Appscale on Xen with 2 virtual hosts

[Nim]

Hi! I'm trying to deploy appscale over two virtual containers with
following configuration:

master - ubuntu 10.04 lts, 2 network bidged interfaces (first for the
private network and the second one for public)
slave - ubuntu 10.04 lts with one network interface looking at the
same private network as the master node
Succsessfuly installed: appscale-core appscale-memcachedb appscale-
tools appscale-voldemort, run appscale-add-keypair.
Trying to deploy example guestbook application with config listed
below:

---
:controller: <some public ipaddr>
:servers:
- 192.168.12.52

public ipaddr is visible for slave node via masquerading, but neither
this, nor changing it to the next private address helps me to resolve
deployment process hanging after the "Please wait for your app to
start up." message.
Any suggestions?

[Chris Bunch]

Hi Nim,
By chance are you running the tools from a machine that can access the
master but not the slave? In our default configuration, we place
application servers (AppServers) on what the slave box, and the tools
try to access them to see if they're running yet. From what your
configuration looks like, it seems as though they wouldn't be reachable
(they're on a private network that only the master can see).

Could you try running a single node deployment with just the master
first to see if that's the problem? Your ips.yaml file should just
require a simple change (see the lack of a 'servers' entry):

> ---
> :controller: <some public ipaddr>
>

Let us know how that works for you.

Thanks,

[Robin Hood]

I also met this problem. I changed the ips.yaml to use private ips for
all instances, then everything is OK. You can have try.

2010/11/15 Nim <nim...@gmail.com>:

> --
> You received this message because you are subscribed to the Google Groups "AppScale Community" group.
> To post to this group, send email to appscale_...@googlegroups.com.
> To unsubscribe from this group, send email to appscale_commun...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/appscale_community?hl=en.
>
>

[Nim]

So, I've edited my ips.yaml to leave only :controller: section with
it's public ip address and relaunched run-instances utility (appscale-
run-instances --ips ips2.yaml -v --table memcachedb --file). Nothing
seems changed. Deployment process hangs on "Please wait for your app
to start up." message.

[Nim]

I've tried the same steps while spot this behaviour, but it didn't
help

On 15 ноя, 21:49, Robin Hood <tangpu...@gmail.com> wrote:
> I also met this problem. I changed the ips.yaml to use private ips for
> all instances, then everything is OK. You can have try.
>

> 2010/11/15 Nim <nimn...@gmail.com>:

[Chris Bunch]

Hi Nim,
When you run it with -v, there should be a log file placed on your
master node at /tmp/{master node ip}.log that should contain more
information about the problem. Could you e-mail me that directly so that
we can look it over?

Thanks,

[Nim]

Thanks. Cris. But its seems nothing to email you cause 1-node
configuration has been started after investigating on to the /tmp/
{master node ip}.log My problem was in the nginx misconfiguration. It
missed mime.types file and was unable to start.

[Nim]

Ok, then. I've tried to expand appscale through the several hosts.
I've read wiki about "how to place critical system services manually".
So I need some help to suggest roles for the nodes, in case I have one
master with public and private net interfaces and few slaves connected
to the private subnet (same with the master's one).

Of course, I could install appscale tools directly in to the xen dom0
and just manage domU containers, but it is still one error I've got
with ssl encryption, when I've tried to map application ports from
public dom0 to the private domU.

[Chris Bunch]

Hello,
The main point of the placement support is to offer you the ability to
place components if you already know the roles you want to experiment
with - if you don't have a particular set of roles in mind and the wiki
page doesn't give you any ideas, I would stick with the default roles
the tools set up for you.

I'm not sure I understand the second part of your mail - some users
install the tools on their master node to get around network problems
that can sometimes show up, but I'm not sure if that's what you're
saying. If you see network error messages in that verbose output,
though, that would be a solution I would recommend.

Thanks,