I created a table with 100 splits. Then started ingesting via a
map-reduce job which read from AccumuloInputFormat and wrote to
AccumuloOutputFormat. There were over 250 mappers. However, at the
start of the job, there was just one tablet. As the job continued,
more tablets were created. Is there a way to pre-create the tablets so
I can have 100 tablets accepting entries from the start of the
map-reduce job?

You created a table with 100 splits but you only had one tablet? That
doesn't quite add up. Your output table only had one tablet?

Regardless, you should be able to use the addsplits command whenever you
want more split points/tablets.

On 12/31/12 1:17 PM, David Medinets wrote:
> I created a table with 100 splits. Then started ingesting via a
> map-reduce job which read from AccumuloInputFormat and wrote to
> AccumuloOutputFormat. There were over 250 mappers. However, at the
> start of the job, there was just one tablet. As the job continued,
> more tablets were created. Is there a way to pre-create the tablets so
> I can have 100 tablets accepting entries from the start of the
> map-reduce job?

I'm confused.
I created a table with 100 splits.

 OK

> However, at the
> start of the job, there was just one tablet.

?

> Is there a way to pre-create the tablets so
> I can have 100 tablets accepting entries from the start of the
> map-reduce job?
>

So, you created a table with splits, then you didn't have have multiple
tablets?

Did you read from a split tablet and output to a new table?

Or are all your tablets sitting on one tablet server?

The current load balancer isn't terribly aggressive about moving new splits
off to other servers.

One way to work around this problem is to create the table, with splits,
offline the table and bring it online again.  That will spread them around
pretty fast.

-Eric

My mistake... The command I used was "addsplits 1,2,3,4,5,6, etc..."
and, of course, I should have used a space between split values.
Accumulo is working fine. This was a user error.

On Mon, Dec 31, 2012 at 1:24 PM, Eric Newton <[EMAIL PROTECTED]> wrote:
> I'm confused.
>
>
>> I created a table with 100 splits.
>
>  OK
>>
>> However, at the
>> start of the job, there was just one tablet.
>
> ?
>>
>> Is there a way to pre-create the tablets so
>> I can have 100 tablets accepting entries from the start of the
>> map-reduce job?
>
>
> So, you created a table with splits, then you didn't have have multiple
> tablets?
>
> Did you read from a split tablet and output to a new table?
>
> Or are all your tablets sitting on one tablet server?
>
> The current load balancer isn't terribly aggressive about moving new splits
> off to other servers.
>
> One way to work around this problem is to create the table, with splits,
> offline the table and bring it online again.  That will spread them around
> pretty fast.
>
> -Eric
>