Best Practices for Implementing High Availability for SAS

Best Practices for Implementing High Availability for SAS® 9. 4 Cheryl Doninger, SAS Zhiyong Li, SAS Bryan Wolfe, SAS

What is HA and Why is it Important for SAS? § Analytics are mission critical to organizations § More and more SAS deployments are operational systems § Large user populations depend on these systems § SAS is mission critical to organizations § Analysts must be able to access the SAS environment at all times 2

The Goals of this HA Best Practice § Consistent HA strategy for all components § Minimize the number of technologies used for HA § SAS Grid Manager § Clustering § Start with smallest scenario and build to most scalable 3

SAS Deployment Tiers § SAS Metadata Server Tier § SAS Application Server Tier § SAS Web or Middle Tier § SAS Client Tier 4

Critical Components § SAS® Metadata Server (Metadata Server) § SAS Web Server (Web Server) § SAS Web Application Server (Web Application Server) § SAS Web Infrastructure Platform Data Server (WIP Data Server) § SAS JMS Broker (JMS Broker) § SAS Cache Locator (Cache Locator) § SAS Object Spawner (Object Spawner) § SAS OLAP Server (OLAP Server) § SAS Environment Manager Server (EV Server) § SAS Environment Manager Agent (EV Agent) § SAS Deployment Agent (Deployment Agent) 5

Some Terminology 6

Bad Things Happen When a Service Goes Down 7

Active-Passive Mode Cold Standby Node Primary Node 8

Master-Slave Mode Warm Standby Node Primary Node 9

Active-Active Mode Cluster 10

SAS Grid Manager § Monitor any/all critical services § Restart on the same node if the service fails § Failover to a standby node if the primary node fails 11

SAS Grid Manager: Active-Passive 12

SAS Grid Manager: Active-Active 13

SAS Metadata Server Cluster § A cluster is three or more metadata server nodes § Each node is a full server with a complete copy of all metadata § One node is designated the master to coordinate the cluster § All other nodes are slave nodes § Clients connect to slave nodes § Once connected the cluster behaves like a normal server 14

Three Node SAS Metadata Server Cluster MASTER NODE 2 SHARED BACKUP SLAVE NODE 1 NODE 3 15

SAS Web Application Server Cluster SAS Web Server (Load-balance / reverseproxy) SAS Web Application Server Web Applications* Web Services SAS Web Infrastructure Platform Data Server JMS Broker Cache Locator * Not all SAS Web Applications support clustering due to technical limitations. 16

Legend § “SGM restart/cold failover” § SAS Grid Manager monitors active service instance and restarts on same machine or starts a new instance on a cold standby machine if the original machine dies § Requires a failover machine § Can apply to both active/passive and active/active § Parentheses ( ) denote cold standby for failover § “Warm failover” § Master-slave mode § Chevrons < > denote warm standby, the slave node of master/slave cluster 17

Legend (continued) § “Clustered and SGM Restart” § Clustering capability native to the service will be used § SAS Grid Manager monitors active service instance and restarts on same machine » There is no standby machine if the original machine dies § Services on multiple machines without ( ) indicate clustering § “SGM restart” § SAS Grid manager monitors a service instance and restarts the service on the same machine § These services have 1 -1 mapping per node 18

Company Confidential - For Internal Use Only Copyright © 2012, SAS Institute Inc. All

Talking Points. . . Component Shared Config Per-node Config Mode Metadata Server X active-passive Object Spawner X active-passive OLAP Server X active-passive Web Server X active-passive EV Agent X n/a WIP Data Server X master-slave EV Server X master-slave JMS Broker X master-slave Web Application Server X active-active Cache Locator X active-active 20

Talking Points. . . Component Shared Config Metadata Server Per-node Config Mode X active-active Object Spawner X active-passive OLAP Server X active-passive Web Server X active-passive EV Agent X n/a WIP Data Server X master-slave EV Server X master-slave JMS Broker X master-slave Web Application Server X active-active Cache Locator X active-active 22

Talking Points. . . Component * active-active w/ proper licensing Shared Config Metadata Server Per-node Config Mode X active-active Object Spawner X active-passive * OLAP Server X active-passive * Web Server X active-passive EV Agent X n/a WIP Data Server X master-slave EV Server X master-slave JMS Broker X master-slave Web Application Server X active-active Cache Locator X active-active 24

Talking Points. . . Component * active-active w/ proper licensing Shared Config Metadata Server Per-node Config Mode X active-active Object Spawner X active-passive * OLAP Server X active-passive * Web Server X active-passive EV Agent X n/a WIP Data Server X master-slave EV Server X master-slave JMS Broker X master-slave Web Application Server X active-active Cache Locator X active-active 26

Summary § SAS is mission critical and, therefore, must be highly available § SAS Grid Manager + Clustering § HA across entire software stack with most functionality § SAS Grid Manager – Clustering § No load balancing capabilities § Clustering – SAS Grid Manager § No automatic restart and failover capabilities 27

More Information. . . § http: //support. sas. com/rnd/scalability/gridpapers. html § http: //support. sas. com/rnd/scalability/grid/HA/gridha. html § SAS 9. 4 Intelligence Platform: Middle-Tier Administration Guide, Second Edition. § “Using Metadata Server Clustering. ” In SAS 9. 4 Intelligence Platform: System Administration Guide § “Understanding Server Load Balancing. ” In SAS 9. 4 Intelligence Platform: Application Server Administration Guide. 28

? ? ? Questions ? ? ? 29