- Aggregate Managers
Inventory Attribute Structure
Our layout will try to capture the relationship between nodes (which for our current purposes means motherboards) and devices (any network/communication device). The relationship hierarchy looks like this:
TESTBED - | -> Node - -INV_... -- Inventory Attributes (e.g. motherboard serial number, disk size, cpu type,...) -INF_... -- Infrastructure Attributes (Control IP, Control interface,...) -CM_... -- CM attributes (e.g. CM IP, CM Version, ...) | -> Device -name = FQDNofNode_dev_unique# -INV_if_name= "name the os thinks this device is. (e.g. eth0)" -INV_dev_id = "xxxx:xxxx" where the x's are hex numbers that represent the device identifiers.
To reflect these relationships we're going to establish a few naming conventions. These will be enumerated in the following table
|Hardware||resource type||name||attributes (all prefixed with INV_)|
|Mother board||Node||FQDNofNode||disk, memory, mother board serial number, disk serial number,cpu type, checkin date, cpu tag|
|Wired ethernet card||Device||FQDNofNode_DEV_unique#||mac, device id, device string, bus type, device tag|
|Wireless ethernet card||Device||FQDNofNode_DEV_unique#||mac, device id, device string, bus type, device tag|
|USB connected device||Device||FQDNofNode_DEV_unique#||mac (if it exists), device id, device string, bus type, device tag|
Inventory collection is now done via a diskless netboot. The PXE links are set for the whole domain and a script that works with CMC service to boot the nodes (usually in multiple waves); the netboot image automatically starts the gathering procedure. The two scripts,gatherer.rb and inventory.sh are the core of the process and stored in the /root directory of the image. Inventory.sh is merely a launchers that specifies where to redirect logs, and where the lspci/lsusb binaries are (set via flags when invoking gatherer.rb). Inventory.sh is added to the node startup routine at priority s99. When the node boots it should immediately begin the inventory process.