Corosync and Pacemaker in Slackware

This will be multi part post about high availability solution for Slackware. My first post will be about Corosync and Pacemaker.

You need to combine Corosync and Pacemaker with other distributed storage system such as DRBD/OCFS2/GFS. I’ll talk about these stacks in another post.

Goal:

Environments:

The MySQL data is not syncronized, this post is just about Corosync and Pacemaker.

Guides:

  1. Download and install these packages (by this order) in both nodes:

    I strongly suggest you build these packages one by one just to be sure there are no missing dependencies. BTW, some script adjustments are needed for Cluster Resource Agents but I’m sure you guys can handle it ;-)

  2. It would be easier for the next steps if password-less login with OpenSSH is enabled. In your Node 1:

    ssh-keygen -t rsa
    ssh-copy-id -i ~/.ssh/id_rsa.pub [email protected]
  3. Generate an authentication key for Corosync:

    corosync-keygen

    If you’re connecting remotely, pressing your keyboard won’t do any good. The fastest way would be typing directly into the server. The other way is running find . on your / directory (press Control + C when the key has been generated).

  4. Copy the new generated authentication key to Node 2:

    scp /etc/corosync/authkey 192.168.1.102:/etc/corosync
  5. Copy the default corosync configuration file:

    cp /etc/corosync/corosync.conf.example /etc/corosync/corosync.conf
  6. Replace bindnetaddr and logfile (optional):

    bindnetaddr: 192.168.1.0
    logfile: /var/log/corosync

    You can checkout the reference about those values. From corosync:

    If the local interface was 10.12.12.93 and the netmask was 255.0.0.0, Totem would execute the logical operation 10.12.12.93 & 255.0.0.0 and produce the value 10.0.0.0. This value would be compared against bindnetaddr and bind Totem to the NIC that matches.
    This can cause confusion if netmask or bindnetaddr are not set properly.
    In the example above, if bindnetaddr is 10.12.12.0, the network interface will never be matched. If bindnetaddr is 10.0.0.0 the interface will be matched.
  7. Copy corosync.conf to Node 2:

    scp /etc/corosync/corosync.conf 192.168.1.102:/etc/corosync
  8. Create pacemaker file so that Corosync will automatically load Pacemaker when it’s started:

    touch /etc/corosync/service.d/pacemaker

    Put these configs in that file:

    service {
      # Load the Pacemaker Cluster Resource Manager
      name: pacemaker
      ver:  0
    }
  9. Copy the pacemaker file to Node 2:

    scp /etc/corosync/service.d/pacemaker 192.168.1.102:/etc/corosync/service.d/
  10. Start your Corosync and let the magic begins:

    /etc/rc.d/rc.corosync start
  11. Check your log for any error:

    tail -f /var/log/corosync

    Check your process list:

    ps auxf

    Corosync should also load other processes automatically:

    root      2008  0.5  3.4  52668  3964 ?        Ssl  13:55   0:00 corosync
    root      2015  0.0  1.9  12140  2248 ?        S    13:55   0:00  \_ /usr/lib/heartbeat/stonithd
    226       2016  0.3  3.3  13004  3796 ?        S    13:55   0:00  \_ /usr/lib/heartbeat/cib
    root      2017  0.0  1.6   6812  1848 ?        S    13:55   0:00  \_ /usr/lib/heartbeat/lrmd
    226       2018  0.1  2.2  12404  2540 ?        S    13:55   0:00  \_ /usr/lib/heartbeat/attrd
    226       2019  0.0  1.7   8664  2032 ?        S    13:55   0:00  \_ /usr/lib/heartbeat/pengine
    226       2020  0.1  2.5  12528  2904 ?        S    13:55   0:00  \_ /usr/lib/heartbeat/crmd
  12. Monitor your cluster using Pacemaker tools:

    crm status

    It should be something like this:

    ============
    Last updated: Sun May 13 13:57:43 2012
    Stack: openais
    Current DC: node1 - partition with quorum
    Version: 1.1.1-b9b672590e79770afb63b9b455400d92fb6b5d9e
    2 Nodes configured, 2 expected votes
    0 Resources configured.
    ============
    
    Online: [ node1 node2 ]

    Give them some time to online if they’re offline.

  13. Put some main configurations to your cluster:

    crm
    configure
    property stonith-enabled=false
    property no-quorum-policy=ignore
    commit
    quit

    If you’re getting some errors such as ERROR: cib-bootstrap-options: attribute last-lrm-refresh does not exist, just proceed. It maybe a bug.

    We had to disable stonith since we just want our Pacemaker to be running. However, in real production environment, you really need to configure stonith, you can read more about it here.

    We also need to ignore quorum policy since we’re only using 2 nodes and you can read more about it here.

    You can see your new configuration by running:

    	crm configure show

    Which will output:

    	node node1
    	node node2
    	property $id="cib-bootstrap-options" \
    		dc-version="1.1.1-b9b672590e79770afb63b9b455400d92fb6b5d9e" \
    		cluster-infrastructure="openais" \
    		expected-quorum-votes="2" \
    		stonith-enabled="false" \
    		last-lrm-refresh="1336919205" \
    		no-quorum-policy="ignore"

    If you accidentally put some wrong configurations and don’t know how to edit it, you can use crm configure edit to change your configurations directly but this method is highly not recommended since it’s error-prone.

  14. It’s time to configure our main/failover/cluster IP (our client will use this IP, not the nodes IP):

    crm
    configure
    primitive ip ocf:heartbeat:IPaddr params ip="192.168.1.100" op monitor interval=10s
    commit
  15. If everyting goes well, you should be able to ping the cluster IP (192.168.1.100) and crm status should yield this result:

    ============
    Last updated: Sun May 13 14:28:19 2012
    Stack: openais
    Current DC: node1 - partition with quorum
    Version: 1.1.1-b9b672590e79770afb63b9b455400d92fb6b5d9e
    2 Nodes configured, 2 expected votes
    1 Resources configured.
    ============
    
    Online: [ node1 node2 ]
    
    ip     (ocf::heartbeat:IPaddr):        Started node1
  16. We’ll now setup MySQL monitoring with Pacemaker. But before that, make sure you:

    Installed MySQL in both of the nodes.

    Able to connect to your MySQL from other than localhost:

    	mysql -u root -p -h 192.168.1.101
    	mysql -u root -p -h 192.168.1.102

    You can use this command to allow any host to connect to your MySQL:

    	GRANT ALL PRIVILEGES ON *.* TO 'root'@'%' IDENTIFIED BY 'password' WITH GRANT OPTION;
    	FLUSH PRIVILEGES;

    Created a database in Node 1 and Node 2. For an example, a databased named node1 in Node 1 and node2 in Node 2. This is just for verification.

  17. Add this resource:

    crm
    configure
    primitive mysql ocf:heartbeat:mysql \
    params binary="/usr/bin/mysqld_safe" config="/etc/my.cnf" user="mysql" pid="/var/run/mysql/mysql.pid" datadir="/var/lib/mysql" socket="/var/run/mysql/mysql.sock" \
    op monitor interval="30s" timeout="30s" \
    op start interval="0" timeout="120" \
    op stop interval="0" timeout="120"
    commit
    quit

    The parameter above is purely based on the standard Slackware’s MySQL package. So make sure you’ve created /etc/my.cnf which is not available by default. Just copy from the default file:

    cp /etc/my-small.cnf /etc/my.cnf
  18. Your latest crm status would show something like this:

    ============
    Last updated: Mon May 14 01:13:23 2012
    Stack: openais
    Current DC: node1 - partition with quorum
    Version: 1.1.1-b9b672590e79770afb63b9b455400d92fb6b5d9e
    2 Nodes configured, 2 expected votes
    2 Resources configured.
    ============
    
    Online: [ node1 node2 ]
    
     ip	(ocf::heartbeat:IPaddr):	Started node1
     mysql	(ocf::heartbeat:mysql):	Started node2

    As you can see, mysql has been started on Node 2. Actually it doesn’t matter in which node it will start first (for this tutorial, not for the production server), what important is that if one of the nodes is down, the other node should start its MySQL automatically. You can test this situation by running these commands in your Node 2 to simulate a node failure:

    crm
    node
    standby
    quit

    crm status would show something like this (give Node 1 some time before it starts its MySQL):

    ============
    Last updated: Mon May 14 01:21:12 2012
    Stack: openais
    Current DC: node1 - partition with quorum
    Version: 1.1.1-b9b672590e79770afb63b9b455400d92fb6b5d9e
    2 Nodes configured, 2 expected votes
    2 Resources configured.
    ============
    
    Node node2: standby
    Online: [ node1 ]
    
     ip	(ocf::heartbeat:IPaddr):	Started node1
     mysql	(ocf::heartbeat:mysql):	Started node1

    Right now, your client can use the cluster IP (192.168.1.100) to connect to your MySQL. The client won’t realize which node it connected to. In this case, he/she will connect to Node 2 if both of them (the nodes) are online. If Node 2 is offline, 192.168.1.100 will automatically connect the client to MySQL in 192.168.1.101. If Node 1 is offline, 192.168.1.100 will automatically uses MySQL in Node 2 which is in 192.168.1.102.

    To reonline Node 2, just use these commands in your Node 2:

    crm
    node
    online
    quit
  19. However, usually you want to control which MySQL will be up first, either in Node 1 or in Node 2. To make this happen, you need to use colocation:

    crm
    configure
    colocation ip-mysql inf: ip mysql
    commit
    quit

    crm status would show something like this:

    ============
    Last updated: Mon May 14 01:26:41 2012
    Stack: openais
    Current DC: node1 - partition with quorum
    Version: 1.1.1-b9b672590e79770afb63b9b455400d92fb6b5d9e
    2 Nodes configured, 2 expected votes
    2 Resources configured.
    ============
    
    Online: [ node1 node2 ]
    
     ip	(ocf::heartbeat:IPaddr):	Started node1
     mysql	(ocf::heartbeat:mysql):	Started node1

    That means, your mysql has been started on Node 1. So, everytime corosync is started on both of the nodes, mysql will be started on Node 1 due to the colocation configuration.

  20. Try turning off Node 1 or Node 2 and see how MySQL switches side from both of the nodes.

I think that’s it, next tutorial should be mainly about DRBD. Good luck!



Powered by Jekyll. Get the RSS or the source. Design copied from Sam Vermette.