Migrate resources depending on ocf:pacemaker:ping RA with PacemakerThu 07 June 2012 by Frank Lazzarini
The job we want to accomplish is quite simple. Let's just say you have two nodes nodeA and nodeB on two different locations which both have a individual gateways. Now let's imagine the following scenario, gatewayA goes down, and resulting in this there is no more internet connectivity on nodeA, so ideally you would want the cluster to switch to the node which still has internet connectivity. Using the resource agent ping you can create such a rule in pacemaker.
There are 3 different resource agents to accomplish this tasks, but we are going to use just one specific.
ocf:heartbeat:pingd (Deprecated you shouldn't use this one anymore) ocf:pacemaker:pingd --> ocf:pacemaker:ping
First let's create the ping primitive in crm and create a clone of it, so it runs on both nodes. We are going to ping ip addresses outside of the network, so internet ip addresses. For each successful ping result the p_ping primitive will score 1 point which is going to be multiplied by 100. But we'll come back on scoring later on.
# crm crm(live)# primitive p_ping ocf:pacemaker:ping params host_list="184.108.40.206 220.127.116.11" multiplier="100" dampen="5s" op monitor interval="60" timeout="60" op start interval="0" timeout="60" op stop interval="0" timeout="60" crm(live)# clone c_ping p_ping crm(live)# commit
host_list the ip addresses that are going to be pinged separated by a space e.g "x.x.x.x y.y.y.y"
multiplier each successful ping will result in a score value of 1 which is then multiplied by the multiplier value, usually you should use 100 here.
If you look at the cluster now you should see something similar to this.
Online: [ nodeB nodeA ] Master/Slave Set: ms_drbd [p_drbd] Masters: [ nodeB ] Slaves: [ nodeA ] Resource Group: g_cluster p_fs (ocf::heartbeat:Filesystem): Started nodeB p_ip (ocf::heartbeat:IPaddr2): Started nodeB Clone Set: c_ping [p_ping] Started: [ nodeB nodeA ]
Counting the scores
So we have the following scenario, nodeA and nodeB have the p_ping primitive and both can ping the ip addresses 18.104.22.168 22.214.171.124 so both nodes will get 1 score per pingagble address which will result in score 2 * 100 so 200 score for each node.
nodeA (OK) --> 2 * 100 = 200 nodeB (OK) --> 2 * 100 = 200
Now let's cut the internet connection on nodeA which will result in the following scenario. nodeA can't ping any of the ip addresses anymore so it will get a score of 0 which is multiplied by 100 which will result in a score of 0.
nodeA (NOK) --> 0 * 100 = 0 nodeB (OK) --> 2 * 100 = 200
With these scores we can work to get the cluster to automatically failover. Just add a simple location rule which will migrate a resource over to another node. Here I use my group definition as a resource, which consists of all the services I need to have started.
# crm crm(live)# configure crm(live)# location groupwithping g_cluster rule -inf: not_defined pingd or pingd lte 0 crm(live)# commit
To test this you can simply use iptables to block all the icmp traffic on one of the nodes.
iptables -A OUTPUT -p icmp -j DROP
To disable this iptables rule again simply do
Prefer a NodeA
You might want to have a setup where you would like to prefer nodeA to have always all the resources if the network connectivity is ok, there you need to give this node an extra score. This can also be done with a location rule.
# crm crm(live)# configure crm(live)# location prefernodeA resourceA rule 50: \#uname eq nodeA crm(live)# commit