Changing versions from 3 to 1 in production - 0.90.3

[Vaibhav Puranik <vp...@gmail.com>]

Hi all,

We are planning to change versions of all of our tables (3 tables with 3 to
4 column families each) from 3 (default) to 1.
Here is why we want to change it:
1) We have never used versions and we don't really need them
2) It should reduce amount of data Hbase loads in memory (StoreFiles or
Stores) by 2/3rd.

Here are my questions:
1) Is this correct? Will it really reduce the amount of data it loads in
memory?

2) How do I do this in production? I know that HBase shell supports alter
statement. Do you recommend using that?
If I execute an alter statement, will it immediately start lot of activity
there by degrading our real time performance?

3) Do we need to do anything else before or after we run alter command?

Regards,
Vaibhav

Re: Changing versions from 3 to 1 in production - 0.90.3

[Vaibhav Puranik <vp...@gmail.com>]

Thanks Joey.

We have many counters with long keys. We update them frequently.
That is why I think it should help us.

Thanks again,
Vaibhav,
GumGum

On Sat, Jul 9, 2011 at 6:12 PM, Joey Echeverria <jo...@cloudera.com> wrote:

> > Here are my questions:
> > 1) Is this correct? Will it really reduce the amount of data it loads in
> > memory?
>
> It will only reduce memory usage if you currently have multiple
> versions of the same cell.
>
> > 2) How do I do this in production? I know that HBase shell supports alter
> > statement. Do you recommend using that?
> > If I execute an alter statement, will it immediately start lot of
> activity
> > there by degrading our real time performance?
>
> Yes, you should use the shell. You'll need to disable the table first.
> No activity should happen immediately. If you currently have multiple
> version of any cells, they will disappear during the next major
> compaction.
>
> > 3) Do we need to do anything else before or after we run alter command?
>
> You can run a full compaction if you want to make sure that additional
> versions are deleted right away.
>
> -Joey
>
> --
> Joseph Echeverria
> Cloudera, Inc.
> 443.305.9434
>

Re: Changing versions from 3 to 1 in production - 0.90.3

[Joey Echeverria <jo...@cloudera.com>]

> Here are my questions:
> 1) Is this correct? Will it really reduce the amount of data it loads in
> memory?

It will only reduce memory usage if you currently have multiple
versions of the same cell.

> 2) How do I do this in production? I know that HBase shell supports alter
> statement. Do you recommend using that?
> If I execute an alter statement, will it immediately start lot of activity
> there by degrading our real time performance?

Yes, you should use the shell. You'll need to disable the table first.
No activity should happen immediately. If you currently have multiple
version of any cells, they will disappear during the next major
compaction.

> 3) Do we need to do anything else before or after we run alter command?

You can run a full compaction if you want to make sure that additional
versions are deleted right away.

-Joey

-- 
Joseph Echeverria
Cloudera, Inc.
443.305.9434