[Bug 458904] Re: When installing a node, euca_find_cluster fails to locate the cluster controller if instances are running

[Mathias Gug]

** Description changed:

  Paul Gyermonprez from Intel, reported the following bug for which I do
  not have the necessary hw to try to reproduce:
  
  Adding nodes in several waves after launching VMs is not possible :
  
  1.      The “node mode” is discovered and the step is hanging
  2.      This step is the automatic preseed
  3.      But the answer from the server (via avahi ?) is sending a link to a wrong IP.
  4.      This wrong IP is the last one added to the eth interface.and the last one is the vlan used for VM communication
  
  Workaround : add ALL nodes before you launch the first VM
  Or close all VMs and make sure the only IPv4 for your eth is the one you want
  
  ==================
  Karmic Release notes:
  
  UEC Node Controller installation failure in an existing UEC
  
  Extending an existing Ubuntu Enterprise Cloud may fail during node
  controller installation started from the ISO with the "Install Ubuntu
  Enterprise Cloud" option. The Node installation fails with an error
  message stating that the preseed file cannot be downloaded from the
  Cluster Controller as the wrong IP address is used to connect to the
  Cluster Controller.
  
  The workaround is to perform a standard package installation rather than
  an ISO installation: Install a default Karmic server and then install
  the eucalyptus-nc package. Additionally the system's primary ethernet
  interface needs to be configured as a bridge and the Cloud Controller's
  eucalyptus user's public ssh key needs to be copied into the Node
  Controller's eucalyptus user's authorized_keys file. More detailed steps
  can be found in the UEC Package Install tutorial, Step 3 at
  https://help.ubuntu.com/community/UEC/PackageInstall.
  
  ==================
  Avahi SRU
  
  Impact:
  Public IPs are added to the Cluster Controller (CC) when instances are booted. These IPs are then automatically published by the avahi-daemon running on the CC. The instance public IPs should not be resolvable via avahi.
  
  How the bug has been addressed:
  The default avahi-daemon upstart job has been modified to use /etc/eucalyptus/avahi-daemon.conf (which will be shipped by the eucalyptus package) as the configuration file instead of the default configuration file.
  
  To reproduce the bug: 
  1. Install avahi-daemon and make sure it's started with the default configuration file (ie no configuration file specified on the command line - the process should use the hostname) - both an a new install and on upgrades:
  
    mathiaz at t-avahi:~$ ps -ef | grep avahi
    avahi     1373     1  0 11:30 ?        00:00:00 avahi-daemon: running [t-avahi.local]
    avahi     1375  1373  0 11:30 ?        00:00:00 avahi-daemon: chroot helper
    mathiaz   1422   977  0 11:31 pts/0    00:00:00 grep avahi
  
  2. Copy /etc/avahi/avahi-daemon.conf to  /etc/eucalyptus/avahi-
  daemon.conf and edit /etc/eucalyptus/avahi-daemon.conf to change the
  published hostname to eucalyptus. Restart avahi-daemon. Make sure the
  process has been restarted and running with the new eucalyptus avahi-
  daemon configuration file:
  
    mathiaz at t-avahi:~$ ps -ef | grep avahi
    avahi     2070     1  1 11:35 ?        00:00:00 avahi-daemon: running [eucalyptus.local]            
    avahi     2071  2070  0 11:35 ?        00:00:00 avahi-daemon: chroot helper                          
    mathiaz   2073   977  0 11:35 pts/0    00:00:00 grep avahi
  
  Regression potential:
  eucalyptus is not installed by default - avahi-daemon should just restart normally. On existing eucalyptus Cluster Controllers, resolution via avahi for the CC will not work until the eucalyptus-cc-publication-ip job has started.
  
  ==================
+ eucalyptus sru
+ 
+ Impact:
+ Adding new Node Controllers by installing them via the -server ISO will fail if instances with public IPs are running. This is due to the fact that public IPs are added on the Cluster Controller whose avahi-daemon will automatically publish. Since the installer relies on Avahi resolution to find the IP address of the Cluster Controller, the install will fail while trying to download the preseed file from the last public IP added to the CC.
+ 
+ How the bug has been addressed:
+ A specific avahi daemon configuration has been added that will not publish addresses by default - instead a specific upstart job has been added that will publish the CC IPs.
+ 
+ To reproduce the bug:
+ 1. Install a Cluster Controller and one Node Controller. Start an instance with a public IP and note its public IP.
+ 2. Boot from the -server iso and select to install an Ubuntu Entreprise Cloud.
+ 3. Before the update, the installation process will fail with an error message stating that the preseed file cannot be downloaded from the public IP address of the instance.
+ 4. After the update on the CC, the installation process is successful (ie the preseed file is correctly downloaded from the CC IP).
+ 
+ Regression potential:
+ The avahi-daemon running on the CC will stop publishing any IP addresses configured on any interfaces from the CC. Connection from a local network relying on avahi dns resolution will be broken. Note that it only affects system which have the eucalyptus-cc package installed.
+ 
+ ==================

-- 
When installing a node, euca_find_cluster fails to locate the cluster controller if instances are running
https://bugs.launchpad.net/bugs/458904
You received this bug notification because you are a member of Ubuntu
Server Team, which is subscribed to eucalyptus in ubuntu.