 |
|||
 |
|||
 |
|||
 |
|||
 |
|||
 |
|||
 |
|||
|
|||
     |
|
 |
|
oneSIS HOWTO: The simplest setup:
Since one of the major functions of oneSIS is to configure differences
between nodes, the easiest possible setup is when every node is the same.
The only thing we'll have oneSIS do here is configure a
class of nodes
and tell a few directories to live in /ram, the oneSIS RAM disk.
|
| Example 1: Regular expression class definition | |
# /etc/sysimage.conf |
This directive says that any node matching the regular expression 'node\d+' is a compute node. This regular expression (regexp) implicity includes the ^ and $ anchors (ie: /^node\d+$/), so the entire hostname must match. For instance, a node with the hostname 'node12b' does not match the regexp above, so it would not be considered a compute node. Look here for more information on defining node classes.
Create any needed writable directories
If you are running any daemons or applications that need to have some writable space in the root filesystem, you might still need to add those to the configuration. For instance, it is necessary in several distributions to have the ability to write to /var/run and /var/lock/subsys at boot time. We'll make these directories writable by simply directing them to live in the RAM disk with some LINKDIR directives. We'll make /var/log and /tmp writable in the same way. We'll impose an upper bound on how big our RAM disk can grow though, so it doesn't get out of control and eat all the memory. With system logs going to a remote log server, a RAMSIZE of 10MB is more than enough space.
| Example 2: Creating more writeable directories | |
# /etc/sysimage.conf |
This will allocate up to 10 megabytes of space on each node for directories such as /tmp to use. Now, since we have added new LINK* directives to the configuration, don't forget to run mk-sysimage on the image.
# mk-sysimage /var/lib/oneSIS/image |
Now, since we only have one type of node (ie: a bunch of compute nodes), thats all that needs to be done! oneSIS isn't needed to configure any differences between nodes, because there aren't any. All nodes will come up NFSroot and run any start scripts and daemons that are configured in the image. You could now boot as many nodes as you want into this image. All changes to the image are global to all nodes, and all the nodes will behave in exactly the same way.
Continue on with 'a typical setup for a cluster image' to learn how to set up multiple classes of nodes with different behaviors.