Red Hat GFS 6.0 -, Руководство администратора

Страница: 93 / 192

Chapter 7. Using the Cluster Configuration System 79
7.4. Changing CCS Configuration Files
Based on the LOCK_GULM locking protocol, the following list defines what can or cannot be
changed in a CCS archive while a cluster is running. There are no restrictions to making changes
to configuration files when the cluster is offline.
•
New nodes can be defined in the
nodes.ccs
file.
•
Unused node definitions can be removed from the
nodes.ccs
file.
•
New fencing devices can be defined in the
fence.ccs
file.
•
The locking servers array (
servers =
) in
cluster.ccs:cluster/lock_gulm
cannot be
changed.
•
The fencing parameters for an existing node definition in
nodes.ccs
can be changed.
•
The IP address of an existing node definition in the
nodes.ccs
file can only be changed if the
node does not have any GFS file systems mounted and is not running a LOCK_GULM server.
7.4.1. Example Procedure
This example procedure shows how to change configuration files in a CCS archive.
1. Extract configuration files from the CCA device into temporary directory
/root/alpha-new/
.
ccs_tool extract /dev/pool/alpha_cca /root/alpha-new/
2. Make changes to the configuration files in
/root/alpha-new/
.
3. Create a new CCS archive on the CCA device by using the
-O
(override) flag to forcibly over-
write the existing CCS archive.
ccs_tool -O create /root/alpha-new/ /dev/pool/alpha_cca
7.5. Alternative Methods to Using a CCA Device
If it is not possible to reserve shared storage for use as a CCA device, you can use two alternative
methods:
•
Section 7.5.1 CCA File and Server
•
Section 7.5.2 Local CCA Files
Neither of these methods requires shared storage to store CCS data.
7.5.1. CCA File and Server
The first alternative to a CCA device is to use a single network server to serve CCS configuration
files to all nodes in the cluster. If used, this CCS server is a single point of failure in a cluster. If a
single (non-redundant) LOCK_GULM server daemon is being used, it would be reasonable to run a
CCS server on the same node as the LOCK_GULM server. The CCS server does not have failover
capabilities.
The CCS server is called
ccs_servd
, it can be run on any node in or out of the cluster. When
CCS daemons (
ccsd
) are started on cluster nodes, the IP address of the node running
ccs_servd
is
specified instead of the name of the CCA device. The name of the cluster is also passed to
ccsd
.
The CCS server does not read CCS files directly; rather, it reads a CCA file that is a local file contain-
ing a CCS archive.