From mboxrd@z Thu Jan 1 00:00:00 1970 From: Xiaopong Tran Subject: Re: ceph init script does not start Date: Mon, 16 Jul 2012 10:09:07 +0800 Message-ID: <500377C3.3030407@gmail.com> References: <50012AF1.4040300@gmail.com> <3046989.AgMSmANe5F@mranderson> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from mail-pb0-f46.google.com ([209.85.160.46]:45419 "EHLO mail-pb0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751234Ab2GPCKX (ORCPT ); Sun, 15 Jul 2012 22:10:23 -0400 Received: by pbbrp8 with SMTP id rp8so9231759pbb.19 for ; Sun, 15 Jul 2012 19:10:23 -0700 (PDT) In-Reply-To: Sender: ceph-devel-owner@vger.kernel.org List-ID: To: Sage Weil Cc: =?UTF-8?B?U3rDqWtlbHlpIFN6YWJvbGNz?= , ceph-devel@vger.kernel.org On 07/16/2012 07:06 AM, Sage Weil wrote: > On Sun, 15 Jul 2012, Sz?kelyi Szabolcs wrote: >> On 2012. July 14. 07:42:34 Sage Weil wrote: >>> On Sat, 14 Jul 2012, Xiaopong Tran wrote: >>>> I'm getting this funny issue. I had setup two test clusters, and >>>> mkcephfs and the ceph start up script worked just fine. We are >>>> now ready to go production, we have 6 nodes, with 10 disks >>>> each, and one osd per disk, with 3 mds and 3 mons. >>>> >>>> The script mkcephfs ran without problem, everything was created >>>> properly. See attached log file. However, when I run >>>> >>>> /etc/init.d/ceph start >>>> >>>> nothing happens, not even a line of message, not on concole, >>>> neither in system log. >>>> >>>> But can I manually start up each individual osd, mds, and mon. >>> >>> This is usually related to the 'host = ...' lines in ceph.conf. They need >>> to match the output of the `hostname` command in order for that daemon to >>> be automatically started or stopped. >> >> Just a humble remark here: actually the host= setting has to match the >> hostname *until the first dot*. If your hostname contains a dot, this will not >> work. This can be useful in a number of cases, like setting the hostname to >> indicate the group of a hierarchy level in the name so that the administrator >> can extactly know what he's working on after logging in (I mean hostnames like >> node.rack in a datacenter or node. when running a >> geographically distributed cluster). >> >> From this comes my request: do you think it's possible to change (I mean, in >> the repository) ceph_common.sh not to cut the output of `hostname` at the first >> dot? I'm running a cluster with hostnames like that, and now I have to edit >> that file after each upgrade. > > I think the trick here is keeping things compatible with both methods. > Maybe it should match any prefix of the host fqdn? e.g., for a host > a.b.c.d, any of host = a, host = a.b, host = a.b.c, or host = a.b.c.d will > match, but host = x.a will not. > It would be great if it can handle fqdn. I tried to use fqdn, along with an internal DNS, and the /etc/hostname and /etc/hosts file set, and it still does not start. host = a.b.c does not do, so far, only host = a works for me. Thanks Xiaopong