hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dima Spivak <dspi...@cloudera.com>
Subject Re: Using HBase in standalone mode in production
Date Tue, 08 Jul 2014 00:21:24 GMT
Does MySQL's BLOB or VARBINARY satisfy your use case?

As for converting a pseudo-distributed cluster to a distributed one, unless
I'm mistaken, you should have no problem doing so. HDFS is quite good with
scaling, whether it's from 10 machines to 20 or 1 to 10 and I don't know of
any reason that HBase would cause any problems in this regard.

-Dima


On Mon, Jul 7, 2014 at 5:09 PM, Arun Allamsetty <arun.allamsetty@gmail.com>
wrote:

> I understand. But for example, my use case is where even if I don't have a
> lot of data, what if I would rather store serialized objects. For this
> traditional RDBMS are not suitable. If I can forego the fail safe
> capabilities, then what is a good choice (if not HBase).
>
> Also, on a different note, if I have a HBase installation in pseudo
> distributed mode, then can I convert it into a distributed setup by adding
> more machines without any loss in data?
>
> Thanks,
> Arun
> On Jul 7, 2014 6:02 PM, "Dima Spivak" <dspivak@cloudera.com> wrote:
>
> > In general, production systems run in distributed mode because they
> > leverage HBase's scalability and reliability; HBase really only shows its
> > worth when it's charged with managing terabytes of data on a
> fault-tolerant
> > file system like HDFS. You lose both of these when you run in standalone
> > mode, so I'd be a bit worried about using such a setup for any production
> > use.
> >
> > -Dima
> >
> >
> > On Mon, Jul 7, 2014 at 4:25 PM, Arun Allamsetty <
> arun.allamsetty@gmail.com
> > >
> > wrote:
> >
> > > Hi Ted,
> > >
> > > I have. So the book says there are two types of distributed modes. One
> is
> > > pseudo distributed, which is used when we want to test HBase's
> > distributed
> > > capabilities using a single machine. As far as I understood, this is
> just
> > > to verify the use cases and the requirements. Then we have the fully
> > > distributed mode in which HBase can be installed over multiple
> machines.
> > >
> > > I understand both the scenarios. But what if my application is not
> large
> > > enough to leverage the distributed mode and the pseudo distributed mode
> > is
> > > pretty much for a PoC. Since the pseudo distributed mode won't be able
> to
> > > provide any fault tolerance, can one use the standalone mode in
> > production.
> > >
> > > I hope my question is clear even if it does not make much sense.
> > >
> > > Thanks,
> > > Arun
> > > On Jul 7, 2014 5:17 PM, "Ted Yu" <yuzhihong@gmail.com> wrote:
> > >
> > > > Have you read http://hbase.apache.org/book.html#standalone_dist ?
> > > >
> > > > Cheers
> > > >
> > > >
> > > > On Mon, Jul 7, 2014 at 3:55 PM, Arun Allamsetty <
> > > arun.allamsetty@gmail.com
> > > > >
> > > > wrote:
> > > >
> > > > > Hi all,
> > > > >
> > > > > So this question might be stupid, retarded even, but it has been
> > > bugging
> > > > me
> > > > > for a while and I cannot think of a better place to ask this. I am
> > > really
> > > > > impressed with the way HBase works (as a key-value store). Since
it
> > > > stores
> > > > > everything as a byte array, I find it really convenient to store
> > > > serialized
> > > > > objects. Also, I understand that HBase is supposed to be used when
> > you
> > > > have
> > > > > too much data to be handled by a single machine, so we can scale
> our
> > > > > application by running it in distributed mode.
> > > > >
> > > > > But what if I want to use it because its HashMap kind of
> capabilities
> > > > with
> > > > > an added feature to track versions. Is it recommended that I use
it
> > > for a
> > > > > small application (in standalone mode) with maybe 100K users and
> > > storage
> > > > > needs which probably won't exceed 100G.
> > > > >
> > > > > I know it is never recommended to be used as a transactional
> database
> > > (I
> > > > > have read that in a million places) but I would like to know more
> > about
> > > > it.
> > > > >
> > > > > Thanks,
> > > > > Arun
> > > > >
> > > >
> > >
> >
>

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