+++ Perry N. Myers [21/09/08 01:30 -0400]:
Darryl L. Pierce wrote:This works with the patch for the server the generates a configuration file for the node based on the interface names passed in with each mac address. Additionally, the node will pull down the configuration and process it as a script. Since the configuration will be composed of two separate parts: 1. a script to load one or more kernel modules 2. a configuration file to pass to augtool The script is downloaded to a temp directory and then passed to bash. If either of the two expected files is produced then it is further processed to configure the node.NACK This breaks networking on the nodes currently.A node boots for the first time, brings up eth0, contacts the server and goes to get its config bundle using its hostname, mac and iface name. Since the node has never been configured in the server before the config bundle it pulls down is empty meaning ovirtbr0 is never set up. eth0 stays up so normal network connectivity works but since the bridge is never configured you can't start vm's on the node.What needs to happen is a 'default' configuration needs to be applied in the event that the Node has not been explicitly configured or in the event that there is no server to contact. In the configure_from_network function we could detect the 'empty' config file and exit out of the loop to apply the default configuration at the end of that function.The controller on the server side should put something in the output when the wget it run so that the ovirt-early script can easily determine that there was no configuration on the server and the Node should drop to the 'default config' of one bridge per interface.Once this is fixed we can retest these two patches.
I added code to return an error status and not an error message when the request is bad. That way the node should not attempt to process the error message text itself. -- Darryl L. Pierce <dpierce redhat com> : GPG KEYID: 6C4E7F1B
Description: PGP signature