Two-node RAC on OCI Base Database Service: HA design and operations

#What you get

• Two VMs in different fault domains, shared ASM storage, Oracle Clusterware.
• Both instances open the same database (active-active).
• Rolling patching of OS, Grid Infrastructure and database: no full outage for patches.
• RAC requires Enterprise Edition Extreme Performance (or BYOL with RAC option) and ASM.

#1. Create

Console → Base Database → Create DB system → shape (flexible VM), Node count: 2, Enterprise Edition Extreme Performance, ASM storage. Pick the private subnet; OCI assigns node IPs, VIPs and SCAN IPs.

#2. Services: the key to RAC HA

Don't connect apps to the default service. Create a service per workload with preferred instances and failover settings:

srvctl add service -db SALES_xyz -service sales_oltp -preferred SALES1,SALES2 \
  -failovertype TRANSACTION -commit_outcome TRUE -failover_restore LEVEL1 \
  -drain_timeout 300 -stopoption IMMEDIATE -notification TRUE -pdb SALESPDB
srvctl start service -db SALES_xyz -service sales_oltp
srvctl status service -db SALES_xyz

-failovertype TRANSACTION enables Application Continuity (replay of in-flight work after an instance failure for supported drivers). For apps without AC support, use -failovertype SELECT or at least FAN-aware pools.

#3. Connection string

(DESCRIPTION=(CONNECT_TIMEOUT=90)(RETRY_COUNT=50)(RETRY_DELAY=3)(TRANSPORT_CONNECT_TIMEOUT=3)
 (ADDRESS=(PROTOCOL=TCP)(HOST=sales-scan.dbsubnet.vcn.oraclevcn.com)(PORT=1521))
 (CONNECT_DATA=(SERVICE_NAME=sales_oltp.dbsubnet.vcn.oraclevcn.com)))

Use the SCAN name, never node IPs. Use the Universal Connection Pool / ODP.NET pool with FAN enabled.

#4. Rolling patching

oci db patch list by-db-system --db-system-id <dbsys-ocid> --output table
oci db system patch --db-system-id <dbsys-ocid> --patch-id <patch-ocid> --patch-action APPLY   # GI/OS
oci db database patch --database-id <db-ocid> --patch-id <patch-ocid> --patch-action APPLY   # database

OCI patches one node at a time. Services with -drain_timeout let sessions finish (or move) before an instance stops.

#5. Check cluster health

crsctl stat res -t
srvctl status database -db SALES_xyz
srvctl config scan

#6. What RAC doesn't cover

Tip

Test node failure with real application load (reboot one node) before go-live. RAC HA only works if the application connects through services and the pool handles FAN events.