Thanks Elliott, I was trying to understand how come the patch for HBASE-14689 <https://issues.apache.org/jira/browse/HBASE-14689> was causing an issue in 1.1, but not in 0.98. I was not able to reproduce the condition. It looks to be a deadlock condition caused by un-ordered acquisition of row locks ( https://issues.apache.org/jira/browse/HBASE-14689?focusedCommentId=15010417&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15010417 ).
Andrew, also in 0.98, we do not have any guarantee for the lock order in doMiniBatchMutation(), so I think it is safest to roll another RC although we cannot reproduce the deadlock. Enis On Wed, Nov 18, 2015 at 9:07 AM, Andrew Purtell <[email protected]> wrote: > This makes me nervous so I am going to make a patch release that is the > 0.98.16 change set plus this one revert. Let me do that today. Appreciate > it if we can get a quick turn around on voting. I'll aim for 72 hours. > > > > On Nov 17, 2015, at 8:47 PM, Enis Söztutar <[email protected]> wrote: > > > > Thanks Andrew, > > > > Maybe we can hold off the release announcement at least, and wait until > we > > investigate further. I'll look at the batching and how row locks are > > different between 0.98 and branch-1 tomorrow. > > > > Enis > > > > On Tue, Nov 17, 2015 at 8:31 PM, Andrew Purtell < > [email protected]> > > wrote: > > > >> Thanks Enis. Appreciate the investigation. > >> > >> Yes I did release this. We can make a patch release though if needed. > >> > >> > >>> On Nov 17, 2015, at 8:04 PM, Enis Söztutar <[email protected]> wrote: > >>> > >>> I have tried the same pe command without small values, and cannot > >> reproduce > >>> the row lock issue in this RC. > >>> > >>> Setup is single node with default settings, using a single SSD. > >>> > >>> bin/hbase pe --latency --nomapred --presplit=10 randomWrite 10 > >>> This same exact command, I can reproduce the hang in 1.0.3RC and > 1.1.3RC. > >>> > >>> Enis > >>> > >>>> On Tue, Nov 17, 2015 at 7:44 PM, Enis Söztutar <[email protected]> > >> wrote: > >>>> > >>>> Andrew did you push the artifacts yet? > >>>> > >>>> Let me see whether this RC is also affected by > >>>> https://issues.apache.org/jira/browse/HBASE-14689 > >>>> > >>>> Enis > >>>> > >>>> On Tue, Nov 17, 2015 at 12:35 PM, Andrew Purtell <[email protected] > > > >>>> wrote: > >>>> > >>>>> With three +1 votes and no 0 or -1 votes, this vote passes and we > have > >>>>> 0.98.16. I will send the artifacts onward for distribution and send > >> out a > >>>>> release announcement shortly. > >>>>> > >>>>> Thanks to all who voted on this candidate. > >>>>> > >>>>> > >>>>> On Tue, Nov 10, 2015 at 10:31 AM, Andrew Purtell < > [email protected]> > >>>>> wrote: > >>>>> > >>>>>> > >>>>>> The 1st HBase 0.98.16 release candidate (RC0) is available for > >> download > >>>>> at > >>>>>> http://people.apache.org/~apurtell/0.98.16RC0/ and Maven artifacts > >> are > >>>>>> also available in the temporary repository > >> https://repository.apache.org/content/repositories/orgapachehbase-1118/ > >>>>> . > >>>>>> > >>>>>> This release does not contain a doc update. Usually I copy back docs > >>>>> from > >>>>>> master. This time I encountered a build error, reported on > >> HBASE-14792. > >>>>>> This isn't a critical (or even major) issue for any given release > >> since > >>>>> doc > >>>>>> updates for 0.98 are rare now, but should be fixed for the next > >> release. > >>>>>> > >>>>>> The API compatibility report with respect to the previous release > can > >> be > >>>>>> found at > >> > http://people.apache.org/~apurtell/0.98.15_0.98.16RC0_compat_report.html > >>>>> . > >>>>>> There are no reported binary or source compatibility issues with > >>>>>> user facing public or LimitedPrivate interfaces. Users should not be > >>>>>> implementing the HConnection interface so will not be impacted by > the > >>>>>> addition of a new abstract method. > >>>>>> > >>>>>> Signed with my code signing key D5365CCD. > >>>>>> > >>>>>> The 47 issues resolved in this release can be found at > >>>>>> http://s.apache.org/Xbv . > >>>>>> > >>>>>> Please try out the candidate and vote +1/0/-1. This vote will be > open > >>>>> for > >>>>>> at least 72 hours. Unless objection I will try to close it Monday > >>>>> October > >>>>>> 16, 2015 if we have sufficient votes. Three +1 votes from PMC will > be > >>>>>> required to release. > >>>>> > >>>>> -- > >>>>> Best regards, > >>>>> > >>>>> - Andy > >>>>> > >>>>> Problems worthy of attack prove their worth by hitting back. - Piet > >> Hein > >>>>> (via Tom White) > >> >
