InnoDB Cluster in a Nutshell: Part 2 MySQL Router

MySQL Router MySQL ClusterMySQL InnoDB Cluster is an Oracle High Availability solution that can be easily installed over MySQL to provide high availability with multi-master capabilities and automatic failover. In the previous post we presented the first component of InnoDB Cluster, group replication. Now we will go through the second component, MySQL Router.  We will address MySQL Shell in a final instalment of this three-part series. By then, you should have a good overview of the features offeed by MySQL InnoDB Cluster.

MySQL Router

This component is responsible for distributing the traffic between members of the cluster. It is a proxy-like solution to hide cluster topology from applications, so applications don’t need to know which member of a cluster is the primary node and which are secondaries.

The tool is capable of performing read/write splitting by exposing different interfaces. A common setup is to have one read-write interface and one read-only interface. This is default behavior that also exposes 2 similar interfaces to use x-protocol (i.e. used for CRUD operations and async calls).

The read and write split is done using a concept of roles: Primary for writes and Secondary for read-only. This is analogous to how members of cluster are named. Additionally, each interface is exposed via a TCP port so applications only need to know the IP:port combination used for writes and the one used for reads. Then, MySQL Router will take care of connections to cluster members depending on the type of traffic to server.

MySQL Router is a very simple tool, maybe too simple as it is a layer four load balance and lacks some of the advanced features that some of it’s competitors have (e.g.. ProxySQL).

Here is a short list of the most important features of MySQL Router:

  • As mentioned, read and write split based on roles.
  • Load balancing both for reads and writes use different algorithms.
  • Configuration is stored in a configuration test file.
  • Automatically detects cluster topology by connecting and retrieving information, based on this information the router configures itself with default rules.
  • Automatically detects failing nodes and redirects traffic accordingly.

Algorithms used for routing

An important thing to mention is the routing_strategy algorithms that are available, as they are assigned by default depending on the routing mode:

  • For PRIMARY mode (i.e. writer node – or nodes): uses the first-available algorithm that picks the first writer node from a list of writes and in case of failure moves to the next in the list. If the failing node comes back to life, it’s automatically added to the list of servers and become PRIMARY again when cluster assign this status. When no writers are available then write routing is stopped
  • For read-only mode (i.e. read nodes): uses the round-robin algorithm between servers listed in the destinations variable. This mode splits read traffic between all servers in an even manner.

Additional routing_strategy algorithms :

  • next-available: similar to first-available but in this case a failing node is marked as crashed and can’t get back into the rotation.
  • round-robin-with-fallback: same as round-robin but it includes the ability in this case of using servers from the primary list (writers) to distribute the read traffic.

A sample configuration

For performance purposes it’s recommended to setup MySQL Router in the same place as the application, considering an instance per application server.

Here you can see a sample configuration file auto-generated by --bootstrap functionality:

We are almost done now, only one post left. The final post is about our third component MySQL Shell, so please keep reading.

You May Also Like

Percona is synonymous with optimized MySQL performance. Watch our webinar on MySQL scaling and high availability for insight on how to make your database run faster and how to make it more optimized than before. Percona CEO Peter Zaitsev hosts the webinar.

Percona XtraDB Cluster (PXC) is a high availability MySQL clustering solution. It is open source and has multi-master capabilities. Percona has integrated PXC with ProxySQL to further meet your high availability and clustering needs. Watch our webinar to learn more and click here to learn how to perform schema upgrades using PXC.

Share this post

Comments (4)

  • lefred

    Hi Francisco,
    thank you for your interest in MySQL InnoDB Cluster, however, I would like to comment on 2 points:

    X Protocol is not for JSON, you can perform JSON operations also in SQL in the standard protocol. X Protocol is used for CRUD operations and for example asynchronous call.

    The second point it related to the router, when the primary comes back it doesn’t go back at the end of the list as it may become the next PRIMARY and therefore used again. So when PRIMARY is used, the router send the traffic to the Primary-Master of the Group and only the Group decide which one it is.

    July 22, 2018 at 12:48 pm
  • Francisco


    Thanks for your comments, I’ve updated the post to reflect them.

    July 23, 2018 at 12:38 pm
  • Krish

    If we have 10 app servers, will router on only 1 app server will be used ? Or all 10 routers on all app servers be used?

    August 20, 2018 at 8:39 pm
    • Francisco

      Krish, it mostly depends on the app configuration, basically each router will expose it’s one read/write interfaces but all of them will point to the same cluster so you will need to configure each app server to connect to local router so you’ll end up using the 10 instances.
      Here you have a graphic with typical deployment

      August 21, 2018 at 9:54 am

Comments are closed.

Use Percona's Technical Forum to ask any follow-up questions on this blog topic.