Practical limit to number of keyspaces?

[Sourabh Agrawal <ii...@gmail.com>]

Hi,

We are using 2.0.x in production env. Each node has 8G heap. Is there a
practical limit to how many keyspaces I can have? Please note that many
keyspaces remain mostly dormant while others have very high level of
activity.

-- 
Sourabh Agrawal
Bangalore
+91 9945657973

Re: Practical limit to number of keyspaces?

[Sourabh Agrawal <ii...@gmail.com>]

Hi Jack,

I read about arena allocators. So, does allocation happen only for
memtables that are active(have atleast 1 update). What happens if the CF is
inactive and does not have any updates in the memtable? Will cassandra
still allocate 1MB for the CF?


On Fri, Jul 11, 2014 at 5:50 PM, Jack Krupansky <ja...@basetechnology.com>
wrote:

>   Technically, key spaces themselves are relatively cheap, so there is no
> limit there per se. But... you probably want to put tables in them, which
> are not so cheap. Generally, the guidance is for no more than “low
> hundreds” of tables, regardless of whether you have one table per key space
> or all in one key space.
>
> If you want to go beyond “low hundreds” of tables, tread carefully. The
> official line is that “Although having more than dozens or hundreds of
> tables defined is almost certainly a Bad Idea (just as it is a design smell
> in a relational database), it's relatively straightforward to allow
> disabling the SlabAllocator.” Emphasis on “almost certainly a Bad Idea.”
>
> See:
> https://issues.apache.org/jira/browse/CASSANDRA-5935
> “Allow disabling slab allocation”
>
> -- Jack Krupansky
>
>  *From:* Sourabh Agrawal <ii...@gmail.com>
> *Sent:* Thursday, July 10, 2014 11:22 PM
> *To:* user@cassandra.apache.org
> *Subject:* Practical limit to number of keyspaces?
>
>  Hi,
>
> We are using 2.0.x in production env. Each node has 8G heap. Is there a
> practical limit to how many keyspaces I can have? Please note that many
> keyspaces remain mostly dormant while others have very high level of
> activity.
>
> --
> Sourabh Agrawal
> Bangalore
> +91 9945657973
>



-- 
Sourabh Agrawal
Bangalore
+91 9945657973

Re: Practical limit to number of keyspaces?

[Jack Krupansky <ja...@basetechnology.com>]

Technically, key spaces themselves are relatively cheap, so there is no limit there per se. But... you probably want to put tables in them, which are not so cheap. Generally, the guidance is for no more than “low hundreds” of tables, regardless of whether you have one table per key space or all in one key space.

If you want to go beyond “low hundreds” of tables, tread carefully. The official line is that “Although having more than dozens or hundreds of tables defined is almost certainly a Bad Idea (just as it is a design smell in a relational database), it's relatively straightforward to allow disabling the SlabAllocator.” Emphasis on “almost certainly a Bad Idea.”

See:
https://issues.apache.org/jira/browse/CASSANDRA-5935
“Allow disabling slab allocation”

-- Jack Krupansky

From: Sourabh Agrawal 
Sent: Thursday, July 10, 2014 11:22 PM
To: user@cassandra.apache.org 
Subject: Practical limit to number of keyspaces?

Hi, 

We are using 2.0.x in production env. Each node has 8G heap. Is there a practical limit to how many keyspaces I can have? Please note that many keyspaces remain mostly dormant while others have very high level of activity.


-- 

Sourabh Agrawal 
Bangalore
+91 9945657973