From: Matt Helsley Subject: Re: [RFC][PATCH 4/4] Represent RPC Callers Date: Tue, 06 Jan 2009 15:05:19 -0800 Message-ID: <1231283119.14345.130.camel@localhost> References: <20090106011314.534653345@us.ibm.com> <20090106011315.229045829@us.ibm.com> <1231247062.7127.36.camel@heimdal.trondhjem.org> Mime-Version: 1.0 Content-Type: text/plain Cc: Linux Containers , "J. Bruce Fields" , Cedric Le Goater , Linux Kernel Mailing List , linux-nfs@vger.kernel.org, Chuck Lever , "Eric W. Biederman" , Linux Containers To: Trond Myklebust Return-path: Received: from e2.ny.us.ibm.com ([32.97.182.142]:38900 "EHLO e2.ny.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751630AbZAFXaj (ORCPT ); Tue, 6 Jan 2009 18:30:39 -0500 In-Reply-To: <1231247062.7127.36.camel-rJ7iovZKK19ZJLDQqaL3InhyD016LWXt@public.gmane.org> Sender: linux-nfs-owner@vger.kernel.org List-ID: On Tue, 2009-01-06 at 08:04 -0500, Trond Myklebust wrote: > On Mon, 2009-01-05 at 17:13 -0800, Matt Helsley wrote: > > plain text document attachment (move-rpc-client-nodename-cache.patch) > > Currently RPC needs to know the nodename (often the same as the hostname) which > > should be used for UNIX-style authentication and file-lock tracking. Because > > hostname can change between RPC calls and some sequences of RPC calls may > > require consistent names between calls RPC currently saves the nodename with > > the RPC client structure. > > > > This is doesn't always work because RPC clients may be discarded over the > > lifetime of a higher level service -- like those that compose NFS. Specifically > > this is known to happen during shutdown. > > > > Hence RPC should expect the nodename to be saved by the caller when sequences > > of RPC calls requiring consistent nodenames may be needed (e.g. NFS). To enable > > this we introduce an RPC caller structure that allows RPC to query the caller > > for this information. > > > > This patch is not complete but is meant to indicate the direction I'm planning > > on going. I'd like to know if there are any objections or if anyone sees a > > better way to handle this. > > You're planning on slowing down every RPC call in order to fix a problem > on client shutdown? Why? I figured that the network latencies would be much larger than the amount of time it takes to make a function call which, in the cases I know of, would reduce to a strcpy(). Cheers, -Matt Helsley