LLI Wiki — My Digital Knowledge Base

Checklist: SQL Server Always On deployment on VMware

Prerequisites to validate before deploying an Always On availability group on a VMware failover cluster: AD, network, accounts, storage and instance.

This checklist validates the prerequisites before deploying a SQL Server Always On availability group on a Windows failover cluster running on VMware virtual machines. Sizing values are examples to adapt.

General scope

Item Example / expected value
SQL Server version SQL Server 2019 Enterprise
Number of servers / instances 4 servers, 4 instances
Number of replicas per database 3
Location 2 x 2 nodes on the same site, in different datacenters
Estimated number, size and activity of databases to be documented before deployment

Active Directory and group policies

Network

Item Note
Subnets a public network, a private network (cluster heartbeat), optionally an administration network
Cluster IP address one dedicated IP address for the failover cluster
Listener one IP address per availability group
Listener name follow a naming convention (for example a lsn_ prefix)
Listener port one port per listener, for example in the 15000 to 15010 range
Mirroring endpoint port one port per instance; 5022 by default
Heartbeat dedicated VLAN with fixed IP addresses to improve stability
DNS decide whether DNS aliases are used

Service accounts

Plan the following accounts in advance:

Servers (virtual machines)

Item Example
CPU depends on the workload (for example 4 vCPU for the OS + 12 per instance)
RAM 64 GB
Virtual SCSI controllers 2 (VMware storage)
Virtual network adapters 3 (public, private, administration)

To verify on every node:

Failover cluster

Storage

Volume Format Note Indicative size
System (OS) GPT 800 GB (example)
Data (per instance) GPT 64 KB allocation unit 100 GB
System databases GPT 64 KB allocation unit 50 GB
Tempdb GPT SSD recommended 50 GB
Transaction logs GPT 64 KB allocation unit 50 GB

Confirm with the system administrators that volumes can be extended online, including at the virtual machine level.

Backup

Instance

Item Value
Trace flag to enable permanently -T3226 (suppresses successful backup messages in the error log)
Named instance yes (for example PROD)
Collation define it consistently on all instances
Tempdb number of data files equal to the number of CPUs
Monitoring to plan (Centreon, Zabbix...)

Naming

Define a naming convention before deployment for databases, availability groups (resource name in the cluster) and listeners.

References

Mastodon LLIMastodon SocialFosstodon