[ClusterLabs] Can Pacemaker monitor geographical separated servers
Jason A Ramsey
jason at eramsey.org
Wed Aug 10 10:43:06 EDT 2016
I can’t answer all of this because I’m still working out how to do fencing, but I’ve been setting up a Pacemaker cluster in Amazon Web Services across two separate availability zones. Naturally, this means that I have to bridge subnets, so I’ve battled through a good bit of this already.
Imagine that you have a cluster node in each of two IP subnets: 10.100.0.0/24 and 10.200.0.0/24. This configuration prevents you from doing two things:
1. You can’t use multicast
2. You can’t pick an IP in either subnet as the cluster VIP
The way that I got around this was to pick an arbitrary subnet that exists _outside_ of all configured subnets in my environment: 10.0.0.0/24. I then created routes from each of my cluster node subnets to VIP(s) (I’m trying to make my cluster Active/Active, so I want 2) on this subnet:
10.0.0.100/32 network interface for cluster node 1
10.0.0.101/32 network interface for cluster node 2
I then set up Pacemaker cluster resources for the VIPs:
pcs resource create cluster_vip1 ocf:heartbeat:IPAddr2 ip=10.0.0.100 cidr_netmask=32 nic=eth0 op monitor interval=15s
pcs resource create cluster_vip2 ocf:heartbeat:IPAddr2 ip=10.0.0.101 cidr_netmask=32 nic=eth0 op monitor interval=15s
The voodoo in this is that you specify the device name of the network interface that you’re mapping to rather than just the IP address. Otherwise Pacemaker will throw an error about how 10.0.0.100 isn’t an address that exists on the cluster nodes. Then you need to make sure that the right VIP is running on the right cluster node by probably moving resources around.
pcs resource move cluster_vip1 clusternode1
pcs resource move cluster_vip2 clusternode2
At this point, if everything is working properly, you should be able to ping (assuming no firewall rules are in the way) the VIP IPs as long as they are associated with the appropriate node and the routes are correct. You’ll find that if you move the VIP to another node without updating the routing table, the pings will no longer work. Success! Well, almost…
While I know logically that to make this work I need to sort out a fencing method that detects node failure and then have a fencing script that updates the routing table as appropriate to get the traffic moving around properly to the failover node in the event of a…well…event… ☺
sudo -u hacluster aws ec2 replace-route --route-table-id rtb-99999999 --destination-cidr-block 10.0.0.100/32 --network-interface-id eni-11111111
pcs resource move cluster_vip1 clusternode2
Obviously, this would only work in an AWS deployment, and, like I said, I still haven’t figured out how to detect an outage to make this failover occur. Hopefully, though, this should get you pointed in the right direction.
[ jR ]
M: +1 (703) 628-2621
@: jason at eramsey.org<mailto:jason at eramsey.org>
there is no path to greatness; greatness is the path
From: "bhargav M.P" <mpbhargav at gmail.com>
Reply-To: Cluster Labs - All topics related to open-source clustering welcomed <users at clusterlabs.org>
Date: Tuesday, August 9, 2016 at 2:40 PM
To: "users at clusterlabs.org" <users at clusterlabs.org>
Subject: [ClusterLabs] Can Pacemaker monitor geographical separated servers
I have deployment where we have two Linux servers that are geographically separated and they are across different subnets . I want the server to work in Active/Standby mode . I would like to use pacemaker-corsync for performing switch over when active fails.
would like to have a single virtual IP address for accessing the Linux servers and only active must have VIP(virtual IP address)
Can pacemaker transfer the virtual IP address to new active when the current active fails.? If so, how can the same virtual IP address be accessible from the client since it has now moved to a different subnet.?
If the above use case cannot be supported by the pacemaker what are the possible options i need to look at.?
Thank you so much for the help,
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Users