From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Samudrala, Sridhar" Subject: Re: [PATCH net] failover: eliminate callback hell Date: Tue, 5 Jun 2018 16:52:22 -0700 Message-ID: References: <20180605034231.31610-1-sthemmin@microsoft.com> <20180605211927-mutt-send-email-mst@kernel.org> <20180605115305.502a7ebb@xeon-e3> <20180605221049-mutt-send-email-mst@kernel.org> <20180605145222.477e5ae8@xeon-e3> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 7bit Cc: kys@microsoft.com, haiyangz@microsoft.com, davem@davemloft.net, netdev@vger.kernel.org, Stephen Hemminger To: Stephen Hemminger , "Michael S. Tsirkin" Return-path: Received: from mga09.intel.com ([134.134.136.24]:22316 "EHLO mga09.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932185AbeFEXwZ (ORCPT ); Tue, 5 Jun 2018 19:52:25 -0400 In-Reply-To: <20180605145222.477e5ae8@xeon-e3> Content-Language: en-US Sender: netdev-owner@vger.kernel.org List-ID: On 6/5/2018 2:52 PM, Stephen Hemminger wrote: > On Tue, 5 Jun 2018 22:38:43 +0300 > "Michael S. Tsirkin" wrote: > >>> See: >>> https://patchwork.ozlabs.org/patch/851711/ >> Let me try to summarize that: >> >> You wanted to speed up the delayed link up. You had an idea to >> additionally take link up when userspace renames the interface (standby >> one which is also the failover for netvsc). >> >> But userspace might not do any renames, in which case there will >> still be the delay, and so this never got applied. >> >> Is this a good summary? >> >> Davem said delay should go away completely as it's not robust, and I >> think I agree. So I don't think we should make all failover users use >> delay. IIUC failover kept a delay option especially for netvsc to >> minimize the surprise factor. Hopefully we can come up with >> something more robust and drop that option completely. > The timeout was the original solution to how to complete setup after > userspace has had a chance to rename the device. Unfortunately, the whole network > device initialization (cooperation with udev and userspace) is a a mess because > there is no well defined specification, and there are multiple ways userspace > does this in old and new distributions. The timeout has its own issues > (how long, handling errors during that window, what if userspace modifies other > device state); and open to finding a better solution. > > My point was that if name change can not be relied on (or used) by netvsc, > then we can't allow it for net_failover either. I think the push back was with the usage of the delay, not bringing up the primary/standby device in the name change event handler. Can't netvsc use this mechanism instead of depending on the delay?