MariaDB MaxScale 2.4.0 Release Notes -- 2019-06-29
MariaDB MaxScale 2.4.0 Release Notes -- 2019-06-29
Release 2.4.0 is a Beta release.
This document describes the changes in release 2.4.0, when compared to release 2.3.
For any problems you encounter, please consider submitting a bug report at Jira.
Section and object names
Section and object names starting with
@@ are now reserved for
use by MaxScale itself. If any such names are encountered in
configuration files, then MaxScale will not start.
Whitespace in section names that was deprecated in 2.2 will now be rejected and cause the startup of MaxScale to fail.
Binding on network ports
MaxScale 2.4.0 will now use the SO_REUSEPORT capability offered by newer kernels that allows reuse of network listener ports. In practice this means improved connection creation speed with more dynamic balancing of connections.
As a side-effect of this, it is possible for two MaxScale instances to bind on
the same listener port on systems that have Linux kernels newer than 3.9. This
can only happen if the MaxScale instances use completely different directory
structures (i.e. different
--basedir arguments). Normal use of MaxScale still
detects multiple MaxScales trying to bind to the same ports. Almost always, this
will not have any negative side-effects.
Stronger hashing algorithm for admin user passwords
The administrative user passwords are now stored as SHA2-512 hashes which is an improvement over the older MD5 hashing algorithm. New users will use the stronger algorithm but old users will continue using the weaker one. To upgrade administrative users, recreate the user.
protocol parameter on server creation
protocol parameter must now always be defined when a server is
created. The previously undocumented default value of
mariadbbackend now must
be explicitly defined when a server is created via the REST API.
TLS on server creation
To create encrypted connection to a server, the TLS parameters must be defined at server creation time. To enable TLS for a server that doesn't have it, destroy the old one and recreate it afterwards.
Enabling server TLS via MaxAdmin
As TLS for servers must now be defined at creation time, enabling TLS at runtime via MaxAdmin is no longer possible. Use MaxCtrl to create servers with TLS enabled.
debugcli router and the
telnetd protocol module it uses have been
ndbclustermon module has been removed.
mmmon module has been removed as the
mariadbmon monitor largely does
what it used to do.
The following settings have been removed and cause a startup error
log_to_shm parameter that was removed in 2.3 will be treated as an unknown
parameter in 2.4.0.
mqfilter has been deprecated and it will be removed in a future version
We advise against using it.
MaxScale no longer ships the example scripts and configuration files for Nagios.
MaxScale now contains support for Clustrix in the form of a Clustrix monitor that is capable of monitoring a Clustrix cluster.
Please see the documentation for details.
MaxScale has now a new router SmartRouter that is capable of routing a query to different kinds of backends, containing the same data, depending on which backend can best handle that particular kind of query.
Please see the documentation for details.
Servers can be drained
It is now possible to drain a server, which means that existing connections to the server can continue to be used but new connections are no longer created to the server.
In the output of
maxctrl, the fact that a server is being drained
is visible in the
State column as the value
┌─────────┬─────────────────┬──────┬─────────────┬───────────────────────────────┬───────┐ │ Server │ Address │ Port │ Connections │ State │ GTID │ ├─────────┼─────────────────┼──────┼─────────────┼───────────────────────────────┼───────┤ │ Server1 │ 192.168.121.159 │ 3306 │ 2 │ Master, Running │ 0-1-6 │ ├─────────┼─────────────────┼──────┼─────────────┼───────────────────────────────┼───────┤ │ Server2 │ 192.168.121.80 │ 3306 │ 1 │ Draining, Slave, Running │ 0-1-6 │ ├─────────┼─────────────────┼──────┼─────────────┼───────────────────────────────┼───────┤ │ Server3 │ 192.168.121.122 │ 3306 │ 2 │ Slave, Running │ 0-1-6 │ ├─────────┼─────────────────┼──────┼─────────────┼───────────────────────────────┼───────┤ │ Server4 │ 192.168.121.144 │ 3306 │ 2 │ Slave, Running │ 0-1-6 │ └─────────┴─────────────────┴──────┴─────────────┴───────────────────────────────┴───────┘
A server is set in the Draining state the same way as it is set in the Maintenance state:
$ maxctrl set server Server2 drain
Note that although the state is displayed as
Draining, when setting
and clearing the state, the word
drain is used.
Note that the full implication of draining a server depends upon both on the role of the server and on the router being used, and its configuration.
For instance, if readwritesplit is used and the server being drained
is a slave, then from a client's perspective there will be no difference;
readwritesplit will simply not use that server. However, if the server
being drained is the master, then it will not be possible to connect
master_failure_mode has been set to something else but the
Once the server has been drained, the state will be
weightby Replacement for Servers:
rank parameter is
the replacement for the deprecated
weightby parameter. It allows explicit
groupings of servers into primary and secondary groups. Servers configured with
rank=secondary will only be used if no primary servers are available.
UNIX Domain Socket for Servers
Servers can now use the
socket parameter to define
a local UNIX domain socket through which the connections will be created.
The servers a service uses can now be specified using the
parameter of the service.
[TheService] ... cluster=TheMonitor
In this case, the servers of the service will be defined by the
referred to monitor. Note that the parameters
are mutually exclusive.
In the MaxScale configuration file, durations can now be suffixed with
ms to indicate that the duration is specified as
hours, minutes, seconds or milliseconds.
Please see the configuration guide for details.
Not providing an explicit unit is strongly discouraged as it will be deprecated in MaxScale 2.5.
Query Classifier Cache
It is now possible to examine the contents of the query classifier cache. The REST-API endpoint is
and the equivalent maxctrl command
maxctrl show qc_cache
The output shows the statements (the canonical version) in the cache, the number of times they have been encountered and how they have been classified.
Connection Attempt Throttling
If a user fails to authenticate multiple times, the host from where the user is
connecting from will be blocked for 60 seconds. See
for more information.
REST API & MaxCtrl
Default API Version
The API version prefix is now optional and if not present, will be assumed to be
the latest version which currently is
Hard maintenance mode
--force option for the
set server command in MaxCtrl allows all
connections to the server in question to be closed when it is set into
maintenance mode. This causes idle connections to be closed immediately.
For more information, read the
REST-API documentation for
The interactive mode for MaxCtrl now has command history.
alter commands in MaxCtrl now accept multiple key-value pairs in one
command. See output of
maxctrl help alter for more information.
For more information on the readwritesplit router, refer to the documentation.
The transaction replay functionality will now also be applied in conjunction with server initiated transaction rollbacks.
transaction_replay_attempts parameter controls how many errors the
transaction replay mechanism tolerates before giving up on the replay
attempt. The number of transaction replay attempts is now capped to a default
value of 5.
Lazy connection creation delays the opening of all connections until they are needed. This reduces the load that is placed on the backend servers when the client connections are short. This feature is disabled by default.
The servers where new connections are created at the start of a session are now
always use connection counts. This allows the use of
Readwritesplit will now load balance master connections in case there are multiple master servers. This is mainly of relevance only with Clustrix clusters.
Readwritesplit now allows open transactions to finish if the master is put into
maintenance mode. To forcefully close all connections to a server use the
maxctrl set server <name> maintenance --force command.
If a slave server is replicating from a Galera node, galeramon will now correctly assign it the Slave status.
Galera nodes will now display their GTID positions in the output of
maxctrl list servers.
Avrorouter Direct Replication
By defining the
servers parameter for the avrorouter service, the replication
is done directly from a remote master server. This skips the binlogrouter
definition completely making the conversion process faster and more space
This MariaDB-Monitor setting allows the monitor greater freedom in managing the backend servers. Please see MariaDB-Monitor documentation for more information.
Known Issues and Limitations
There are some limitations and known issues within this version of MaxScale. For more information, please refer to the Limitations document.
RPM and Debian packages are provided for the Linux distributions supported by MariaDB Enterprise.
Packages can be downloaded here.
The source code of MaxScale is tagged at GitHub with a tag, which is identical with the version of MaxScale. For instance, the tag of version X.Y.Z of MaxScale is X.Y.Z. Further, master always refers to the latest released non-beta version.
The source code is available here.