From mboxrd@z Thu Jan 1 00:00:00 1970 From: Joao Eduardo Luis Subject: Re: A Transformation of our Global Context Date: Sat, 11 Jun 2016 00:57:23 +0100 Message-ID: <575B53E3.90102@suse.de> References: <20160610185033.GA29233@ultraspiritum.eng.arb.redhat.com> <20160610195837.GB29233@ultraspiritum.eng.arb.redhat.com> <20160610201543.GC29233@ultraspiritum.eng.arb.redhat.com> <20160610204549.GB7952@ultraspiritum.eng.arb.redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from mx2.suse.de ([195.135.220.15]:56455 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750888AbcFJX50 (ORCPT ); Fri, 10 Jun 2016 19:57:26 -0400 In-Reply-To: Sender: ceph-devel-owner@vger.kernel.org List-ID: To: Sage Weil , "Adam C. Emerson" Cc: Gregory Farnum , The Sacred Order of the Squid Cybernetic On 06/10/2016 09:53 PM, Sage Weil wrote: > On Fri, 10 Jun 2016, Adam C. Emerson wrote: >> On 10/06/2016, Sage Weil wrote: >>> FWIW the time thing is a bit of a red herring. The only reason time takes >>> cct is for the configuration clock skew. That was added for debugging >>> purposes but has never (to my knowledge) been used. I'd be happy to rip >>> it out. >> >> I wasn't aware of that. I'd asked once if the clock skew debug option >> was really important (back with the original time patch) and was told >> it was. The person I was talking to might have thought I was talking >> about monitor clock skew or something. > > Ah, there is *one* teuthology test that sets this: > > $ git grep clock\ offset | cat > suites/rados/multimon/tasks/mon_clock_with_skews.yaml: clock offset: 10 > > I'd be quite happy to drop this test, or kludge around it some other way > (e.g., put a mon on a different host and change the system time), if it > saves us as much code churn as I suspect it will... i've been hacking a mon debug interface via the ceph tool to test features and do all sorts of nasty things (provided a given experimental feature is enabled and the user is really, really sure about it), and I think we could then use it to test clock skews just as well. It'd be a matter of telling a given monitor to just skew the response to the leader during the timecheck round. -Joao