Oracle RAC VIP and SCAN IPs

Question

I've read the Oracle RAC documentation a couple of times but SCAN and VIP are still confusing me. Can someone help me understand how this needs to be configured technically so that I can explain it my network admin.

1. VIP in Oracle RAC, should each VIP bind to the node or just require a DNS A record without allocating it to node1 or node2 and an entry in the host file? I know while performing Grid cluster installation Oracle will bind the VIP automatically, but should this be part of DNS assigned to one of the nodes or should it be free and unassigned?

2. Oracle SCAN IPs need to be created in DNS record; is this an A record to 3 IPs with reverse lookup or round robin way and this should not be part of hosts file?

I need to explain this to my network admin to add it on the DNS server.

Answer 1

First, VIPs:

A VIP is a Virtual IP address, and should be defined in DNS and not assigned to any host or interface. When you install GRID/ASM home, you'll specify the VIP names that were assigned in DNS. When Oracle Clusterware starts up, it will assign a VIP to each node in the cluster. The idea is, if a node goes down (crashes), clusterware can immediately re-assign that VIP to a new (surviving) node. This way, you avoid TCP timeout issues.

Next, SCAN:

A SCAN (Single Client Access Name) is a special case of VIP. The SCAN should also be defined in DNS, and not assigned to any host or interface. There should be three IPs associated with the SCAN name in DNS, and the DNS entry should be defined so that one of the three IPs is returned each time DNS is queried, in a round robin fashion.

At clusterware startup time, each of the three VIPs that make up the SCAN will be assigned to a different node in the cluster. (Except in the special case of a two node cluster, one of the nodes wil have a 2 SCAN VIPs assigned to it.) The point of the SCAN, is that no matter how many nodes are added to or removed from the cluster, all the Net Service Name definitions in your tnsnames.ora (or LDAP equivalent) will not need to ever change, because they all refer to the SCAN, which doesn't change, regardless of how many node additions or drops are made to the cluster.

For example, in the three node cluster, you may have:

Physical and virtual hostnames/IPs assigned as follows:

Hostname    Physical IP    Virtual hostnmae   Virtual IP
rac1        10.1.1.1       rac1-vip           10.1.1.4
rac2        10.1.1.2       rac2-vip           10.1.1.5
rac3        10.1.1.3       rac3-vip           10.1.1.6

Additionally, you may have the SCAN defined as: rac-scan with three IPs, 10.1.1.7, 10.1.1.8, 10.1.1.9. Again, the DNS definition would be defined so those IPs are served up in a round robin order.

Note that the SCAN VIPs, Host VIPs, and the Physical IPs are all in the same subnet.

Finally, though you didn't ask about it, to complete the picture, you'd also need one private, non-routable IP assigned per host, and that IP would be associated with the private interconnect. So, you may have something like:

rac1-priv  172.16.1.1
rac2-priv  172.16.1.2
rac3-priv  172.16.1.3

Note that the '-priv' addresses should not be in DNS, only in the /etc/hosts file of each host in the RAC cluster. (They are private, non-routable, and only clusterware will ever know about or use those addresses, so adding to DNS doesn't make sense.)

Note also, that '-priv' and physical IP/hostname definitions should go in /etc/hosts, and the physical IPs and VIPs should be in DNS. So, physical IPs in both DNS and /etc/hosts, VIPs only in DNS, '-priv' addresses only in /etc/hosts.