You are viewing a plain text version of this content. The canonical link for it is here.

Posted to user@phoenix.apache.org by Alexander Litvak <al...@marketwired.com> on 2015/03/27 19:38:47 UTC

"CREATE TABLE" for existing large HBase tables

Hi,

I have large HBase tables (>10 billion rows). RowIDs are salted with byte prefix and the table splitted into 64…256 regions.

1. When I run CREATE TABLE “table” (…) SALT_BUCKETS=256, Phoenix adds “0:_0” key to each row that add huge update load to the HBase cluster and eventually HBase clients timeout.
The HBase cluster is tuned for other tasks and I don’t want to change it just to “create” Phoenix table.
I tried "CREATE VIEW” but it doesn’t support “SALTED_BUCKETS” option.

2. Running CREATE INDEX for existing large tables also overload HBase cluster.

Is there a way to run CREATE TABLE and CREATE INDEX as a mapreduce job on YARN?
Is 10bil tables fit to Phoenix use cases?

Any thoughts on this?

Thanks,
Alex