MHA helper is a Python module that supplements in doing proper failover using MHA. MHA is responsible for executing the important failover steps such as finding the most recent slave to failover to, applying differential logs, monitoring master for failure, etc. But it does not deal with additional steps that need to be taken before and after failover. These would include steps such as setting the read-only flag, killing connections, moving writer virtual IP, etc.
There are three functions of MHA Helper:
- Execute pre-failover and post-failover steps during an online failover. An online failover is one in which the original master is not dead and the failover is performed, for example, for maintenance purposes.
- Execute pre-failover and post-failover steps during a hard master failover. In this case the original master is dead, meaning either the host is dead or the MySQL server process has died.
- Daemonize the monitor that monitors the masters for failure (Currently not implemented)
First and foremost MHA itself needs to be installed. You need the MHA manager and node packages installed. You can read more about installing MHA and its dependencies here: http://code.google.com/p/mysql-master-ha/wiki/Installation
MHA Helper has been developed and tested against Python 2.6 and 2.7. Versions < 2.6 are not supported.
In addition to Python 2.6, the following Python modules are needed:
- paramiko
- PyMySQL
There are other MHA specific requirements, please go through the link below to read about them: https://code.google.com/p/mysql-master-ha/wiki/Requirements
MHA Helper packages are available in RPM format. Currently packages are only available for CentOS/RHEL 6.x
Before installing the package you will have to configure the TwinDB package repository:
curl -s https://packagecloud.io/install/repositories/twindb/main/script.rpm.sh | sudo bash
Once the repository has been configured you can install the package as follows:
yum install python-mha_helper
MHA Helper is also available as a python module mha_helper and can be installed using pip as follows:
pip install mha_helper
The most easiest way to install and configure both MHA and MHA Helper is to use the Chef cookbook mysql-mha.
Below are some of the benefits of using the cookbook:
- Installation of MHA and MHA Helper
- Automatic discovery of MySQL replication clusters and automatic configuration of MHA as well as MHA Helper
- Ability to discover new nodes joining existing MySQL replication clusters and automatic reconfiguration of MHA and MHA Helper
- System configuration related to other MHA prerequisites such as system user creation, SSH setup to allow password-less login to MySQL replication nodes
MHA Helper uses ini-style configuration files.
The Helper expects one configuration file per MySQL replication cluster present in the directory /etc/mha-helper.
The configuration file has a 'default' section and then one section per host for every host in the MySQL replication cluster. The following configuration options are supported:
- writer_vip_cidr
- The virtual IP that is assigned to the MySQL master. This must be in CIDR format.
- vip_type
- The type of VIP which can be anyone of these:
- none : When this is set then MHA Helper does not do VIP management
- metal : When this is set then traditional baremetal-style VIP management is done using the standard ip command
- aws : When this is set then VIP management is done in a way relevant to AWS (Currently not implemented)
- openstack : When this is set then VIP management is done in a way relevant to OpenStack (Currently not implemented)
- super_read_only
- Certain MySQL flavors (such as Percona Server: https://www.percona.com/doc/percona-server/5.6/management/super_read_only.html) have super_read_only which also disallows users with SUPER privileges to perform any writes. Set this to yes to use this feature.
- read_only_config_file
- In case of a restart of a slave, the read_only configuration set by MHA is lost. Define a configuration file (which is read by my.cnf with !includedir, and has permissions for the mha user), to enable creation of this file to ensure a node that restarts remains read_only. Leave empty when you do not want to use this feature.
- report_email
- The email address which receives the email notification when a MySQL failover is performed
- smtp_host
- The SMTP host that is used to send the failover report email
- requires_sudo
- Some of the system commands executed as part of the failover process require either the use of a privileged user or a user with sudo privileges. Set this to no when the system user does not need to execute commands using sudo, set to yes otherwise
- requires_arping
- Some environments do not need to do anything with ARP caches. Set this to no in order to not send those arping commands.
- cluster_interface
- The ethernet interface on the machine that gets the Virtual IP assigned or removed
- kill_after_timeout
- How many seconds do we want to give the application to close MySQL connections gracefully before killing still active connections on the old master. Set this to 0 to disable waiting and kill all connections immediately.
All the options above can be specified either in the default section or in the host specific sections. Values specified in host specific sections override the values specified in the default section.
Let me show you an example configuration file:
[default] requires_sudo = yes requires_arping = yes vip_type = metal writer_vip_cidr = 192.168.10.155/24 cluster_interface = eth1 super_read_only = no read_only_config_file = /etc/my.cnf.d/read_only.cnf report_email = [email protected] smtp_host = localhost kill_after_timeout = 5 [db10] cluster_interface = eth10 [db11] [db12] report_email = [email protected] smtp_host = localhost2 requires_sudo = no
Apart from the configuration file needed by MHA Helper, you also need to setup the MHA specific application configuration file which defines the master-slave hosts. You can find details on how the application configuration file should be written here: https://code.google.com/p/mysql-master-ha/wiki/Configuration#Writing_an_application_configuration_file
I would also suggest that you go through this link to see all the available MHA configuration options: https://code.google.com/p/mysql-master-ha/wiki/Parameters
Following are the important options that must be specified in the MHA application configuration file:
- user
- password
- ssh_user
- manager_workdir
- manager_log
- master_ip_failover_script
- master_ip_online_change_script
- report_script
Below is an example application configuration file:
[server default] user = mha_helper password = helper ssh_user = mha_helper ssh_port = 2202 repl_user = replicator repl_password = replicator master_binlog_dir = /var/log/mysql manager_workdir = /var/log/mha/test_cluster manager_log = /var/log/mha/test_cluster/test_cluster.log remote_workdir = /var/log/mha/test_cluster master_ip_failover_script = /usr/bin/master_ip_hard_failover_helper master_ip_online_change_script = /usr/bin/master_ip_online_failover_helper report_script = /usr/bin/master_failover_report [server1] hostname = db10 candidate_master = 1 check_repl_delay = 0 [server2] hostname = db11 candidate_master = 1 check_repl_delay = 0 [server3] hostname = db12 no_master = 1
To make sure that the failover is safe and does not cause any data inconsistencies, MHA Helper takes the following steps before the failover:
- Set read_only on the new master to avoid any data inconsistencies
- Remove the writer VIP from the original master if vip_type != none
- Set read_only=1 on the original master
- Wait up to 5 seconds for all connected threads to disconnect on the original master
- Terminate all the connections except those that are replication-related, the connection made by MHA Helper and the connections opened by the 'system user'
- Disconnect from the original master
If any of the above steps fail, any changes made during pre-failover are rolled back.
Once MHA has switched the masters and reconfigured replication, the MHA Helper takes the following steps:
- Remove the read_only flag from the new master
- Assign the writer VIP to the new master if vip_type != none
If the original master is accessible via SSH, i.e. in cases where MySQL crashed and stopped but the host is still up, then MHA Helper takes the following step:
- Remove the writer VIP from the original master if vip_type != none
Once MHA has switched the masters and reconfigured replication, the MHA Helper takes the following steps:
- Remove the read_only flag from the new master
- Assign the writer VIP to the new master if vip_type != none
TODO
Once everything is configured and running, doing the failover is pretty simple.
Do a failover when the master db1 goes down:
/usr/bin/mysql_failover -d db1 -c /etc/mha/test_cluster.conf
Do an online failover:
/usr/bin/mysql_online_failover -c /etc/mha/test_cluster.conf
If you are using non-root user to connect to master-slave hosts via ssh (the user that you use for this purpose is taken from the ssh_user option) then you need to make sure that the user can execute the following commands: - /sbin/ip - /sbin/arping
The user should be able to execute the above commands using sudo, and should not have to provide a password. This can accomplished by editing the file /etc/sudoers using visudo and adding the following lines:
mha_helper ALL=NOPASSWD: /sbin/ip, /sbin/arping
In the example above I am assuming that ssh_user=mha_helper.
There are some general recommendations that I want to make, to prevent race-condition that can cause data inconsistencies:
- Do not persist interface with writer VIP in the network scripts. This is important for example in cases where both the candidate masters go down i.e. hosts go down and then come back online. In which case we should need to manually intervene because there is no automated way to find out which MySQL server should be the source of truth
- Persist read_only in the MySQL configuration file of all the candidate masters as well. This is again important for example in cases where both the candidate masters go down. It is possible to use the read_only_config_file setting.