You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hbase.apache.org by mac fang <ma...@gmail.com> on 2011/01/16 15:36:50 UTC
One question about HBase Client
Hi, guys,
one question want to get the awnser: i read the HBase Client code, seems the
logic to get a table:
1. Get -ROOT- RegionLoc from Zk,
2. Get .META. RegionLoc from -ROOT- Table.
3. Get Table RegionLoc from .META. Table.
So seems the -ROOT- is NOT necessary, we can put the .META. location
directly into -ROOT-, is it true? Thanks for any replies.
regards
macf
Re: One question about HBase Client
Posted by Stack <st...@duboce.net>.
Thanks for the input. You might want to add your thinking to
https://issues.apache.org/jira/browse/HBASE-3171. Thats one of the
issues where there is talk of moving away from the BT way of doing
-ROOT- and .META.
St.Ack
On Mon, Jan 17, 2011 at 6:21 AM, mac fang <ma...@gmail.com> wrote:
> Thanks, Stack, I read the Big Table paper again. It used the SSTable for
> both -ROOT- and .META. However in my mind, the -ROOT- table is very
> important and the whole system would be out-of-service if the SSTable of
> -ROOT- can NOT be opened. So if we can locate the location of .META.
> directly into Zookeeper. It would be more safer. Just one suggestion.
>
> regards
> macf
>
>
> On Mon, Jan 17, 2011 at 5:37 AM, Stack <st...@duboce.net> wrote:
>
>> Currently there can be only one region in the HBase .META. table.
>> ROOT is there for the day when .META. has many regions of its own;
>> i.e. big clusters. Thats the idea from the BigTable at least.
>>
>> That said, there's a few conversations that have gone on where some
>> propose doing as you suggest, doing away with -ROOT- altogether, and
>> if .META. splits, just keep its regions up in zookeeper (Some have
>> even gone so far as to propose putting .META. content up in
>> zookeeper).
>>
>> St.Ack
>>
>> On Sun, Jan 16, 2011 at 6:36 AM, mac fang <ma...@gmail.com> wrote:
>> > Hi, guys,
>> >
>> > one question want to get the awnser: i read the HBase Client code, seems
>> the
>> > logic to get a table:
>> >
>> > 1. Get -ROOT- RegionLoc from Zk,
>> > 2. Get .META. RegionLoc from -ROOT- Table.
>> > 3. Get Table RegionLoc from .META. Table.
>> >
>> > So seems the -ROOT- is NOT necessary, we can put the .META. location
>> > directly into -ROOT-, is it true? Thanks for any replies.
>> >
>> > regards
>> > macf
>> >
>>
>
Re: One question about HBase Client
Posted by mac fang <ma...@gmail.com>.
Thanks, Stack, I read the Big Table paper again. It used the SSTable for
both -ROOT- and .META. However in my mind, the -ROOT- table is very
important and the whole system would be out-of-service if the SSTable of
-ROOT- can NOT be opened. So if we can locate the location of .META.
directly into Zookeeper. It would be more safer. Just one suggestion.
regards
macf
On Mon, Jan 17, 2011 at 5:37 AM, Stack <st...@duboce.net> wrote:
> Currently there can be only one region in the HBase .META. table.
> ROOT is there for the day when .META. has many regions of its own;
> i.e. big clusters. Thats the idea from the BigTable at least.
>
> That said, there's a few conversations that have gone on where some
> propose doing as you suggest, doing away with -ROOT- altogether, and
> if .META. splits, just keep its regions up in zookeeper (Some have
> even gone so far as to propose putting .META. content up in
> zookeeper).
>
> St.Ack
>
> On Sun, Jan 16, 2011 at 6:36 AM, mac fang <ma...@gmail.com> wrote:
> > Hi, guys,
> >
> > one question want to get the awnser: i read the HBase Client code, seems
> the
> > logic to get a table:
> >
> > 1. Get -ROOT- RegionLoc from Zk,
> > 2. Get .META. RegionLoc from -ROOT- Table.
> > 3. Get Table RegionLoc from .META. Table.
> >
> > So seems the -ROOT- is NOT necessary, we can put the .META. location
> > directly into -ROOT-, is it true? Thanks for any replies.
> >
> > regards
> > macf
> >
>
Re: One question about HBase Client
Posted by Stack <st...@duboce.net>.
Currently there can be only one region in the HBase .META. table.
ROOT is there for the day when .META. has many regions of its own;
i.e. big clusters. Thats the idea from the BigTable at least.
That said, there's a few conversations that have gone on where some
propose doing as you suggest, doing away with -ROOT- altogether, and
if .META. splits, just keep its regions up in zookeeper (Some have
even gone so far as to propose putting .META. content up in
zookeeper).
St.Ack
On Sun, Jan 16, 2011 at 6:36 AM, mac fang <ma...@gmail.com> wrote:
> Hi, guys,
>
> one question want to get the awnser: i read the HBase Client code, seems the
> logic to get a table:
>
> 1. Get -ROOT- RegionLoc from Zk,
> 2. Get .META. RegionLoc from -ROOT- Table.
> 3. Get Table RegionLoc from .META. Table.
>
> So seems the -ROOT- is NOT necessary, we can put the .META. location
> directly into -ROOT-, is it true? Thanks for any replies.
>
> regards
> macf
>