Use keepalived to Establish Master/Slave Clusters

Last updated: 2020-02-26 16:38:39

    This document shows how to build a highly available master/slave cluster in Tencent Cloud VPC with keepalived.


    To provide a clear description of the practical use of keepalived on Tencent Cloud CVMs, this document will:

    1. Give a brief introduction to keepalived and explain the different ways it is used on CVMs and physical networks.
    2. Explain the steps for establishing the modes of non-standing-master mode and standing master/slave mode.
    3. Provide many "keepalived configuration and script files + methods for configuration in different scenarios" to guide the user to put it into practice on CVMs.
    4. The practical guide mainly introduces how to configure keepalived VRRP Instances as unicast VRRP messages.

    Basic principle

    Typically, the high availability master/slave cluster consists of two servers, with the master server being in active status for a service and the slave server being in standby status for the service. The two servers share the same VIP (virtual IP) which is only valid for the master device if there is no failure. In case of the master server failure, the slave server will take over the VIP to continue providing services. The high availability master/slave mode is widely used in MySQL master/slave switchover and Ngnix web access.

    Keepalived on CVMs vs. Keepalived on Physical Networks

    • In traditional physical networks, the master-slave state can be negotiated using the VRRP protocol of keepalived based on the following principles:
      The master device periodically sends free-of-charge ARP messages to purge the MAC table or terminal ARP table of the uplink switch, triggering the VIP migration to the master device.
    • The keepalived can be deployed in Tencent Cloud VPCs to build a high availability master/slave cluster. The difference between this mode and the deployment in physical networks is:
    • The VIP used must be a HAVIP applied for by Tencent Cloud.
    • There is a subnet attribute that can only be bound by announcement of a machine under the same subnet.

    Procedure Overview

    1. Apply for a VIP, which can only be migrated within a subnet (the master and slave servers must be in the same subnet).
    2. Install and configure keepalived ( Version 1.2.24 or above ) on master and slave servers, and modify the configuration files.
    3. Edit the notify mechanism that uses keepalived and do simple log recording using
    4. Verify whether or not the VIP is switched normally when the master/slave switchover occurs.

    Detailed Steps

    Step 1: Apply for VIP

    Apply for VIP.

    Step 2: Install keepalived (**Version 1.2.24 or above) on master and slave CVMs.

    Take CentOS as an example:

    • Using yum installation
      Check whether the "yum list keepalived" version number meets requirements. If so, then go to the next step. If not, then use the method of source package installation with "yum-y install keepalived".
    • Using source package installation
    tar zxvf keepalived-1.2.24.tar.gz
        cd keepalived-1.2.24
        ./configure --prefix=/
        make; make install
        chmod +x /etc/init.d/keepalived **Prevent occurrence of env: /etc/init.d/keepalived: Permission denied**

    Step 3: Confirm master and slave requirements

    This document introduces two modes:

    • Non-standing master mode: The priorities of the two devices to be chosen as the master device are the same.
    • Standing master/slave mode: Keep one of the devices as the master as long as no failure occurs.

    Compared to the non-standing master mode, this mode increases the number of switchovers between the master and the slave. It is recommended to use the non-standing master mode (non-standing master/slave mode, also called the dual slave mode).

    Step 4: Modify the configuration keepalived.conf

    Modifying the configuration file:

    • In the standing master/slave mode, modify keepalived.conf, taking the master device as the example:
           0) state            The initial role. Enter MASTER as the master device, and BACKUP as the slave device.
           1) interface        Change to the ENI name of the local device, such as eth0.
           2) priority         The value of the master is higher than that of the slave, for example, 50 for the master and 30 for the slave. 
           3) unicast_src_ip   Change to the private IP of the local device.
           4) unicast_peer     Change to the private IP of the peer device.
           5) virtual_ipaddress    Change to the private VIP. 
           6) track_interface  Change to the ENI name of the local device, such as eth0.
    • In non-standing master/slave mode, modify keepalived.conf with the same changes made to master and slave devices:
           0) state            The initial role. Enter BACKUP for both master and slave devices.
           1) interface        Change to the ENI name of the local device, such as eth0.
           2) priority         Both devices are configured with the same integer, such as 50.
           3) unicast_src_ip   Change to the private IP of the local device.
           4) unicast_peer     Change to the private IP of the peer device.
           5) virtual_ipaddress    Change to the private VIP.  
           6) track_interface  Change to the ENI name of the local device, such as eth0.

    The practical guide in this document demonstrates the unicast mode, which requires specifying the IP address of the peer device.

    ! Configuration File for keepalived
    global_defs {
       notification_email {
       notification_email_from Alexandre.Cassen@firewall.loc
       smtp_connect_timeout 30
       router_id LVS_DEVEL
       vrrp_garp_interval 0
       vrrp_gna_interval 0
    #vrrp_script checkhaproxy
    #    script "/etc/keepalived/"
    #    interval 5
    vrrp_instance VI_1 {
        # Select proper parameters for the master and slave servers
        # state MASTER            #Master device   #Modification item. "MASTER" is for the master device and "BACKUP" is for the slave device.
    state BACKUP           #Slave device
        interface eth0          #Change it to the ENI name of the local device, such as eth0.  
        virtual_router_id 51
        nopreempt                   #Non-preemptive mode
    #    preempt_delay 10
        priority 30             #The priority of the master should be greater than that of the slave, for example, 50 for the master and 30 for the slave.
        advert_int 1        
        authentication {
            auth_type PASS
            auth_pass 1111
        unicast_src_ip   #Private IP of local device
        unicast_peer {
             #IP address of the peer device, for example:
        virtual_ipaddress {
            #The private network VIP 
        notify_master "/etc/keepalived/ MASTER"
        notify_backup "/etc/keepalived/ BACKUP"
        notify_fault "/etc/keepalived/ FAULT"
        notify_stop "/etc/keepalived/ STOP"
        garp_master_delay 1
        garp_master_refresh 5
            track_interface {
                    eth0                #Change it to the ENI name of the local device, such as eth0.
    #    track_script {
    #        checkhaproxy 
    #    }

    Step 5: Use for simple log recording

        Modify for the standing-master/slave mode:
            1) N/A
        Modify for the non-standing-master/slave mode:
            1) N/A
        keepalived main log is still recorded in /var/log/message
        echo "[`date '+%Y-%m-%d %T'`] $1" >> $log_file
    [ ! -d /var/keepalived/ ] && mkdir -p /var/keepalived/
    case "$1" in
        "MASTER" )
            echo -n "$1" > /var/keepalived/state
            log_write " notify_master"
            echo -n "0" > /var/keepalived/vip_check_failed_count
        "BACKUP" )
            echo -n "$1" > /var/keepalived/state
            log_write " notify_backup"
        "FAULT" )
            echo -n "$1" > /var/keepalived/state
            log_write " notify_fault"
        "STOP" )
            echo -n "$1" > /var/keepalived/state
            log_write " notify_stop"
            log_write " STATE ERROR!!!"

    Step 6: Scenario in which primary IPs of local master and slave CVMs do not have public IPs

    The CVM or its ENI does not require a public IP.

    Step 7: Verify whether the VIP and public IP are switched normally during master/slave switchover.

    1. Enable keepalived: /etc/init.d/keepalived start Or systemctl start keepalived Or service keepalived start
    2. Verify the disaster recovery capability of master/slave switchover: Simulate the CVM failure by restarting the keepalived process or restarting the CVM to check whether the VIP can be migrated. The corresponding logs will be written in /var/log/keepalived.log . You can also view the interval from the network suspension to recovery by pinging VIP.
    • For each switchover, the time of ping suspension should be about 4 seconds. It may take 6 seconds if it is in standing master/slave mode. This type of situation usually occurs when the time of master "failure" is Very short , and the master/slave switchover may occur twice for a short time. Then, the VIP will be re-migrated to the old master server that was just recovered.
    • The script log will be written to "/ var/log/keealived.log" and takes up your disk space. You can clear the accumulated logs with logrotate and other tools. The log related to the keepalived process will be written to "/ var/log/message".


    Each cluster uses a unique VRRP router ID in the VPC

    MAC address conflict will be caused by different clusters using the Same Router ID when in VMAC mode, and this could lead to a communications exception. The reasons for this are:

    • Tencent Cloud VPC provides Multicast ability .
    • Tencent Cloud's multicast domain is the entire VPC, and if the ENIs of different subnets are added into the same multicast group, messages within the group can be received.
    • The default multicast mode of keepalived is to use a fixed multicast address and identify different clusters using the IDs in the VRRP messages.
      -If different clusters use the same router ID, then the messages of different masters will interfere with each other and lead to some clusters having no master.
    • The router ID will be used to generate the MAC addresses of VMAC devices, and Tencent Cloud requires that the MAC addresses of VPC private network card devices be unique.
    • Therefore, when using VMAC mode, having different clusters with the same router ID will cause MAC address conflict, leading to communications exceptions.

    How to use multicast for VRRP communications

    • Submit a ticket to apply for the multicast feature, which will automatically turn on the multicast switch of the VPC for which multicast functionality is desired.
    • Do not configure unicast_peer In the keepalived configuration file.

    Recommendation to use VMAC device

    It is recommend that VMAC mode be used for the following reason:
    When keepalived is running, if starting and stopping of network subsystems occurs in a CVM, keepalived may configure the HAVIP earlier than the network subsystem to the ENI so that the virtual IP becomes the primary IP of the ENI. Whether unicast or multicast mode is being used, the subsequent VRRP messages will use the VIP as the source IP for delivery. The keepalived processes in other CVMs will ignore this announced message, causing split brain.

    Controlling the number of VIPs configured in an individual ENI

    • To allow virtual IPs to switch more smoothly, the Tencent Cloud platform sends a free-of-charge ARP announcement to each ENI, and the frequency of VIPs will be limited to a certain degree.
    • We recommend that the number of HAVIPs bound to a single EIP not exceed 5, or else the switching latency of some IPs could increase.
    • If there is a need to use multiple VIPs, we recommend that the configuration vrrp_garp_master_repeat 1 In the keepalived configuration file's global_defs section be added or modified.

    How to use it in the situation of having multiple ENIs

    • For CVMs with multiple ENIs, it is recommended that the unicast mode be used to configure keepalived.
    • If multicast mode is used, it is recommended that the ENI on the default route be used, or else the VRRP multicast messages may encounter exceptions when sent and received by other ENIs.

    Was this page helpful?

    Was this page helpful?

    • Not at all
    • Not very helpful
    • Somewhat helpful
    • Very helpful
    • Extremely helpful
    Send Feedback