Software requirements

Important:

HAA only supports 64-bit operating systems and must be installed on a clean host with no other applications. All HAA nodes must be synchronized with the same NTP server.

The following table lists the OS versions that we tested and confirmed to be compatible with HAA. However, due to the high number of operating system updates and variations, we cannot test compatibility for each combination of HAA version and minor OS version. While we generally expect HAA to work correctly with newer minor OS versions that we did not test, we cannot guarantee compatibility.

If you plan to upgrade to a minor OS version that we do not list in the table, you must perform the upgrade in a testing environment before deploying in a production environment.

Platform	Supported Versions
RHEL/CentOS 8	8.6, 8.8, 8.9
RHEL/CentOS 9	9.3
Ubuntu	20.04 Server version is recommended for production installations. Desktop version is only recommended for development deployments.

General cluster requirements

The HAA cluster needs three servers for a healthy operation.

Under normal circumstances, the HAA cluster servers have the following roles:

one principal server – it holds a data shard, and it accepts read and write database operations from the outside world;
one secondary server – it holds a copy of the data shard;
one secondary server – it holds no data, it exists for quorum purposes only.

The HAA cluster supports a single server failure only.

If one of the servers goes down, no matter which one, the HAA cluster continues working, and a warning is shown in the Web management interface. You can find more information by running the rlcheck command, usually found in /opt/redislabs/bin. You can also append --continue-on-error to the rlcheck command.
If two servers fail, the HAA cluster fails as well, even if the principal server is still online.

Building an HAA cluster with more than three servers is possible but offers no real benefit.

There is no increase in the number of servers that can fail. In the case of a five-node HAA cluster, if the principal node and the secondary node keeping the data shard copy both fail, the entire cluster fails as well, and there is no data shard reallocation to other nodes.