incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Edward Capriolo <edlinuxg...@gmail.com>
Subject Re: Multiple keyspaces vs Multiple CFs
Date Fri, 09 Nov 2012 02:40:02 GMT
It is not as bad with hector, but still each Keyspace object is
another socket open to Cassandra. If you have 500 webservers and 10
keyspaces. Instead of having 5000 connections you now have 5000.

On Thu, Nov 8, 2012 at 6:35 PM, sankalp kohli <kohlisankalp@gmail.com> wrote:
> I think this code is from the thrift part. I use hector. In hector, I can
> create multiple keyspace objects for each keyspace and use them when I want
> to talk to that keyspace. Why will it need to do a round trip to the server
> for each switch.
>
>
> On Thu, Nov 8, 2012 at 3:28 PM, Edward Capriolo <edlinuxguru@gmail.com>
> wrote:
>>
>> In the old days the API looked like this.
>>
>>   client.insert("Keyspace1",
>>              key_user_id,
>>                new ColumnPath("Standard1", null,
>> "name".getBytes("UTF-8")),
>>                   "Chris Goffinet".getBytes("UTF-8"),
>>                timestamp,
>>                    ConsistencyLevel.ONE);
>>
>> but now it works like this
>>
>> /----pay attention to this below -------------/
>> client.set_keyspace("keyspace1");
>> /----pay attention to this above -------------/
>>   client.insert(
>>          key_user_id,
>>                      new ColumnPath("Standard1", null,
>> "name".getBytes("UTF-8")),
>>                       "Chris Goffinet".getBytes("UTF-8"),
>>                       timestamp,
>>                       ConsistencyLevel.ONE);
>>
>> So each time you switch keyspaces you make a network round trip.
>>
>> On Thu, Nov 8, 2012 at 6:17 PM, sankalp kohli <kohlisankalp@gmail.com>
>> wrote:
>> > I am a bit confused. One connection pool I know is the one which
>> > MessageService has to other nodes. Then there will be incoming
>> > connections
>> > via thrift from clients. How are they affected by multiple keyspaces?
>> >
>> >
>> > On Thu, Nov 8, 2012 at 3:14 PM, Edward Capriolo <edlinuxguru@gmail.com>
>> > wrote:
>> >>
>> >> Any connection pool. Imagine if you have 10 column families in 10
>> >> keyspaces. You pull a connection off the pool and the odds are 1 in 10
>> >> of it being connected to the keyspace you want. So 9 out of 10 times
>> >> you have to have a network round trip just to change the keyspace, or
>> >> you have to build a keyspace aware connection pool.
>> >> Edward
>> >>
>> >> On Thu, Nov 8, 2012 at 5:36 PM, sankalp kohli <kohlisankalp@gmail.com>
>> >> wrote:
>> >> > Which connection pool are you talking about?
>> >> >
>> >> >
>> >> > On Thu, Nov 8, 2012 at 2:19 PM, Edward Capriolo
>> >> > <edlinuxguru@gmail.com>
>> >> > wrote:
>> >> >>
>> >> >> it is better to have one keyspace unless you need to replicate
the
>> >> >> keyspaces differently. The main reason for this is that changing
>> >> >> keyspaces requires an RPC operation. Having 10 keyspaces would
mean
>> >> >> having 10 connection pools.
>> >> >>
>> >> >> On Thu, Nov 8, 2012 at 4:59 PM, sankalp kohli
>> >> >> <kohlisankalp@gmail.com>
>> >> >> wrote:
>> >> >> > Is it better to have 10 Keyspaces with 10 CF in each keyspace.
or
>> >> >> > 100
>> >> >> > keyspaces with 1 CF each.
>> >> >> > I am talking in terms of memory footprint.
>> >> >> > Also I would be interested to know how much better one is
over
>> >> >> > other.
>> >> >> >
>> >> >> > Thanks,
>> >> >> > Sankalp
>> >> >
>> >> >
>> >
>> >
>
>

Mime
View raw message