[Spread-users] Spread startup
Theo Schlossnagle
jesus at omniti.com
Tue Sep 26 12:07:39 EDT 2006
On Sep 26, 2006, at 11:51 AM, Ian Eiloart wrote:
>
>
> --On 26 September 2006 09:26:04 -0400 Theo Schlossnagle
> <jesus at omniti.com> wrote:
>
>> On Sep 26, 2006, at 6:14 AM, Ian Eiloart wrote:
>>> I have spread configured with this segment only:
>>> Spread_Segment 139.184.15.255:4803 {
>>> karpinski 139.184.14.85
>>> chip 139.184.14.86
>>> lynndie 139.184.14.87
>>> sivits 139.184.14.88
>>> }
>>>
>>> I'm using wackamole, configured to share about 20 IP addresses in
>>> virtual interfaces ("VIFs") between those four
>>> machines. Sometimes, when I use wackatrl -l, I get output like this:
>>>
>>> Owner: 127.0.0.1
>>> * lo0:139.184.14.101/32
>>> -> lo0:139.184.14.70/32
>>>
>>
>> Make sure that "ping `hostname`" on all your boxes pings the expected
>> publicly accessible IP address. (85-88). If not, set your
>> hostname and
>> make sure you have an /etc/hosts entry that points that hostname
>> to the
>> right IP. That, to me, looks like a wackamole issue more than a
>> Spread
>> issue.
>
> Well, hostname is always set correctly by the time I can get to log
> in, it's set in /etc/hostconfig. The IP address is set by DHCP, so
> it's not in /etc/hosts.
>
> Perhaps that's an alternative solution, but I think Spread would
> ideally check that it has local access to one of the IP addresses
> in the spread_segment definition before starting. Does it not do that?
Most server software is pretty particular about startup order /
dependencies. It sounds like you are trying to compensate for a lack
of this on the Mac. I was unaware that Mac launchd suffered form
this (I thought it was a rather thorough startup system). Spread and
wackamole require certain things in the system to be functioning
correctly before they start -- otherwise, the outcome will likely not
be the desired outcome.
I would not be using DHCP to run these things. The IP addresses are
hardcoded into the spread.conf, but not into your hosts'
configurations. This is an inconsistency.
// Theo Schlossnagle
// CTO -- http://www.omniti.com/~jesus/
// OmniTI Computer Consulting, Inc. -- http://www.omniti.com/
More information about the Spread-users
mailing list