MaxScale and Clustrix Tutorial

MaxScale and Clustrix Tutorial

Since version 2.4, MaxScale has built-in support for Clustrix. This tutorial explains how to setup MaxScale in front of a Clustrix cluster.

There is no Clustrix specific router, but both the readconnroute and the readwritesplit routers can be used.

Clustrix and Readconnroute

With readconnroute you get simple connection based routing, where each new connection is created (by default) to the Clustrix node with the least amount of existing connections. That is, with readconnroute the behaviour will be very similar to the behaviour when HAProxy is used as the Clustrix load balancer.

Bootstrap servers

The Clustrix monitor is capable of autonomously figuring out the cluster configuration, but in order to get going there must be at least one server-section referring to a node in the Clustrix cluster.


That server defintion will be used by the monitor in order to connect to the Clustrix cluster. There can be more than one such "bootstrap" definition to cater for the case that the node used as a bootstrap server is down when MaxScale starts.

NOTE These bootstrap servers should only be referred to from the Clustrix monitor configuration, but never from a service.


In the Clustrix monitor section, the bootstrap servers are referred to in the same way as "ordinary" servers are referred to in other monitors.


The bootstrap servers are only used for connecting to the Clustrix cluster; thereafter the Clustrix monitor will dynamically find out the cluster configuration.

The discovered cluster configuration will be stored (the ips and ports of the Clustrix nodes) and upon subsequent restarts the Clustrix monitor will use that information if the bootstrap servers happen to be unavailable.

With the configuration above maxctrl list servers might output the following:

│ Server            │ Address      │ Port │ Connections │ State           │ GTID │
│ @@Clustrix:node-7 │ │ 3306 │ 0           │ Master, Running │      │
│ @@Clustrix:node-8 │ │ 3306 │ 0           │ Master, Running │      │
│ @@Clustrix:node-6 │ │ 3306 │ 0           │ Master, Running │      │
│ Bootstrap-1       │ │ 3306 │ 0           │ Master, Running │      │

All servers whose name start with @@ have been detected dynamically.

Note that the address appears twice; once for Bootstrap-1 and another time for @@Clustrix:node-6. The Clustrix monitor will create a dynamic server instance for all nodes in the Clustrix cluster; also for the ones used in bootstrap server sections.


The service is specified as follows:


Note that the service does not list any specific servers, but instead refers, using the argument cluster, to the Clustrix monitor.

In practice this means that the service will use the servers of the monitor named Clustrix and in the case of a Clustrix monitor those servers will be the ones that the monitor has detected dynamically. That is, when setup like this, the service will automatically adjust to any changes taking place in the Clustrix cluster.

NOTE There is no need to specify any router_options, but the default router_options=running provides the desired behaviour. In particular do not specify router_options=master as that will cause only a single node to be used.


To complete the configuration, a listener must be specified.


Clustrix and Readwritesplit

The primary purpose of the router readwritesplit is to split statements between one master and multiple slaves. In the case of Clustrix, all servers will be masters, but readwritesplit may still be the right choise.

Namely, as readwritesplit is transaction aware and capable of replaying transactions, it can be used for hiding certain events taking place in Clustrix from the clients that use it.

For instance, whenever a node is removed from or added to a Clustrix cluster there will be a group change, which is visible to a client as a transaction rollback. However, if readwritesplit is used and transaction replay is enabled, then MaxScale may be able to hide the group change so that the client only detects a slight delay.

Apart from the service section, the configuration when using readwritesplit is identical to the readconnroute configuration described above.


The service is specified as follows:


With this configuration, subject to the boundary conditions of transaction replaying, a client will neither notice group change events nor the disappearance of the very node the client is connected to. In that latter case, MaxScale will simply connect to another node and replay the current transaction (if one is active). For detailed information about the transaction replay functionality, please refer to the readwritesplit documentation.

NOTE It is vital to have slave_selection_criteria=LEAST_GLOBAL_CONNECTIONS, as otherwise connections will not be distributed evenly across all Clustrix nodes.

As a rule of thumb, use readwritesplit if it is important that changes taking place in the cluster configuration are hidden from the applications, otherwise use readconnroute.


Comments loading...
Content reproduced on this site is the property of its respective owners, and this content is not reviewed in advance by MariaDB. The views, information and opinions expressed by this content do not necessarily represent those of MariaDB or any other party.