System Failover Strategies
Caché provides several high availability (HA) solutions, and easily integrates with all common HA configurations supplied by operating system providers.
The primary mechanism for maintaining high system availability is called failover. Under this approach, a failed primary system is replaced by a backup system; that is, processing fails over to the backup system. Many HA configurations also provide mechanisms for disaster recovery, which is the resumption of system availability when failover mechanisms have been unable to keep the system available.
There are five general approaches to Caché instance failover for HA (including not implementing an HA strategy). This chapter provides an overview of these approaches, while the remainder of this guide provides procedures for implementing them.
It is important to remember that in all of these approaches except mirroring, a single storage failure can be disastrous. For this reason, disk redundancy, database journaling as described in the “Journaling” chapter of the Caché Data Integrity Guide, and good backup procedures, as described in the “Backup and Restore” chapter of the Caché Data Integrity Guide, must always be part of your approach, as they are vital to mitigating the consequences of disk failure.
If you require detailed information to help you develop failover and disaster recovery strategies tailored to your environment, or to review your current practices, please contact the InterSystems Worldwide Response Center (WRC).
This guide does not address disaster recovery (DR), except in regard to the DR capabilities of Caché mirroring, as described in the “Mirroring” chapter.
No Failover Strategy
The integrity of your Caché database is always protected from production system failure by the features described in the Caché Data Integrity Guide. Structural database integrity is maintained by Caché write image journal (WIJ) technology, while logical integrity is maintained through journaling and transaction processing. Automatic WIJ and journal recovery are fundamental components of Intersystems’ “bulletproof” database architecture.
With no failover strategy in place, however, a failure can result in significant down time, depending on the cause of the failure and your ability to isolate and resolve it. For many applications that are not business-critical, this risk may be acceptable.
Customers that adopt this approach share the following traits:
Clear and detailed operational recovery procedures, including journaling and backup and restore
Disk redundancy (RAID and/or disk mirroring)
Ability to replace hardware quickly
24x7 maintenance contracts with all vendors
Management acceptance and application user tolerance of moderate downtime caused by failures
A common approach to achieving HA is the failover cluster, in which the primary production system is supplemented by a (typically identical) standby system, with shared storage and a cluster IP address that follows the active member. In the event of a production system failure, the standby assumes the production workload, taking over the programs and services formerly running on the failed primary, including Caché.
Caché is designed to integrate easily with failover solutions provided at the operating system level, such as Microsoft Windows Server Clusters, IBM PowerHA SystemMirror, Red Hat Enterprise Linux HA, and others. A single instance of Caché is installed on the shared storage device so that both cluster members recognize the instance, then added to the failover cluster configuration so it will be started automatically as part of failover. If the active node becomes unavailable for a specified period of time, the failover technology transfers control of the cluster IP address and shared storage to the standby and restarts Caché on the new primary. On restart, the system automatically performs the normal startup recovery, with WIJ, journaling, and transaction processing maintaining structural and data integrity exactly as if Caché had been restarted on the failed system.
The standby server must be capable of handling normal production workloads for as long as it may take to restore the failed primary. Optionally, the standby can become the primary, with the failed primary becoming the standby once it is restored.
Under this approach, failure of the shared storage device is disastrous. For this reason, disk redundancy, journaling and good backup and restore procedures are critically important to providing adequjate recovery capability.
Virtualization platforms generally provide HA capabilities, which typically monitor the status of both the guest operating system and the hardware it is running on. On the failure of either, the virtualization platform automatically restarts the failed virtual machine, on alternate hardware as required. When the Caché instance restarts, it automatically performs the normal startup recovery, with WIJ, global journaling, and transaction processing maintaining structural and data integrity as if Caché had been restarted on a physical server.
In addition, virtualization platforms allow the relocation of virtual machines to alternate hardware for maintenance purposes, enabling upgrade of physical servers, for example, without any down time. Virtualization HA shares the major disadvantage of the failover cluster and concurrent cluster, however: failure of shared storage is disastrous.
Caché database mirroring with automatic failover provides an effective and economical high availability solution for planned and unplanned outages. Mirroring relies on data replication rather than shared storage, avoiding significant service interruptions due to storage failures.
A Caché mirror consists of two physically independent Caché systems, called failover members. Each failover member maintains a copy of each mirrored database in the mirror; application updates are made on the primary failover member, while the backup failover member’s databases are kept synchronized with the primary through the application of journal files from the primary. (See the “Journaling” chapter of the Caché Data Integrity Guide for information about journaling.)
The mirror automatically assigns the role of primary to one of the two failover members, while the other failover member automatically becomes the backup system. When the primary Caché instance fails or otherwise becomes unavailable, the backup automatically and rapidly takes over and becomes primary.
A third system, called the arbiter, maintains continuous contact with the failover members, providing them with the context needed to safely make failover decisions when they cannot communicate directly. Agent processes running on each failover system host, called ISCAgents, also help with automatic failover logic. The backup cannot take over unless it can confirm that the primary is really down or unavailable and will not attempt to operate as primary. Between the arbiter and the ISCAgents, this can be accomplished under almost every outage scenario.
Alternatively, when using a hybrid virtualization and mirroring HA approach (as discussed later in this section), the virtualization platform can restart the failed host system, allowing mirroring to determine the status of the former primary instance and proceed as required.
When the mirror is configured to use a virtual IP address (VIP), redirection of application connections to the new primary is transparent. If connections are by ECP, they are automatically reset to the new primary. Other mechanisms for redirection of application connections are available.
When the primary instance is restored to operation, it automatically becomes the backup. Operator-initiated failover can also be used to maintain availability during planned outages for maintenance or upgrades.
The use of mirroring in a virtualized environment creates a hybrid high availability solution combining the benefits of both. While the mirror provides the immediate response to planned or unplanned outages through automatic failover, virtualization HA software automatically restarts the virtual machine hosting a mirror member following an unplanned machine or OS outage. This allows the failed member to quickly rejoin the mirror to act as backup (or to take over as primary if necessary).
For complete information about Caché mirroring, see the “Mirroring” chapter of this guide.
Using ECP with a Failover Strategy
Whatever approach you take to HA, Enterprise Cache Protocol (ECP) can be used to provide a layer of insulation between the users and the database server. Users remain connected to ECP application servers when the database server fails; user sessions that are actively accessing the database during the outage will pause until the database server becomes available again via completion of failover or restart of the failed system.
Bear in mind, however, that adding ECP to your HA strategy can increase complexity and introduce additional points of failure.
For information about ECP features and implementation, see the “ECP and High Availability” chapter of this guide and the Caché Distributed Data Management Guide.