From mboxrd@z Thu Jan 1 00:00:00 1970 From: Yang Honggang Subject: RGW placement multi data pool support Date: Fri, 18 Mar 2016 19:07:09 +0800 Message-ID: <56EBE15D.6020906@xtaotech.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Received: from m199-177.yeah.net ([123.58.177.199]:46909 "EHLO m199-177.yeah.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751042AbcCRLHO (ORCPT ); Fri, 18 Mar 2016 07:07:14 -0400 Sender: ceph-devel-owner@vger.kernel.org List-ID: To: ceph-devel Cc: Javen Wu Hello, In the current RGW implementation, only one =E2=80=98data pool=E2=80=99= is allowed in the placement target. So the size of a bucket is limited by the backend poo= l size. We can add osds into the backend pool, but it would trigger PG da= ta rebalance. Also, we might have to do PG split when the OSD count is clo= se to the pre-defined PG number, however, it would trigger data rebalance = as well. We try to avoid data rebalance as many as we can by adding multip= le pools into one placement target. Then a scheduler between pools of a placement target is needed. I have a immature proposal would like to discuss with you. We can exten= d placement's data pool to a pool list. When a new object is created thro= ugh RGW, we can choose one data pool from placement's data pool list. In or= der to do that, the pool ID need to be added into the bucket index for each object. How do you guys think? thx, Joseph -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" i= n the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html