2

I have a working two node juju enabled MAAS that has the beginnings of hadoop installed. I got this working using the following steps:

  • bootstrapping the environment
  • adding a hadoop charm

Once I did added the charm, the second node appeared when I ran juju status. It was never able to install the agent, which always reported as not-started.

To get the agent installed, I destroy-environment'ed my juju. Then I did a bootstrap again. To my surprise all the nodes appeared in the list in MAAS as allocated to me, and then when I rebooted the second node, it all of a sudden wanted to install the image with the juju agent in it.

So, I have two functioning nodes one is the juju environment node, the other is the hadoop master. What I am trying to do is bring a third node up so that I can have some slavecluster nodes. But I am now facing the same problem as with the second node:

MAAS is not installing the juju-agent enabled image to the third node, so the agent status is always not-started.

Is this a juju problem, where juju must inform MAAS that it should install the correct image to the node? Or is this a MAAS problem where states are out of sync?

It really surprising that to get this to work, I have to destroy-environment, and then bootstrap again, and everything is OK.

Anyone have any ideas what I have done here?

Improve this question

2 Answers 2

Reset to default
1

There shouldn't be no need to re-run 'juju bootstrap' to get your nodes working. The bootstrap phase is only needed to get the juju environment started.

You said you rebooted the second node and I think that's the key: did you configure the power parameters of your nodes (I assume you run the quantal version of MAAS)? If you can't do that, once you deploy a charm manually and you see the node only which the charm will get deployed 'allocated' on the MAAS UI, you need to power on the nodes so that Ubuntu can get installed on the machine and the charm installed (this can take a while depending on the hardware you have). What I'm saying is that if your node uses IPMI or any other kind of supported power mechanism and you've configured the credentials, then MAAS will power up the nodes but if you can't do that, then you need to do that manually.

If this does not solve your problem, I suggest running 'juju debug-log' to see why the charm cannot be deployed.

Improve this answer
2
  • I am running the Precise version right now. What is not happening is the agent is not getting installed when I try to join a node to juju. The MAAS part seems to work fine - it can install the necessary components. But when I try to make the node part of the juju cluster, the agent never gets there, until I do a destroy-environment, then a bootstrap. I dont understand why this is happenning. This weekend, its time to start from scratch, and see if I can figure out where I went wrong.
    – David Buttrick
    Nov 14, 2012 at 0:40
  • This problem really isnt about the charm. I'd love to see a charm get installed without an agent. The problem really is this: I am in a situation now, where MAAS is working fine, I can commission just fine. After I juju bootstrap, that is when the problems start. I cant find any good troubleshooting for these things: That the juju environment has been installed on the juju node (ssh keys, juju agent, etc.). That the juju agent is running on the node.
    – David Buttrick
    Nov 21, 2012 at 4:01
0

What DNS domain do you have set in MaaS?

Does it end in .local?

If so, it's a problem with avahi-daemon. Edit the MaaS generic preseed and take out avahi-daemon from the install list.

Alternatively, edit /etc/nsswitch.conf and change the hosts: line to: hosts: files dns mdns4

On all nodes.

You'll know if this is the problem by not being able to ping the initial ZK node FQDN from your other nodes by name.

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .