Sun Cluster 3.1 Data Services Developer's Guide

Appendix E Requirements for Non-Cluster Aware Applications

An ordinary, non-cluster-aware application must meet certain requirements to be a candidate for high availability (HA). The section, Analyzing the Application for Suitability, lists these requirements. This appendix provides additional details for certain of the items in that list.

An application is made highly available by configuring its resources into resource groups. The application's data is placed on a highly available global file system, making the data accessible by a surviving server in the event that one server fails. See information regarding cluster file systems in Sun Cluster 3.1 Concepts.

For network access by clients on the network, a logical network IP address is configured in logical host name resources that are contained in the same resource group as the data service resource. The data service resource and the network address resources fail over together, causing network clients of the data service to access the data service resource on its new host.

Multihosted Data

The highly available global file systems' disk sets are multihosted so that when a physical host crashes, one of the surviving hosts can access the disk. For an application to be highly available, its data must be highly available, and thus its data must reside in the global HA file systems.

The global file system is mounted on disk groups that are created as independent entities. The user can choose to use some disk groups as mounted global file systems and others as raw devices for use with a data service, such as HA Oracle.

An application might have command-line switches or configuration files pointing to the location of the data files. If the application uses hard-wired pathnames, you could change the pathnames to symbolic links that point to a files in a global file system, without changing the application code. See Using Symbolic Links for Multihosted Data Placement for a more detailed discussion about using symbolic links.

In the worst case, the application's source code must be modified to provide some mechanism for pointing to the actual data location. You could do this by implementing additional command-line switches.

Sun Cluster supports the use of UNIX UFS file systems and HA raw devices configured in a volume manager. When installing and configuring , the system administrator must specify which disk resources to use for UFS file systems and which for raw devices. Typically, raw devices are used only by database servers and multimedia servers.

Using Symbolic Links for Multihosted Data Placement

Occasionally an application has the path names of its data files hard-wired, with no mechanism for overriding the hard-wired path names. To avoid modifying the application code, you can sometimes use symbolic links.

For example, suppose the application names its data file with the hard-wired path name /etc/mydatafile. You can change that path from a file to a symbolic link that has its value pointing to a file in one of the logical host's file systems. For example, you can make it a symbolic link to /global/phys-schost-2/mydatafile.

A problem can occur with this use of symbolic links if the application, or one of its administrative procedures, modifies the data file name as well as its contents. For example, suppose that the application performs an update by first creating a new temporary file, /etc/ Then it renames the temporary file to have the real file name by using the rename(2) system call (or the mv(1) program). By creating the temporary file and then renaming it to the real file, the data service is attempting to ensure that its data file contents are always well formed.

Unfortunately, the rename(2) action destroys the symbolic link. The name /etc/mydatafile is now a regular file, and is in the same file system as the /etc directory, not in the cluster's global file system. Because the /etc file system is private to each host, the data is not available after a takeover or switchover.

The underlying problem in this situation is that the existing application is not aware of the symbolic link and was not written with symbolic links considered. To use symbolic links to redirect data access into the logical host's file systems, the application implementation must behave in a way that does not obliterate the symbolic links. So, symbolic links are not a complete remedy for the problem of placing data on the cluster's global file systems.

Host Names

You must determine whether the data service ever needs to know the host name of the server on which it is running. If so, the data service might need to be modified to use a logical host name (that is, a host name configured into a logical host name resource that resides in the same resource group as the application resource), rather than that of the physical host.

Occasionally, in the client-server protocol for a data service, the server returns its own host name to the client as part of the contents of a message to the client. For such protocols, the client could be depending on this returned host name as the host name to use when contacting the server. For the returned host name to be usable after a takeover or switchover, the host name should be a logical host name of the resource group, not the name of the physical host. In this case, you must modify the data service code to return the logical host name to the client.

Multihomed Hosts

The term multihomed host describes a host that is on more than one public network. Such a host has multiple host names and IP addresses. It has one host name-IP address pair for each network. Sun Cluster is designed to permit a host to appear on any number of networks, including just one (the non-multihomed case). Just as the physical host name has multiple host name-IP address pairs, each resource group can have multiple host name-IP address pairs, one for each public network. When Sun Cluster moves a resource group from one physical host to another, the complete set of host name-IP address pairs for that resource group is moved.

The set of host name-IP address pairs for a resource group is configured as logical host name resources contained in the resource group. These network address resources are specified by the system administrator when the resource group is created and configured. The Sun Cluster Data Service API contains facilities for querying these host name-IP address pairs.

Most off-the-shelf data service daemons that have been written for the Solaris environment already handle multihomed hosts properly. Many data services do all their network communication by binding to the Solaris wildcard address INADDR_ANY. This binding automatically causes the data services to handle all the IP addresses for all the network interfaces. INADDR_ANY effectively binds to all IP addresses currently configured on the machine. A data service daemon that uses INADDR_ANY generally does not have to be changed to handle the Sun Cluster logical network addresses.

Binding to INADDR_ANY Versus Binding to Specific IP Addresses

Even when non-multihomed hosts are used, the Sun Cluster logical network address concept enables the machine to have more than one IP address. The machine has one IP address for its own physical host and additional IP addresses for each network address (logical host name) resource that it currently masters. When a machine becomes the master of a network address resource, it dynamically acquires additional IP addresses. When it gives up mastery of a network address resource, it dynamically relinquishes IP addresses.

Some data services cannot work properly in a Sun Cluster environment if they bind to INADDR_ANY. These data services must dynamically change the set of IP addresses to which they are bound as the resource group is mastered or unmastered. One strategy for accomplishing the rebinding is to have the starting and stopping methods for these data services kill and restart the data service's daemons.

The Network_resources_used resource property permits the end user to configure a specific set of network address resources to which the application resource should bind. For resource types that require this feature, the Network_resources_used property must be declared in the RTR file for the resource type.

When the RGM brings the resource group online or offline, it follows a specific order for plumbing, unplumbing and configuring network address up or down in relation to when it calls call data service resource methods. See Deciding on the Start and Stop Methods to Use.

By the time the data service's Stop method returns, the data service must have stopped using the resource group's network addresses. Similarly, by the time the Start method returns, the data service must have started to use the network addresses.

If the data service binds to INADDR_ANY rather than to individual IP addresses, the order in which data service resource methods are called and network address methods are called is not relevant.

If the data service's stopping and starting methods accomplish their work by killing and restarting data service's daemons, then the data service stops and starts using the network addresses at the appropriate times.

Client Retry

To a network client, a takeover or switchover appears to be a crash of the logical host followed by a fast reboot. Ideally, the client application and the client-server protocol are structured to do some amount of retrying. If the application and protocol already handle the case of a single server crashing and rebooting, then they also will handle the case of the resource group being taken over or switched over. Some applications might elect to retry endlessly. More sophisticated applications notify the user that a long retry is in progress and enable the user to choose whether to continue.