[Labs-l] Instance scheduler modified, could cause instance creation to fail

Ryan Lane rlane32 at gmail.com
Wed Feb 29 21:06:27 UTC 2012


I think I found the issue. The wikistream project had a security group
rule like this:

From: -1
To : 80
Protocol: tcp
CIDR: 0.0.0.0/0

-1 is only a viable port for icmp, and it really maps to some fake
iptables entry. It seems nova will let you provide -1 for tcp, then
the iptables command will fail when the instance is created, leaving
the instance in a pending state.

I removed the rule from the default security group, created a "web"
security group, and added ports 80 and 443 to it.

- Ryan

On Wed, Feb 29, 2012 at 11:47 AM, Ryan Lane <rlane32 at gmail.com> wrote:
> Hm. Something else went wrong here. It was scheduled (nova-scheduler
> log shows it casted to virt4), and networked (nova-network log shows
> that working properly), but for some reason the instance's directory
> wasn't created. It also has no libvirt log on virt4, which means it
> was never properly launched. I can't find any specific error as to why
> this may have happened.
>
> We'll just need to delete and recreate the instance.
>
> On Wed, Feb 29, 2012 at 10:15 AM, Damian Zaremba
> <damian at damianzaremba.co.uk> wrote:
>> On 23/02/2012 00:44, Ryan Lane wrote:
>>>
>>> I've switched from using the chance scheduler to the simple scheduler
>>> for nova (https://gerrit.wikimedia.org/r/#change,2731). This means
>>> that when instances are created, the compute node with the least
>>> number of instances running will be chosen, rather than before where
>>> any random compute node was chosen.
>>>
>>> The new scheduler has some options which limit which nodes are chosen,
>>> though, and the defaults may be set a little low. If you create an
>>> instance and it stays in the scheduling or pending mode forever (or it
>>> eventually fails), let me know and I'll tweak the configuration
>>> options.
>>>
>>> - Ryan
>>>
>>> _______________________________________________
>>> Labs-l mailing list
>>> Labs-l at lists.wikimedia.org
>>> https://lists.wikimedia.org/mailman/listinfo/labs-l
>>
>>
>> It looks like wikistream-1 is stuck in pending (and nagios is spamming IRC,
>> gah - possibly someone could fix the script so it only gets instances after
>> the first puppet run).
>>
>> If you could magic some tweaking that would be awesome :)
>>
>> --
>> Damian Zaremba
>> http://about.me/damianzaremba
>>
>> _______________________________________________
>> Labs-l mailing list
>> Labs-l at lists.wikimedia.org
>> https://lists.wikimedia.org/mailman/listinfo/labs-l
>>



More information about the Labs-l mailing list