Difference between revisions of "Simple IP Failover with Heartbeat"

From ArchWiki
Jump to: navigation, search
(build a VERY SIMPLE Active/Passive High Availability Heartbeat FailOver cluster using haresources file, without pacemaker!)
Line 78: Line 78:
<br>Please feel free to contact me in #archlinux @freenode if you need any assistance, and PLEASE(for *** sake) update/spell check/and fix the formatting in this article if you have some spare time!!! (I don't right now :P).
<br>Please feel free to contact me in #archlinux @freenode if you need any assistance, and PLEASE(for *** sake) update/spell check/and fix the formatting in this article if you have some spare time!!! (I don't right now :P).
<br>--Sully O,
<br>--[[User:Sulucohmun|Sulucohmun]] 23:07, 11 August 2011 (EDT)Sully O. / Suluc Ohmun.
--Suluc Ohmun.

Revision as of 03:07, 12 August 2011

This article illustrates a method of implementing VERY simple active/passive IP failover using heartbeat(Available in AUR). I intend this to be very short and to the point, as such I will not be discussing any of the core concepts/principles of the individual compononents of a High Availability Linux Cluster--I will assume you have at-least a basic understanding of what a High Availability cluster is, and the individual components that come together to provide High Availability using heartbeat. If you have any trouble understanding anything in this article--I HIGHLY reccomended that you first familiarize yourself with the concepts of High Availability Clustering in Linux before proceeding to implement these instructions in a live/production environment.(SEE: Linux-HA)

For the purposes of this article we will not be configuring pacemaker, we will be using the older style haresources file/method to define our Highly Available Services with heartbeat.

We will NOT be using a load balancer, or any external resource agents with heartbeat; because of this, this setup will only allow for a 2 node ACTIVE/PASSIVE cluster, and we will be using two PHYSICAL machines. You may however try this on two virtual machines/hosts to test first, I see no reason why this setup wouldn't work in a virtual machine/environment.

We will have two machines, and at-least 3 IP addresses, for my setup I have 3 publically accessible/WAN IP Adresses, but this may also be done using two internal/LAN IP's & a single(1) Public/WAN IP address.

The machines we will be using are as follows:

  • ha1.example.com: (WAN)
  • node1.example.com (WAN OR LAN)
  • node2.example.com (WAN OR LAN)

* & (the IP's our two heartbeat nodes will be using to communicate with eachother over our Local Area Network, these may be public(WAN) or private(LAN) addresses. Each of these two nodes should preferably be on the same subnet but all that is needed is that each node is able to communicate with the other/vice versa.)

* (our VIRTUAL IP Address that the two node's will "share" & monitor/bring alive if one node should stop communicating, this IP Address may be on ANY subnet. This IP Address should be reachable from the public internet(WAN) if you plan for your Highly Available services to be reachable from outside of your private Local Area Network (we will be using this ip address to connect to our Highly Available Services on the 'currently' active node))

*It is NOT neccesary to have a genuine registered domain name (FQDN) or a DNS server for the purposes of these instructions, heartbeat will use our /etc/hosts file for all heartbeat related hostname/domain name lookups locally(Irregardless of what /etc/host.conf tells it to use.); but, if you would like to reach your highly available services via a domain name from outside your Local Area Network, you will have to register a domain and fix it up with the proper zone/ns/mx/A record/CNAME definitions, and, optionally install/run ICS Bind with the proper/relevant zone definitions & replicate those definitions in /etc/hosts.

Cross your fingers, here we go!

01. Install the neccesary packages(Available in AUR):

yaourt -S ha-glue ha-heartbeat ha-pacemaker ha-resourceagent inetutils net-tools

Note: on inetutils/net-tools: heartbeat seems to still use ifconfig/telnet for SOME things.) also the order in which we install these components seems to be crucial, I believe this is the proper order but if you get any dependency issues try switching up the order that these are installed. :P

02. Edit the main configuration file for heartbeat /etc/ha.d/ha.cf and make it look like this:

deadtime 5 warntime 10 initdead 15 bcast eth0 auto_failback on node node1 node node2 use_logd yes

Note: 'auto_failback on' will tell heartbeat to prefer node1. If node1 should go down, node2 will take over our Virtual IP Address and start our Highly Available services. When/if node1 comes back alive node2 will transfer(automatically failback) our Virtual IP and Highly Available services back to node1 and stop our H.A services on node 2. If auto_failback off is used, in the previous scenario, node2 would not give our virtual IP and services back to node1 when node1 comes back alive--node1 would remain inactive and node2 would continue to serve our virtual IP/HA services until node2 eventually fails and node1 takes over for it...auto_failback on provides for more session disconnects/connection drops from a user perspective but this may be a neccesary option for you if you do not have shared storage(E.G For user email, NFS/shared storage, etc)... For example, you want to also serve Highly Available NFS exports(data) using your VIRTUAL IP and all of your data is on node1--if auto_failback off is used and node1 fails, users will browse to their NFS share and not see any of the data that exists on node1's NFS shares!
Note: if you want to use auto_failback off & still share NFS exports/etc this can be overcome by using something like rsync, fam & ssh/scp to frequently copy over (only the new data/see FAM) and keep the two nodes in sync, one could also use DRBD on both nodes to mirror two partitions over ethernet--essentially raid1 over ethernet. DRBD(Avail. in AUR) is the most commonly used tool to solve this problem, if you choose to use drbd for this please note that you want to set your two drbd partitions secondary/secondary(initially) and have heartbeat/haresources file mount your drbd partition as primary on the 'currently' active node ONLY, I'd also HIGHLY reccomend using a cluster friendly FileSystem like GFS2 or OCFS2 for a shared drbd device/partition! Another possible solution is to use pNFS(if you choose to use NFS to share data that will be simultaneously accessed/written to/altered on/by both nodes at the same time to prevent file corruption, or at the least, prevent any drastic issues with .lock files...
Template error: are you trying to use the = sign? Visit Help:Template#Escape template-breaking characters for workarounds.
Note: If you should choose to use unicast in your /etc/ha.d/haresources file, for say, a dedicated heartbeat connection with a cat5e crossover cable & extra NICS in our two nodes, the /etc/ha.d/haresources file on each node should represent the other nodes IP Address, for example, the unicast line in node1's haresources file would look like this 'ucast eth0 node2.ip.address.here', and the unicast line in haresources on node2 would be the opposite like such 'ucast eth0 node1.ip.address.here', in ALL other scenarios, our haresources file should be 100% identical on both nodes.

03. Edit/create the file /etc/ha.d/haresources as follows:

node1 IPaddr:: named httpd mysqld

Note: this file should be EXACTLY the same on both nodes!!! ...explanation: node1 is the hostname of our first/primary node(see above), IPaddr:: is the virtual IP address, and the gateway/device we are expecting to assign this 'virtual' WAN IP address to us. In most circumstances the gateway definition will not be neccesary(EG. remove '/ so it reads IPaddr::' but for my purposes, because of some funky behaviour of my att 2wire u-verse gateway and how it assigns it's static IP addresses, heartbeat would fail to bring up the virtual IP address/eth0:0 alias without this specifically defined in the haresources file. Please remove 'named httpd mysqld' from haresources if you will not be running BIND, Apache/other web server, MySQL Server, otherwise heartbeat will try to start bind, apache, and mysql server and COMPLETELY FAIL to start if you do not have startup scripts for these in /etc/rc.d/(heartbeat looks for startup scripts/resource agents in BOTH: /etc/rc.d/* & /etc/ha.d/resource.d, the resource agent 'IPaddr is in /etc/ha.d/resource.d for example)
Note: You MAY define more then one(as many as you like) VIRTUAL IP Adresses, this is desirable in MOST scenarios, for example, you dns registrar requires that you have two name servers and you would like to host your own dns zone. For this our haresources file would look like: 'node1 IPaddr:: IPaddr:: named httpd mysqld' where we have two highly available IP's &

04. after the line 'node1 IPaddr::' our /etc/ha.d/haresources file append the name of the startup scripts/resource agents for whatever service you would like heartbeat to make highly available. In the example above in step 03. We have told heartbeat to manage named(BIND) httpd(APACHE) mysqld(MYSQL Server), see here:

IPaddr:: IPaddr:: named httpd mysqld

Note: any services you define in your /etc/ha.d/haresources file MUST NOT be defined in your /etc/rc.conf daemons array, or heartbeat will fail when trying to start the service(it will already be running). It is not reccomended to tell heartbeat to manage services like sshd(SSH) eg. if we tell heartbeat to manage sshd(SSH), we will only be able to ssh into whatever node is currently active as our master node, and not the slave node.

05. Edit /etc/rc.conf and place heartbeat at the END of your daemons array...

That's it! Fire up both nodes, pull the plug on your primary node/node1, and check node two to see that it has taken over your H.A Services & Virtual IP Address. E.G. 'ip addr show','ps aux'...

Please feel free to contact me in #archlinux @freenode if you need any assistance, and PLEASE(for *** sake) update/spell check/and fix the formatting in this article if you have some spare time!!! (I don't right now :P).

--Sulucohmun 23:07, 11 August 2011 (EDT)Sully O. / Suluc Ohmun.