Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Plain View
Accumulo, mail # dev - Given a tablet id, is there a function to split?


+
David Medinets 2013-01-25, 14:42
+
John Vines 2013-01-25, 15:01
+
Keith Turner 2013-01-25, 16:47
+
David Medinets 2013-01-25, 17:08
Copy link to this message
-
Re: Given a tablet id, is there a function to split?
Keith Turner 2013-01-25, 17:14
if you lower the tables split threshhold, it should cause that function to run

On Fri, Jan 25, 2013 at 12:08 PM, David Medinets
<[EMAIL PROTECTED]> wrote:
> The issue is that I know the tablet to split, but not the split value.
> I think I need something like the findSplitRow method In Tablet.java.
> Is that functionality exposed to the client?
>
> On Fri, Jan 25, 2013 at 11:47 AM, Keith Turner <[EMAIL PROTECTED]> wrote:
>> On Fri, Jan 25, 2013 at 10:01 AM, John Vines <[EMAIL PROTECTED]> wrote:
>>> What about the adsplits command in the client API?
>>
>> Also can use tableOperations().tableIdMap() to convert the table id to
>> a table name.  Then use the table name to call addSplits()
>>
>>>
>>> Sent from my phone, please pardon the typos and brevity.
>>> On Jan 25, 2013 9:43 AM, "David Medinets" <[EMAIL PROTECTED]> wrote:
>>>
>>>> I am seeing some tablets with say 2 million records and some tablets
>>>> with 10 million records. I'd like to split the larger tablets so my
>>>> mappers all take approximately the same time to run. I'm Ok with
>>>> simply cutting the tablet in half. At least for now. I know how to
>>>> identify the tablets that I want to split. Is there an existing method
>>>> that I can call. I am leery of simply grabbing some code from the
>>>> middle of Accumulo and experimenting because I'll probably miss some
>>>> nuance causing the system to crash unexpectedly at some future time.
>>>>