From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Garzik Subject: Re: [Patch 2/2] tabled: rotate the input chunkserver Date: Fri, 23 Oct 2009 16:40:32 -0400 Message-ID: <4AE214C0.2050901@garzik.org> References: <20091022232602.31374d6d@redhat.com> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20091022232602.31374d6d@redhat.com> Sender: hail-devel-owner@vger.kernel.org List-ID: Content-Type: text/plain; charset="us-ascii"; format="flowed" To: Pete Zaitcev Cc: Project Hail List On 10/23/2009 01:26 AM, Pete Zaitcev wrote: > Another step to data redundancy: select the input chunkserver more > properly than just using the first one. Note: presently, if the > chunkserver dies while we're accessing it, we do not recover. > However, application can retry and hit an available one. > > Signed-off-by: Pete Zaitcev applied 1-2 As long as we (a) return an error documented in S3 spec as needing a retry, or (b) close the TCP connection before Content-Length is satisfied, the S3 client should retry, or at least, notice an error and not silently corrupt data. Have you verified that (b) occurs, if chunk->tabled->client data pipeline is interrupted after data transfer is under way? Jeff