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

Switch to Threaded View
Accumulo >> mail # user >> Merge Tablets [SEC=UNOFFICIAL]


Copy link to this message
-
Re: Merge Tablets [SEC=UNOFFICIAL]
Matt,

You interpreted the prompt correctly. That was added as a failsafe to
keep you from merging your entire table into one tablet.

The merge command doesn't take into account the split threshold, it
simply merges all tablets specified by the rows you provide (or entire
table) and merge them into one tablet. In other words, the configured
table.split.threshold you have defined won't work as you expect.

If you know how many tablets you want this table to have in the end (I
believe the current recommendation is an upper-bound of ~200 tablets per
tabletserver), you can use the current splits and some shell magic to
create a script that you can pass the shell that will merge series of
tablets together.

e.g.

merge -t table -b 1 -e 3
merge -t table -b 4 -e 6
merge -t table -b 7 -e 9

etc.

- Josh

On 09/30/2013 11:17 PM, Dickson, Matt MR wrote:
>
> *UNOFFICIAL*
>
> I have a table with 125K tablets and would like to merge the smaller
> tablets together.  The documentation states;
>
> Merging away small tablets works on one section at a time. If your
> table contains many sections of small split points, or you are
> attempting to change the split size of the entire table, it will be
> faster to set the split point and merge the entire table:
>
> Based on this I went to run *'merge -t /tablename/'* on a table with
> table.split.threshold set to 1G and was prompted to confirm if I
> wanted to merge the entire table into a single tablet.
>
> I had thought Accumulo would find small tablets and merge these upto a
> size of 1G, rather than attempt to create a single tablet.  Am Ir
> using the wrong command for this or misinterpreting the prompt?
>
> Thanks in advance,
>
> Matt
>