hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Yu <yuzhih...@gmail.com>
Subject Re: meta region written by 2.0 cannot be opened by 1.x
Date Wed, 07 Dec 2016 10:29:11 GMT
Shouldn't HBASE-14949 be marked Incompatible Change ?

On Wed, Dec 7, 2016 at 2:17 AM, 张铎 <palomino219@gmail.com> wrote:

> See this one
>
> https://issues.apache.org/jira/browse/HBASE-14949
>
> 2016-12-07 18:09 GMT+08:00 Ted Yu <yuzhihong@gmail.com>:
>
> > Which other features ?
> > Can you name them ?
> >
> > Thanks
> >
> > On Wed, Dec 7, 2016 at 2:06 AM, 张铎 <palomino219@gmail.com> wrote:
> >
> > > I believe there are other features that require upgrading to a specific
> > > version in each 1.x release before rolling upgrading to 2.0? I think
> this
> > > is a typical trick to add incompatible changes...
> > >
> > > 2016-12-07 18:02 GMT+08:00 ramkrishna vasudevan <
> > > ramkrishna.s.vasudevan@gmail.com>:
> > >
> > > > If its going to be a pain then we need to support writing the older
> > class
> > > > name in the FFT (trailer) and then use our internal mapping.
> > > >
> > > > On Wed, Dec 7, 2016 at 3:22 PM, Anoop John <anoop.hbase@gmail.com>
> > > wrote:
> > > >
> > > > > I read the comments in the issue 16189..  This concern was raised
> > > > > there already..   This is what I replied there.
> > > > > "Ya we may need to make such a need. User on 1.x has to first
> upgrade
> > > > > (rolling upgrade supported) to latest version in 1.x and then to
> 2.0.
> > > > > This was discussed some other place also. I believe in the mail
> > thread
> > > > > where we discussed abt rolling upgrade support in 2.0"
> > > > >
> > > > > Not able to find the original mail thread which discussed abt
> rolling
> > > > > upgrade support in 2.0    Pls correct if took some thing wrong
> way..
> > > > > Ya I agree that this way of indirection might be inconvenience for
> > > > > users.
> > > > >
> > > > >
> > > > > -Anoop-
> > > > >
> > > > > On Wed, Dec 7, 2016 at 3:13 PM, Ted Yu <yuzhihong@gmail.com>
> wrote:
> > > > > > bq. write old name only and within code we will have to map
it
> with
> > > new
> > > > > > name.
> > > > > >
> > > > > > This is a viable approach.
> > > > > > Even if we document upgrading to 1.2.3+, some users may not
> perform
> > > > this
> > > > > > step before upgrading to 2.0
> > > > > >
> > > > > > On Tue, Dec 6, 2016 at 10:04 PM, Anoop John <
> anoop.hbase@gmail.com
> > >
> > > > > wrote:
> > > > > >
> > > > > >> Ya that bug raised by Enis was very valid..  So this means
when
> > > > > >> rolling upgrade happens, if some of the other RSs with version
> > > <1.2.3
> > > > > >> , which is not having this fix,  this issue might come up!
> > > > > >> How to address then?   Do we need to enforce a 1.2.3+ upgrade
> 1st
> > > and
> > > > > >> then only 2.0 rolling upgrade?
> > > > > >>
> > > > > >> Or else we will need to fix in 2.0..  We write the new
> Comparator
> > > > > >> class name in FFT (trunk code)   To fix, we might have to
write
> > old
> > > > > >> name only and within code we will have to map it with new
name.
> It
> > > > > >> will be ugly! It can be fixed only in 3.0 may be..  But
that can
> > > make
> > > > > >> the rolling upgrade story easier for users..   Just saying
the
> > > > > >> possibilities..
> > > > > >>
> > > > > >> Thanks Ted to bring it up again.
> > > > > >>
> > > > > >> -Anoop-
> > > > > >>
> > > > > >> On Wed, Dec 7, 2016 at 10:04 AM, ramkrishna vasudevan
> > > > > >> <ramkrishna.s.vasudevan@gmail.com> wrote:
> > > > > >> > I think when Enis reported the issue (
> > > > > >> > https://issues.apache.org/jira/browse/HBASE-16189),
in
> rolling
> > > > > upgrade
> > > > > >> > regions could move around. So a region created in 2.0
moved
> to a
> > > > > server
> > > > > >> > with 1.x.
> > > > > >> >
> > > > > >> > Regards
> > > > > >> > Ram
> > > > > >> >
> > > > > >> >
> > > > > >> > On Wed, Dec 7, 2016 at 1:27 AM, Stack <stack@duboce.net>
> wrote:
> > > > > >> >
> > > > > >> >> On Tue, Dec 6, 2016 at 10:19 AM, Ted Yu <yuzhihong@gmail.com
> >
> > > > wrote:
> > > > > >> >>
> > > > > >> >> > Looking at http://hbase.apache.org/book.
> > > html#executing.the.0.96.
> > > > > >> upgrade
> > > > > >> >> ,
> > > > > >> >> > there is step of running "bin/hbase upgrade
-check"
> > > > > >> >> >
> > > > > >> >> > How about adding a sample hfile which contains
> > > > > >> >> > CellComparator$MetaCellComparator
> > > > > >> >> > so that the upgrade check can read and verify
?
> > > > > >> >> >
> > > > > >> >> >
> > > > > >> >>
> > > > > >> >> We don't support downgrade. Never have.
> > > > > >> >> St.Ack
> > > > > >> >>
> > > > > >> >>
> > > > > >> >>
> > > > > >> >> > On Tue, Dec 6, 2016 at 8:50 AM, Ted Yu <
> yuzhihong@gmail.com>
> > > > > wrote:
> > > > > >> >> >
> > > > > >> >> > > The build I used yesterday didn't include
HBASE-16189
> > > > > >> >> > > <https://issues.apache.org/jira/browse/HBASE-16189>
> > > > > >> >> > >
> > > > > >> >> > > Once it is included, the cluster can
be downgraded fine.
> > > > > >> >> > >
> > > > > >> >> > > I wonder how users would know that their
existing
> > deployment
> > > > has
> > > > > >> >> > > HBASE-16189 <https://issues.apache.org/
> > > jira/browse/HBASE-16189
> > > > >
> > > > > >> before
> > > > > >> >> > > upgrading to 2.0 release.
> > > > > >> >> > >
> > > > > >> >> > > On Tue, Dec 6, 2016 at 2:29 AM, ramkrishna
vasudevan <
> > > > > >> >> > > ramkrishna.s.vasudevan@gmail.com>
wrote:
> > > > > >> >> > >
> > > > > >> >> > >> @Ted
> > > > > >> >> > >> Does your version have this fix
> > > > > >> >> > >> https://issues.apache.org/jira/browse/HBASE-16189
> > > > > >> >> > >>
> > > > > >> >> > >> Regards
> > > > > >> >> > >> Ram
> > > > > >> >> > >>
> > > > > >> >> > >> On Tue, Dec 6, 2016 at 3:56 PM, Ted
Yu <
> > yuzhihong@gmail.com
> > > >
> > > > > >> wrote:
> > > > > >> >> > >>
> > > > > >> >> > >> > Is the assumption that hbase:meta
would not split ?
> > > > > >> >> > >> >
> > > > > >> >> > >> > In other thread, Francis Liu
was proposing supporting
> > > > > splittable
> > > > > >> >> > >> > hbase:meta in 2.0 release.
> > > > > >> >> > >> >
> > > > > >> >> > >> > Cheers
> > > > > >> >> > >> >
> > > > > >> >> > >> > > On Dec 6, 2016, at 2:20
AM, 张铎 <
> palomino219@gmail.com
> > >
> > > > > wrote:
> > > > > >> >> > >> > >
> > > > > >> >> > >> > > Could this be solved by
hosting meta only on master?
> > > > > >> >> > >> > >
> > > > > >> >> > >> > > BTW, MetaCellComparator
is introduced in
> HBASE-10800.
> > > > > >> >> > >> > >
> > > > > >> >> > >> > > Thanks.
> > > > > >> >> > >> > >
> > > > > >> >> > >> > > 2016-12-06 17:44 GMT+08:00
Ted Yu <
> > yuzhihong@gmail.com
> > > >:
> > > > > >> >> > >> > >
> > > > > >> >> > >> > >> Hi,
> > > > > >> >> > >> > >> When I restarted a
cluster with 1.1 , I found that
> > > > > hbase:meta
> > > > > >> >> > region
> > > > > >> >> > >> > >> (written to by the
previously deployed 2.0)
> couldn't
> > be
> > > > > >> opened:
> > > > > >> >> > >> > >>
> > > > > >> >> > >> > >> Caused by: java.io.IOException:
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.io.
> > > hfile.CorruptHFileException:
> > > > > >> Problem
> > > > > >> >> > >> reading
> > > > > >> >> > >> > >> HFile Trailer from
file hdfs://
> yz1.xx.com:8020/apps/
> > > > > >> >> > >> hbase/data/data/
> > > > > >> >> > >> > >> hbase/meta/1588230740/info/
> > > > 599fc8a37311414e876803312009a9
> > > > > 86
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.regionserver.HStore.
> > > > > openStoreFiles(
> > > > > >> >> > >> HStore.java:
> > > > > >> >> > >> > >> 579)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.regionserver.HStore.
> > > > > loadStoreFiles(
> > > > > >> >> > >> HStore.java:
> > > > > >> >> > >> > >> 534)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.
> regionserver.HStore.<init>(
> > > > > >> >> > HStore.java:275)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.regionserver.HRegion.
> > > > > instantiateHSto
> > > > > >> >> > >> re(HRegion.
> > > > > >> >> > >> > >> java:5150)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.
> regionserver.HRegion$1.call(
> > > > > HRegion.
> > > > > >> >> > >> java:912)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.
> regionserver.HRegion$1.call(
> > > > > HRegion.
> > > > > >> >> > >> java:909)
> > > > > >> >> > >> > >>        at java.util.concurrent.
> > > > FutureTask.run(FutureTask.
> > > > > >> >> java:266)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> java.util.concurrent.Executors$RunnableAdapter.
> call(
> > > > > >> >> > >> Executors.java:511)
> > > > > >> >> > >> > >>        at java.util.concurrent.
> > > > FutureTask.run(FutureTask.
> > > > > >> >> java:266)
> > > > > >> >> > >> > >>        ... 3 more
> > > > > >> >> > >> > >> Caused by: org.apache.hadoop.hbase.io.
> > > > > >> >> hfile.CorruptHFileException:
> > > > > >> >> > >> > Problem
> > > > > >> >> > >> > >> reading HFile Trailer
from file hdfs://
> > > > > >> >> > >> > >> yz1.xx.com:8020/apps/hbase/data/data/hbase/
> > > > > meta/1588230740/
> > > > > >> >> > >> > >> info/599fc8a37311414e876803312009a986
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.io.
> > > > hfile.HFile.pickReaderVersion(
> > > > > >> >> > >> > HFile.java:483)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.io.
> hfile.HFile.createReader(
> > > > > >> >> > HFile.java:511)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.
> > regionserver.StoreFile$Reader.
> > > > > >> >> > >> > >> <init>(StoreFile.java:1128)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.
> regionserver.StoreFileInfo.
> > > > > >> >> > >> > >> open(StoreFileInfo.java:267)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.
> regionserver.StoreFile.open(
> > > > > StoreFil
> > > > > >> >> > >> e.java:409)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.regionserver.StoreFile.
> > > > > >> >> > >> > >> createReader(StoreFile.java:517)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.regionserver.HStore.
> > > > > createStoreFileA
> > > > > >> >> > >> ndReader(
> > > > > >> >> > >> > >> HStore.java:687)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.
> regionserver.HStore.access$
> > > 000(
> > > > > >> >> > >> HStore.java:130)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.
> regionserver.HStore$1.call(
> > > > > >> >> > HStore.java:554)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.
> regionserver.HStore$1.call(
> > > > > >> >> > HStore.java:551)
> > > > > >> >> > >> > >>        ... 6 more
> > > > > >> >> > >> > >> Caused by: java.io.IOException:
java.lang.
> > > > > >> >> ClassNotFoundException:
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.CellComparator$
> > > > MetaCellComparator
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.io.hfile.FixedFileTrailer.
> > > > > getCompara
> > > > > >> >> > >> torClass(
> > > > > >> >> > >> > >> FixedFileTrailer.java:581)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.io.hfile.FixedFileTrailer.
> > > > > >> >> > deserializeFromPB(
> > > > > >> >> > >> > >> FixedFileTrailer.java:300)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.io.hfile.FixedFileTrailer.
> > > > > >> >> > >> > >> deserialize(FixedFileTrailer.java:242)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.io.hfile.FixedFileTrailer.
> > > > > >> >> readFromStream(
> > > > > >> >> > >> > >> FixedFileTrailer.java:407)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.io.
> > > > hfile.HFile.pickReaderVersion(
> > > > > >> >> > >> > HFile.java:468)
> > > > > >> >> > >> > >>        ... 15 more
> > > > > >> >> > >> > >> Caused by: java.lang.ClassNotFoundException:
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.CellComparator$
> > > > MetaCellComparator
> > > > > >> >> > >> > >>        at java.net.URLClassLoader.
> > > > > >> findClass(URLClassLoader.java:
> > > > > >> >> > 381)
> > > > > >> >> > >> > >>        at java.lang.ClassLoader.
> > > > > loadClass(ClassLoader.java:
> > > > > >> 424)
> > > > > >> >> > >> > >>        at sun.misc.Launcher$
> > AppClassLoader.loadClass(
> > > > > >> >> > Launcher.java:
> > > > > >> >> > >> 331)
> > > > > >> >> > >> > >>        at java.lang.ClassLoader.
> > > > > loadClass(ClassLoader.java:
> > > > > >> 357)
> > > > > >> >> > >> > >>        at java.lang.Class.forName0(Native
Method)
> > > > > >> >> > >> > >>        at java.lang.Class.forName(Class.java:264)
> > > > > >> >> > >> > >>        at
> > > > > >> >> > >> > >> org.apache.hadoop.hbase.io.hfile.FixedFileTrailer.
> > > > > getCompara
> > > > > >> >> > >> torClass(
> > > > > >> >> > >> > >> FixedFileTrailer.java:579)
> > > > > >> >> > >> > >>
> > > > > >> >> > >> > >> When user does rolling
upgrade from 1.1 to 2.0, the
> > > above
> > > > > may
> > > > > >> >> cause
> > > > > >> >> > >> > problem
> > > > > >> >> > >> > >> if hbase:meta region
is updated by server running
> 2.0
> > > but
> > > > > >> later
> > > > > >> >> > >> > assigned to
> > > > > >> >> > >> > >> a region server which
still runs 1.1 (due to crash
> of
> > > the
> > > > > >> server
> > > > > >> >> > >> running
> > > > > >> >> > >> > >> 2.0, e.g.)
> > > > > >> >> > >> > >>
> > > > > >> >> > >> > >> I want to get community
feedback on the severity of
> > > this
> > > > > >> issue.
> > > > > >> >> > >> > >>
> > > > > >> >> > >> > >> Thanks
> > > > > >> >> > >> > >>
> > > > > >> >> > >> >
> > > > > >> >> > >>
> > > > > >> >> > >
> > > > > >> >> > >
> > > > > >> >> >
> > > > > >> >>
> > > > > >>
> > > > >
> > > >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message