Buy Percona ServicesBuy Now!

Setting up PXC reference architecture with HAProxy

This manual describes how to set up Percona XtraDB Cluster in a virtualized test sandbox.

The procedure assumes Amazon EC2 micro instances running CentOS 6. However, it should apply to any virtualization technology (for example, VirtualBox) with any Linux distribution.

This manual requires three virtual machines for Percona XtraDB Cluster nodes, and one for HAProxy client, which redirects requests to the nodes. Running HAProxy on an application server, instead of having it as a dedicated entity, removes the unnecessary extra network roundtrip, because the load balancing layer in Percona XtraDB Cluster scales well with application servers.

  1. Install Percona XtraDB Cluster on the three cluster nodes, as described in Installing Percona XtraDB Cluster on Red Hat Enterprise Linux and CentOS.

  2. Install HAProxy and sysbench on the client node:

    yum -y install haproxy sysbench
  3. Make sure that the my.cnf configuration file on the first node contains the following:

  4. Start the first node

  5. Adjust the my.cnf configuration files on the second and third nodes to contain the same configuration settings, except the following:

    • Second node:

    • Third node:



    • server_id can be any unique number
    • wsrep_cluster_address is the IP address of the first node
    • wsrep_node_name can be any unique name, for example, the output of the hostname command
  6. Start the second and third nodes.

    When a new node joins the cluster, SST is performed by taking a backup using XtraBackup, then copying it to the new node with netcat. After a successful SST, you should see the following in the error log:

    120619 13:20:17 [Note] WSREP: State transfer required:
         Group state: 77c9da88-b965-11e1-0800-ea53b7b12451:97
         Local state: 00000000-0000-0000-0000-000000000000:-1
    120619 13:20:17 [Note] WSREP: New cluster view: global state: 77c9da88-b965-11e1-0800-ea53b7b12451:97, view# 18: Primary, number of nodes: 3, my index: 0, protocol version 2
    120619 13:20:17 [Warning] WSREP: Gap in state sequence. Need state transfer.
    120619 13:20:19 [Note] WSREP: Running: 'wsrep_sst_xtrabackup 'joiner' '' '' '/var/lib/mysql/' '/etc/my.cnf' '20758' 2>sst.err'
    120619 13:20:19 [Note] WSREP: Prepared |SST| request: xtrabackup|
    120619 13:20:19 [Note] WSREP: wsrep_notify_cmd is not defined, skipping notification.
    120619 13:20:19 [Note] WSREP: Assign initial position for certification: 97, protocol version: 2
    120619 13:20:19 [Warning] WSREP: Failed to prepare for incremental state transfer: Local state UUID (00000000-0000-0000-0000-000000000000) does not match group state UUID (77c9da88-b965-11e1-0800-ea53b7b12451): 1 (Operation not permitted)
          at galera/src/replicator_str.cpp:prepare_for_IST():439. IST will be unavailable.
    120619 13:20:19 [Note] WSREP: Node 0 (ip-10-244-33-92) requested state transfer from '*any*'. Selected 1 (ip-10-112-39-98)(SYNCED) as donor.
    120619 13:20:19 [Note] WSREP: Shifting PRIMARY -> JOINER (TO: 102)
    120619 13:20:19 [Note] WSREP: Requesting state transfer: success, donor: 1
    120619 13:20:59 [Note] WSREP: 1 (ip-10-112-39-98): State transfer to 0 (ip-10-244-33-92) complete.
    120619 13:20:59 [Note] WSREP: Member 1 (ip-10-112-39-98) synced with group.
    120619 13:21:17 [Note] WSREP: |SST| complete, seqno: 105
    120619 13:21:17 [Note] Plugin 'FEDERATED' is disabled.
    120619 13:21:17 InnoDB: The InnoDB memory heap is disabled
    120619 13:21:17 InnoDB: Mutexes and rw_locks use GCC atomic builtins
    120619 13:21:17 InnoDB: Compressed tables use zlib 1.2.3
    120619 13:21:17 InnoDB: Using Linux native AIO
    120619 13:21:17 InnoDB: Initializing buffer pool, size = 400.0M
    120619 13:21:17 InnoDB: Completed initialization of buffer pool
    120619 13:21:18 InnoDB: highest supported file format is Barracuda.
    120619 13:21:18  InnoDB: Waiting for the background threads to start
    120619 13:21:19 Percona XtraDB ( 1.1.8-rel25.3 started; log sequence number 246661644
    120619 13:21:19 [Note] Recovering after a crash using mysql-bin
    120619 13:21:19 [Note] Starting crash recovery...
    120619 13:21:19 [Note] Crash recovery finished.
    120619 13:21:19 [Note] Server hostname (bind-address): '(null)'; port: 3306
    120619 13:21:19 [Note]   - '(null)' resolves to '';
    120619 13:21:19 [Note]   - '(null)' resolves to '::';
    120619 13:21:19 [Note] Server socket created on IP: ''.
    120619 13:21:19 [Note] Event Scheduler: Loaded 0 events
    120619 13:21:19 [Note] WSREP: Signalling provider to continue.
    120619 13:21:19 [Note] WSREP: Received |SST|: 77c9da88-b965-11e1-0800-ea53b7b12451:105
    120619 13:21:19 [Note] WSREP: |SST| received: 77c9da88-b965-11e1-0800-ea53b7b12451:105
    120619 13:21:19 [Note] WSREP: 0 (ip-10-244-33-92): State transfer from 1 (ip-10-112-39-98) complete.
    120619 13:21:19 [Note] WSREP: Shifting JOINER -> JOINED (TO: 105)
    120619 13:21:19 [Note] /usr/sbin/mysqld: ready for connections.
    Version: '5.5.24-log'  socket: '/var/lib/mysql/mysql.sock'  port: 3306  Percona XtraDB Cluster (GPL), wsrep_23.6.r340
    120619 13:21:19 [Note] WSREP: Member 0 (ip-10-244-33-92) synced with group.
    120619 13:21:19 [Note] WSREP: Shifting JOINED -> SYNCED (TO: 105)
    120619 13:21:20 [Note] WSREP: Synchronized with group, ready for connections

    For debugging information about the SST, you can check the sst.err file and the error log.

    After SST finishes, you can check the cluster size as follows:

    mysql> show global status like 'wsrep_cluster_size';
    | Variable_name      | Value |
    | wsrep_cluster_size | 3     |
    1 row in set (0.00 sec)
  7. When all cluster nodes are started, configure HAProxy on the client node. This will enable the application to connect to localhost as if it were a single MySQL server, instead of a Percona XtraDB Cluster node.

    You can configure HAProxy to connect and write to all cluster nodes or to one node at a time. The former method can lead to rollbacks due to conflicting writes when optimistic locking at commit time is triggered, while the latter method avoids rollbacks.

    However, most good applications should be able to handle rollbacks, so either method is fine in this case.

    To configure HAProxy, add the following to /etc/haproxy/haproxy.cfg:

    log local0
    log local1 notice
    maxconn 4096
    chroot /usr/share/haproxy
    user haproxy
    group haproxy
    log global
    mode http
    option tcplog
    option dontlognull
    retries 3
    option redispatch
    maxconn 2000
    contimeout 5000
    clitimeout 50000
    srvtimeout 50000
    frontend pxc-front
    bind *:3307
    mode tcp
    default_backend pxc-back
    frontend stats-front
    bind *:80
    mode http
    default_backend stats-back
    frontend pxc-onenode-front
    bind *:3306
    mode tcp
    default_backend pxc-onenode-back
    backend pxc-back
    mode tcp
    balance leastconn
    option httpchk
    server c1 check port 9200 inter 12000 rise 3 fall 3
    server c2 check port 9200 inter 12000 rise 3 fall 3
    server c3 check port 9200 inter 12000 rise 3 fall 3
    backend stats-back
    mode http
    balance roundrobin
    stats uri /haproxy/stats
    stats auth pxcstats:secret
    backend pxc-onenode-back
    mode tcp
    balance leastconn
    option httpchk
    server c1 check port 9200 inter 12000 rise 3 fall 3
    server c2 check port 9200 inter 12000 rise 3 fall 3 backup
    server c3 check port 9200 inter 12000 rise 3 fall 3 backup

    In this configuration, three frontend-backend pairs are defined:

    • The stats pair is for HAProxy statistics page (port 80).

      You can access it at /haproxy/stats using the credential specified in the stats auth parameter.

    • The pxc pair is for connecting to all three nodes (port 3307).

      In this case, the leastconn load balancing method is used, instead of round-robin, which means connection is made to the backend with the least connections established.

    • The pxc-onenode pair is for connecting to one node at a time (port 3306) to avoid rollbacks because of optimistic locking.

      If the node goes offline, HAProxy will connect to another one.


    MySQL is checked via httpchk. MySQL will not serve these requests by default. You have to set up the clustercheck utility, which is distributed with Percona XtraDB Cluster. This will enable HAProxy to check MySQL via HTTP.

    The clustercheck script is a simple shell script that accepts HTTP requests and checks the node via the wsrep_local_state variable. If the node’s status is fine, it will send a response with HTTP code 200 OK. Otherwise, it sends 503.

    To create the clustercheck user, run the following:

    mysql> grant process on *.* to 'clustercheckuser'@'localhost' identified by 'clustercheckpassword!';
    Query OK, 0 rows affected (0.00 sec)
    mysql> flush privileges;
    Query OK, 0 rows affected (0.00 sec)

    If you want to use a different user name or password, you have to modify them in the clustercheck script.

    If you run the script on a running node, you should see the following:

    # clustercheck
    HTTP/1.1 200 OK
    Content-Type: Content-Type: text/plain

    You can use xinetd to daemonize the script. If xinetd is not installed, you can install it with yum:

    # yum -y install xinetd

    The service is configured in /etc/xinetd.d/mysqlchk:

    # default: on
    # description: mysqlchk
    service mysqlchk
    # this is a config for xinetd, place it in /etc/xinetd.d/
    disable = no
    flags = REUSE
    socket_type = stream
    port = 9200
    wait = no
    user = nobody
    server = /usr/bin/clustercheck
    log_on_failure += USERID
    only_from =
    # recommended to put the IPs that need
    # to connect exclusively (security purposes)
    per_source = UNLIMITED

    Add the new service to /etc/services:

    mysqlchk 9200/tcp # mysqlchk

    Clustercheck will now listen on port 9200 after xinetd restarts and HAProxy is ready to check MySQL via HTTP:

    # service xinetd restart

If you did everything correctly, the statistics page for HAProxy should look like this:


Testing the cluster with sysbench

After you set up Percona XtraDB Cluster in a sand box, you can test it using sysbench. This example shows how to do it with sysbench from the EPEL repository.

  1. Create a database and a user for sysbench:

    mysql> create database sbtest;
    Query OK, 1 row affected (0.01 sec)
    mysql> grant all on sbtest.* to 'sbtest'@'%' identified by 'sbpass';
    Query OK, 0 rows affected (0.00 sec)
    mysql> flush privileges;
    Query OK, 0 rows affected (0.00 sec)
  2. Populate the table with data for the benchmark:

    sysbench --test=oltp --db-driver=mysql --mysql-engine-trx=yes --mysql-table-engine=innodb --mysql-host= --mysql-port=3307 --mysql-user=sbtest --mysql-password=sbpass --oltp-table-size=10000 prepare
  1. Run the benchmark on port 3307:

    sysbench --test=oltp --db-driver=mysql --mysql-engine-trx=yes --mysql-table-engine=innodb --mysql-host= --mysql-port=3307 --mysql-user=sbtest --mysql-password=sbpass --oltp-table-size=10000 --num-threads=8 run

    You should see the following in HAProxy statistics for pxc-back:


    Note the Cur column under Session:

    • c1 has 2 threads connected
    • c2 and c3 have 3 threads connected
  2. Run the same benchmark on port 3306:

    sysbench --test=oltp --db-driver=mysql --mysql-engine-trx=yes --mysql-table-engine=innodb --mysql-host= --mysql-port=3306 --mysql-user=sbtest --mysql-password=sbpass --oltp-table-size=10000 --num-threads=8 run

    You should see the following in HAProxy statistics for pxc-onenode-back:


    All 8 threads are connected to the c1 server. c2 and c3 are acting as backup nodes.

If you are using HAProxy for MySQL you can break the privilege system’s host part, because MySQL will think that the connections are always coming from the load balancer. You can work this around using T-Proxy patches and some iptables magic for the backwards connections. However in the setup described in this how-to this is not an issue, since each application server has it’s own HAProxy instance, each application server connects to, so MySQL will see that connections are coming from the application servers. Just like in the normal case.

Visit Percona Store

General Inquiries

For general inquiries, please send us your question and someone will contact you.