Why to automate MariaDB deployments and management
MariaDB includes a powerful configuration system. This is enough when we need to deploy a single MariaDB instance, or a small number of instances. But many modern organisations have many database servers. Deploying and upgrading them manually could require too much time, and would be error-prone.
Infrastructure as Code
Several tools exist to deploy and manage several servers automatically. These tools operate at a higher level, and execute tasks like installing MariaDB, running queries, or generating new configuration files based on a template. Instead of upgrading servers manually, users can launch a command to upgrade a group of servers, and the automation software will run the necessary tasks.
Servers can be described in a code repository. This description can include MariaDB version, its configuration, users, backup jobs, and so on. This code is human-readable, and can serve as a documentation of which servers exist and how they are configured. The code is typically versioned in a repository, to allow collaborative development and track the changes that occurred over time. This is a paradigm called Infrastructure as Code.
Automation code is high-level and one usually doesn’t care how operations are implemented. Their implementation is delegated to modules that handle specific components of the infrastructure. For example a module could equally work with apt and yum package managers. Other modules can implement operations for a specific cloud vendor, so we declare we want a snapshot to be done, but we don’t need to write the commands to make it happen. For special cases, it is of course possible to write BASH commands, or scripts in every language, and declare that they must be run.
Manual interventions on the servers will still be possible. This is useful, for example, to investigate performance problems. But it is important to leave the servers in the state that is described by the code.
This code is not something you write once and never touch again. It is periodically necessary to modify infrastructures to update some software, add new replicas, and so on. Once the base code is in place, making such changes is often trivial and potentially it can be done in minutes.
Once replication is in place, two important aspects to automate are load balancing and failover.
Proxies can implement load balancing, redirecting the queries they receive to different server, trying to distribute the load equally. They can also monitor that MariaDB servers are running and in good health, thus avoiding to send queries to a server that is down or struggling.
However, this does not solve the problem with replication: if a primary server crashes, its replicas should point to another server. Usually this means that an existing replica is promoted to a master. This kind of changes are possible thanks to MariaDB [gtid|GTID].
One can promote a replica to a primary by making change to existing automation code. This is typically simple and relatively quick to do for a human operator. But this operation takes time, and in the meanwhile the service could be down.
Automating failover will minimise the time to recover. A way to do it is to use Orchestrator, a tool that can automatically promote a replica to a primary. The choice of the replica to promote is done in a smart way, keeping into account things like the servers versions and binary log format.
Content initially contributed by Vettabase Ltd.