-Re: Migrating a table from HBase older version to newer version.
Michael Webster 2013-09-26, 03:28
I have done a similar migration between clusters using the bulkload
utility. Look at http://hbase.apache.org/book/arch.bulk.load.html for
setting up a mapreduce job to write hfiles. Such a job will write the
hfiles according to existing region boundaries on your destination table.
It would be best to minimize writes to the new cluster as much as
possible, otherwise the region boundaries may change during the bulkload,
it will still work, but the bulkload tool has to recalculate where to put
the hfiles, which isn't that efficient. You may have trouble with the
incompatibilities in 0.90 vs 0.92 though, I haven't tried a cross version
migration using a MR job, just moving between clusters.
You could also just copy the hfiles to your new cluster (using hftp to
avoid RPC version mistmatches) in a temp directory and use bulkload to copy
them into the hbase tables. This will likely be slower than using a MR job
as the region boundaries between the old cluster and the new one probably
won't line up.
On Wed, Sep 25, 2013 at 3:31 PM, hbaseuser hbaseuser <[EMAIL PROTECTED]
> I currently have two HBase clusters (0.90) & (0.92).
> I'm moving data from 0.90 to 0.92 and I'm wondering if I can copy the files
> from /hbase folder in one cluster to the other cluster and do a put in
> I have tried the export / import tools and it doesn't work.
> I was able to successfully do the migration of /hbase folders between the
> two cluster and bring up Hbase and the files from the older version got
> converted to the newer version but I cannot do that here since the
> destination cluster already has data.
*Michael Webster*, Software Engineer
Marketing solutions for commerce. Learn more.<http://www.bronto.com/platform>