How to make Juju High Availability work properly?

[曾建銘]

Hi Marco,

Many thanks for your reply, the juju version I used is 1.24.6-trusty-amd64.

I found that when a juju node failed, not only the services on the node
change to lost, but alse services on other working nodes change to
executing or even error.

I tried to log into the other working node and use "top" to check resource
usage. I found that a lot of CPU power was consumed by jujud in this time.
And it may take hours to become normal again.

Was the juju doing something for fixing specific problem? I think that
service on failed node should only become lost and not interfere services
on workings nodes. But it didn't act as I expected.

By the way, I used Juju to deploy OpenStack, so I deployed a lot of charms
on it. Is that matter?

Sincerely yours,
Leon

On Sun, Oct 18, 2015 at 7:16 AM, Marco Ceppi <marco at ondina.co> wrote:

> Hi Leon,
>
> Sorry to hear you're having issues, I haven't seen this problem before but
> I'm curious what version of Juju you're using (`juju version`) I know there
> was recent work to make ensure-availability more robust. As to how to solve
> the issue, could you run `juju ssh 0` then once on the zero node run:
>
> sudo apt-get install pastebinit
> pastebinit /var/log/juju/machine-0.log
>
> This will provide a URL with the pastebin of the machine-0 log which would
> be helpful in diagnosing this issue further and potentially ways to resolve
> this.
>
> Thanks,
> Marco Ceppi
>
> On Fri, Oct 16, 2015 at 3:56 AM 曾建銘 <godleon at gmail.com> wrote:
>
>> Hi All,
>>
>> I got some problems when I was testing Juju High Availability after
>> deploying OpenStack on my physical servers.
>>
>> I used "juju ensure-availability" to generate 3 state servers. Juju
>> became unnormal after the bootstrap node was shutdown.
>>
>> When the bootstrap node was gone, the whole juju tasks seemed not
>> switched to another state server successfully. I found agent-states of all
>> services became "lost", workload-state of all services become unknown or
>> error.
>>
>> I used "juju debug-log" to check the juju working status, a lot of
>> messages passed by, they looked like there were many communications between
>> services and the state server.
>>
>> I tried to wait for a while, I found that agent-states of services became
>> idle again. But they will become lost again later. Then I try to wait for
>> more a long time(more than 1 hour), I found the agent-state of all services
>> were change from lost to executing, then to idle, then to lost finally.
>>
>> No matter how long I waited, I always found the same result I mentioned
>> above. Then I could use juju commands normally.
>>
>> Did anyone get the same problem? I will be really appreciated if someone
>> can tell me how to solve this issue.
>>
>> Thanks in advanced.
>>
>> Sincerely yours,
>> Leon
>>
>> --
>> Juju mailing list
>> Juju at lists.ubuntu.com
>> Modify settings or unsubscribe at:
>> https://lists.ubuntu.com/mailman/listinfo/juju
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/juju/attachments/20151023/dff4dff1/attachment.html>