Mailing-List: contact user-help@hbase.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@hbase.apache.org
Received-SPF: pass (athena.apache.org: local policy)
From: Sagar Naik <snaik@splunk.com>
To: "user@hbase.apache.org" <user@hbase.apache.org>, Dhaval Shah
	<prince_mithibai@yahoo.co.in>
Subject: Re: HBase Design : Column name v/s Version
Thread-Topic: HBase Design : Column name v/s Version
Thread-Index: AQHPGSmTPPQjbeBu1ECZatoW+9aDpZqUrG6A//9/i4CAAI1XAP//fZyA
Date: Fri, 24 Jan 2014 18:46:36 +0000
Message-ID: <CF07F355.4D03%snaik@splunk.com>
References: <CF07E28E.4CAD%snaik@splunk.com>
 <1390585637.40880.YahooMailNeo@web190103.mail.sg3.yahoo.com>
 <CF07EB23.4CB6%snaik@splunk.com>
 <1390588404.70450.YahooMailNeo@web190102.mail.sg3.yahoo.com>
In-Reply-To: <1390588404.70450.YahooMailNeo@web190102.mail.sg3.yahoo.com>
Accept-Language: en-US
Content-Language: en-US
Content-Type: text/plain; charset="us-ascii"
Content-ID: <042B90A655350D4ABD3E669055ECB0AD@internal>
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0

Thanks for clarifying,

I will be using custom version numbers (auto incrementing on the client
side) and not timestamps.
Two clients do not update the same row


-Sagar

On 1/24/14 10:33 AM, "Dhaval Shah" <prince_mithibai@yahoo.co.in> wrote:

>I am talking about schema 2. Schema 1 would definitely work. Schema 2 can
>have the version collisions if you decide to use timestamps as versions
>=20
>Regards,
>
>Dhaval
>
>
>----- Original Message -----
>From: Sagar Naik <snaik@splunk.com>
>To: "user@hbase.apache.org" <user@hbase.apache.org>; Dhaval Shah
><prince_mithibai@yahoo.co.in>
>Cc:=20
>Sent: Friday, 24 January 2014 1:07 PM
>Subject: Re: HBase Design : Column name v/s Version
>
>I am not sure I understand you correctly.
>I assume you are talking abt schema 1.
>In this case I m appending the version number to the column name.
>
>The column_names are different (data_1/data_2) for value_1 and value_2
>respectively.
>
>
>-Sagar
>
>
>On 1/24/14 9:47 AM, "Dhaval Shah" <prince_mithibai@yahoo.co.in> wrote:
>
>>Versions in HBase are timestamps by default. If you intend to continue
>>using the timestamps, what will happen when someone writes value_1 and
>>value_2 at the exact same time?
>>=20
>>Regards,
>>
>>Dhaval
>>
>>
>>----- Original Message -----
>>From: Sagar Naik <snaik@splunk.com>
>>To: "user@hbase.apache.org" <user@hbase.apache.org>
>>Cc:=20
>>Sent: Friday, 24 January 2014 12:27 PM
>>Subject: HBase Design : Column name v/s Version
>>
>>Hi,
>>
>>I have a choice to maintain to data either in column values or as
>>versioned data.
>>This data is not a versioned copy per se.
>>
>>The access pattern on this get all the data every time
>>
>>So the schema choices are :
>>Schema 1:
>>1. column_name/qualifier =3D> data_1. column_value =3D> value_1
>>1.a. column_name/qualifier =3D> data_2. column_value =3D> value_2,value_2=
.a
>>
>>1.b. column_name/qualifier =3D> data_3. column_value =3D> value_3
>>
>>To get all the values for "data", I will have to use ColumnPrefixFilter
>>with prefix set "data"
>>
>>Schema 2:
>>2. column_name/qualifier =3D> data. version=3D> 1, column_value =3D> valu=
e_1
>>
>>2.a. column_name/qualifier =3D> data. version=3D> 2, column_value =3D>
>>value_2,value_2.a
>>
>>2.b. column_name/qualifier =3D> data. version=3D> 3, column_value =3D> va=
lue_3
>>To get all the values for "data" , I will do a simple get operation to
>>get
>>all the versions.
>>
>>Number of versions can go from: 10 to 100K
>>
>>Get operation perf should beat the Filter perf.
>>Comparing 100K values will be costly as the # versions increase.
>>
>>I would like to know if there are drawbacks in going the version route.
>>
>>
>>
>>
>>-Sagar
>>
>