Troubleshooting
Problem
An Informix Dynamic Server (IDS) Cluster is a set of IDS Instances that are tightly-coupled using High Availability Replication to replicate data across network connections. One of the methods to do this is to share data using a Storage Area Network.
Symptom
Managing a cluster of Informix Dynamic Servers has several considerations. This article discusses the important aspects needed for a Shared Disk Secondary Server, also known as SDS.
Cause
Setting up a shared disk over a network is operating system dependent. What are the options and main considerations needed to set up a Shared Disk?
Environment
In order to set up a Shared Disk over a network, OS-specific software is needed. The category of software is known as Cluster Management Software. The exact product name depends on the operating system and may vary over time. The more common names include:
- IBM HACMP
- HP ServiceGuard
- Sun Cluster
- Veritas Storage Foundation
- Red Hat Linux Cluster Suite
Diagnosing The Problem
Several components of Cluster Management Software are used to assure that things work together:
- Cluster File System
- Multi-Owner Logical Volume Manager (LVM) or Concurrent Logical Volume Manager CLVM)
- Lock Manager
- Heart beat monitoring
- I/O Fencing
Resolving The Problem
Hardware and Operating System Requirements
- The primary and secondary servers must be able to run the same Dynamic Server executable image, even if they do not have identical hardware or operating systems.
- The hardware that runs the primary and secondary database servers must support network capabilities.
Additional considerations:
- It helps if your SAN supports SCSI-3 PR.
- If the SDS will be expected to be a fail-over server, it should be able to handle the same load as the primary.
- If the SDS node is too small, performance can impact the Primary server.
The Informix Connection Manager (CM) serves as the Cluster Manager, and performs several roles:
- It monitors the IDS cluster members (referred to as heart beats).
- In addition, each IDS member monitors the other nodes in the cluster.
- Disruptions can be automatically redirected to another instance.
- It acts as a proxy, if legacy clients are in use or firewall rules are strict.
- It manages fail-over order, if an instance is lost.
- It manages load balancing.
Further details about Cluster File Systems:
- Global File System (GFS) – Is included with the Red Hat Enterprise Cluster Suite (5 and above)
General Parallel File System (GPFS) – This is a high performance Shared Disk Clustered file system developed by IBM. It is available for use with AIX, Linux and Windows, and can be used with the IBM HACMP Cluster Management Software.
Veritas Storage Foundation Cluster File System (CFS) – This is a Cluster Volume Manager by Symantec, available on most commercial UNIX distributions. It can either be used as Cluster Manager Software or it can be used with Sun Cluster Managers also.
HP PolyServe File System (PSFS) – This is a cluster file system from HP. It is available on Windows and Linux. It can be used as a Cluster Manager as well.
Logical Volume Managers (LVM)
If a LVM is required in your environment, in most cases it should be a Multi-User LVM. Examples include:
- The Sun Cluster initial cluster framework includes a LVM that can be used via multiple nodes.
- The Linux Cluster Suite Provides a CLVM (Concurrent Logical Volume Manager )
- Veritas provides it’s own. It is known as VxCLVM
- GPFS (described in previous section) allows a single node LVM when a multi-node LVM is not needed.
I/O Fencing
I/O Fencing is the method used to ensure integrity of critical information by preventing data corruption. It is recommended for the following scenarios:
- If there is a ill-behaving node/application where leftover write operations could make there way to the shared storage area after a fail-over has been initiated.
- The possibility of 'Split-Brain'. During an HDR secondary fail-over, this can occur if more than one node thinks it is Primary.
- Whenever maintenance or testing is being performed, node fencing is strongly encouraged.
Types of Fencing:
- Power Fencing - Power off the node if a problem is detected.
Fiber Channel Switch Fencing (Requires SCSI-3 PR) – Blocks a port on the Fiber Channel device by removing the problematic node's reservation.
Various other methods are available depending on the Cluster Manger Software and/or hardware capabilities. See the vendor's documentation for additional options.
Related Information
Was this topic helpful?
Document Information
Modified date:
03 June 2021
UID
swg21623842