Comment 6 for bug 503180

Thank Thierry,

We've finally been able to (we believe) reproduce this type of condition on our Lucid machines, and have figured out the reason why it is being triggered on lucid UEC installations. The Eucalyptus front-end components (cloud, walrus, sc) require that, on startup, the network interface that was used to register components is up and configured (i.e. has an IP address that was used at component registration time). Our init process ensures this by running after the 'network' init script completes, but in the upstart case it looks like the eucalyptus components can start after the network process has started, but before it completes. In sum, if eucalyptus starts, then the interface becomes configured, this condition will be triggered and the front-end will need to be restarted so that it can go through the initialization process again.

Note that this is a startup time requirement only (i.e., once the system is up, and for some reason the interface goes down and comes up again), the service will stay active; it is just during the startup that eucalyptus requires that the interface is alive and configured.

Is there a way to have the eucalyptus upstart scripts wait until the network interface is fully alive before starting the front-end components?

Thanks!