Linux on HP Integrity Servers: A System Administrators Guide

A ServiceGuard cluster is a networked group of Linux servers that transfer control of applications when hardware or software problems occur. There is hardware reliability designed into a Service cluster in the form of many redundant components, such as LAN cards, disks, networking connections, and so on. ServiceGuard handles much of the software reliability by transferring control of applications from a failed system to an operational system. ServiceGuard uses packages, which are Linux processes grouped together to facilitate the management and fail over of important applications. Packages include the resources needed to support a specific application including disks, networking, and linux processes. The best way to understand the makeup of a ServiceGuard cluster is through a diagram. Figure 12-1 shows a basic ServiceGuard setup.

Figure 12-1 is a simple ServiceGuard cluster consisting of two systems. Note that there are many redundant components in the diagram. The disks for the operating systems (root) are mirrored. The shared disks that contain pkgA and pkgB are also mirrored. The package is setup with Logical Volume Manager (LVM) volume groups. The procedure for creating the storage required for a ServiceGuard cluster is described in the document cited earlier.

There are two LAN cards and two physical networks. The two networks are connected with a hub and not a router. Should one physical network fail, the hub provides instant connectivity to the second physical network.

When a failure occurs on the second node, the pkgB switches to the first system, as shown in Figure 12-2.

After the package switch takes place, the package and its associated IP address(es) are moved to the "adoptive" system. IP addresses are associated with each system and with each application package. When the failover occurs, the IP address of pkg2 moves over to node1 along with the application package pkg2. Users of pkg2 connect to the same IP address after failover even though pkg2 has been moved from node2 to node1.

The application package contains all the information related to applications in the ServiceGuard environment. For each package, there is a package configuration file and package control script. The package configuration file contains all dependencies for the application, such as the package name, service name, subnet, and so on. The package control script is run to start or stop the application package. These are ASCII files.

In addition to the ASCII files used to define the package and start and stop it, there is a cluster configuration file. This file has in it parameters such as the cluster name, maximum number of configured packages, and so on. This file, along with the package configuration file, are combined to produce a binary file that is distributed to all the nodes. There are a series of commands to create a package and cluster. The cmapplyconf command is used to produce the binary file.

ServiceGuard Manager

ServiceGuard clusters can be managed through a graphical interface called ServiceGuard Manager, which is available for free on software.hp.com. ServiceGuard Manager gets configuration data from nodes in a cluster and is invoked with /usr/local/sgmgr/bin/sgmgr -c clustername. ServiceGuard Manager can also be run on HP-UX and Windows systems.

You can perform a variety of cluster-related tasks with ServiceGuard Manager including viewing a cluster as shown in Figure 12-3 for the cluster morningstar.

This screen shows that the node bangalore is not in the cluster and two packages are running on the system minneapolis.

We can get more information about the systems in the cluster, as shown in Figure 12-4.

This figure shows the status of the systems, including the fact that bangalore is up but that it failed as part of the ServiceGuard cluster. The Cluster Properties window shows information about the cluster, including a tab for its nodes as shown in the figure.

You can get specific information about the system bangalore, as shown in Figure 12-5.

This figure shows information on bangalore, including the versions of Linux and ServiceGuard that the node is running. Using the tabs, you can also obtain information about the Packages and Network.

Figure 12-6 shows a summary of the ServiceGuard cluster.

This figure shows the status of the clusters. You can see that the cluster morningstar is indeed up, but that the packages can't failover because the node bangalore is not up.

This was a high-level overview of ServiceGuard on Linux systems. To users of ServiceGuard on HP-UX systems, this functionality is nearly identical. To Linux users who haven't before seen ServiceGuard, this is a highly functional and widely used high-availability tool with tens of thousands of installations on HP-UX that is worth evaluating.