From: Joao Eduardo Luis <joao.luis@inktank.com>
To: Roman Alekseev <rs.alekseev@gmail.com>
Cc: Wido den Hollander <wido@widodh.nl>, ceph-devel@vger.kernel.org
Subject: Re: Monitor issue
Date: Tue, 30 Oct 2012 10:32:36 +0000 [thread overview]
Message-ID: <508FACC4.1080402@inktank.com> (raw)
In-Reply-To: <508F6E55.4020101@gmail.com>
On 10/30/2012 06:06 AM, Roman Alekseev wrote:
> On 29.10.2012 18:59, Wido den Hollander wrote:
>>
>>
>> On 10/29/2012 03:48 PM, Roman Alekseev wrote:
>>> Hello,
>>>
>>> I have 3 monitors on different nodes and when 'mon.a' was stopped whole
>>> cluster stopped work too.
>>> My conf: http://pastebin.com/hT3qEhUF
>>>
>>> Could someone explain how to fix such kind of failure?
>>
>> Could you explain a bit more about the setup?
>>
>> Which version are you running?
>>
>> What do you mean with failure? Is the ceph -s command still working?
>>
>> How sure are you that you didn't catch a bug that killed all three
>> monitors? Are those processes actually up and running?
>>
>> Did you check the logs of the monitors?
>>
>> Could you let us know?
>>
>> Thanks!
>>
>> Wido
>
> Hi Wido,
>
> I'm running ceph version 0.48.1argonaut.
> The "ceph -s" command doesn't work until I start that monitor again.
> Under failure I mean that ceph commands (such as ceph -s , -w, ceph mon
> dump etc) don't respond. I've re-added all three mons and found the
> following situations:
>
> Situation A:
> 1) mon.a is disabled:
> health HEALTH_WARN 1 mons down, quorum 1,2 b,c (cluster works)
> 2) mon.b is disabled:
> health HEALTH_WARN 1 mons down, quorum 0,1 a,c (cluster works)
> 3) mon.c is disabled:
> health HEALTH_WARN 1 mons down, quorum 0,2 a,b (cluster works)
>
> Situation B:
> If 2 mons are disabled all cluster stop working.
> So cluster works only when 2 monitors are running.
>
> Is it correct ?
>
In a nutshell, yes.
You need to have a majority of monitors up and running, and in the
quorum, for the cluster to work.
So, in 'A', you always have 2 out of the 3 existing monitors up; this
means a majority is up and quorum can be formed.
In 'B' however, you have only one monitor. Given there are 3 monitors in
the monmap, you don't have enough monitors to form a quorum (N+1/2, N
being the total number of monitors in your cluster, which is 3 in your
case).
Same rules apply if you had, say, 5 monitors: you'd need 3 up for the
cluster to work; if you only had 2, the monitors wouldn't be able to
form quorum.
And by the way, the cluster doesn't work because unless you have a
healthy monitor cluster, with formed quorum, the monitors that are up
and running will basically refuse to answer any kind of requests made to
them (aside from specific commands made through the admin socket, that
are targeted at one specific monitor).
Hope this helps.
-Joao
prev parent reply other threads:[~2012-10-30 10:33 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-10-29 14:48 Monitor issue Roman Alekseev
2012-10-29 14:59 ` Wido den Hollander
2012-10-30 6:06 ` Roman Alekseev
2012-10-30 10:32 ` Joao Eduardo Luis [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=508FACC4.1080402@inktank.com \
--to=joao.luis@inktank.com \
--cc=ceph-devel@vger.kernel.org \
--cc=rs.alekseev@gmail.com \
--cc=wido@widodh.nl \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.