forces a failure if snapshots/logs are where they shouldn't be. That way we
> prevent people that are running 3.4.11 from experiencing data loss.
>
> Abe
>
> On Wed, Jan 3, 2018, at 03:07, Andor Molnar wrote:
> > Hi all,
> >
> > Looks like we've a pretty serious bug in 3.4.11:
> > https://issues.apache.org/jira/browse/ZOOKEEPER-2960
> >
> > In a nutshell: ZK uses dataDir and dataLogDir parameters in the wrong
> > order, meaning it tries to access logs in the snapshot folder and vica
> > versa.
> >
> > Not sure what's the process in such case, but I believe we've to
> deprecate
> > 3.4.11 and release a maintenance/hotfix/etc release asap.
> >
> > Please advise.
> >
> > In the meantime I'm creating a patch to address the issue on branch-3.4.
> > (other branches are not affected)
> >
> > Thanks,
> > Andor
>
Re: Bug in Zookeeper 3.4.11 release
Posted by Abraham Fine <af...@apache.org>.
I left a comment on the JIRA and I'm not sure what the conventional way to proceed would be either.
I think we should work to get a release out as quickly as possible with this fix. In addition, I think we may want to add a check on startup that forces a failure if snapshots/logs are where they shouldn't be. That way we prevent people that are running 3.4.11 from experiencing data loss.
Abe
On Wed, Jan 3, 2018, at 03:07, Andor Molnar wrote:
> Hi all,
>
> Looks like we've a pretty serious bug in 3.4.11:
> https://issues.apache.org/jira/browse/ZOOKEEPER-2960
>
> In a nutshell: ZK uses dataDir and dataLogDir parameters in the wrong
> order, meaning it tries to access logs in the snapshot folder and vica
> versa.
>
> Not sure what's the process in such case, but I believe we've to deprecate
> 3.4.11 and release a maintenance/hotfix/etc release asap.
>
> Please advise.
>
> In the meantime I'm creating a patch to address the issue on branch-3.4.
> (other branches are not affected)
>
> Thanks,
> Andor