September 27, 2013 Leave a comment
Today we will explore little more about SCAN listeners. The common question comes in mind like: How many SCAN listeners could be configured in a multi-node RAC environment? , How many SCAN listeners would actually be required and how many would be enough? Etc.
Let’s find out the answers for these common queries.
I am assuming that by now most of DBAs are pretty comfortable with SCAN listeners; in case need detailed explanation then following URL would a good choice to visit here. As usual, awesome documentation from Oracle for SCAN in 11g as well as 12c releases.
Basically SCAN Listeners are introduced to create another connection handler on top of existing node listeners to overcome the failover and load balancing issues till 11gR1 architecture for RAC. Till 11gR1 Node listeners were defined on VIPs to facilitate NACK (Negative Acknowledgement) to resolve the issue of TCP timeouts.
Pre 11gR2 database environments, clients may take up to 2 minutes to decide (on TCP level) a node is down. This is purely because of the TCP Timeouts, which can differ from platform to platform. These 2 minutes are unacceptable, and it was a good thing of Oracle to understand and address this issue. Oracle designed a virtual IP address to be assigned to the public interface. Under normal circumstances, the VIP will be located to its designated NIC, and the listener will be bound to this VIP. Whenever there is a need to failover the VIP to another node in the cluster, when clients want to connect to this VIP (due to tns alias addresses pointing to this VIP), the VIP will immediately respond (because it has failed over and the TCP stack is running against it), and the client is able to get negative acknowledgement (NACK) confirming that no listener is active at its designated port. Within few seconds the client will know and fail over to the alternative address in its TNS alias. This makes failover a lot faster.
From 11gR2 onwards oracle enhanced the administration as well as availability of database from client’s perspective, load balancing etc by introducing SCAN (Single Client Access Name). With SCAN, clients could use SCAN-NAME (resolved by 3 VIPs, for default configuration) rather than list of all rac nodes in connect string. By default SCAN listeners (defined on SCAN VIPs) created as first point which co-ordinate with Node Listeners (defined on VIPs), so even if failure of node scan listener running on that node (if any) would be relocated to any surviving node while available SCAN listeners would be serving as normal.
In practice any complex environment designed with multi-tier architecture involving connection handling via connection pool mechanism. It reduces connectivity time for application or web based end users. Normally 3 SCANs are capable enough to handle hundreds of new connections in a span of few seconds. Still if you feel that default 3 SCAN listeners are not enough for your environment then you have an option to add few more scan listeners. GNS based dynamic IP scheme still has no way to change number of SCAN listeners in your environment. Today we will elaborate the process of adding one extra scan listener in our DNS based static IP configuration. Read more of this post