Services as Workload Boundaries for RAC/Data Guard/Applications and Operational Routing

Learning outcomes

ServiceHub has three connection strings that differ only by host and SID. During maintenance, the DBA stops one instance and expects clients to “figure it out.” A database service should instead be the stable workload contract: it names the business class, carries service attributes, can be created in a PDB, can move among RAC instances, and can start only on the database role where the workload is valid.

1. A service is the application contract

Clients should request a service such as servicehub_oltp, not an instance SID. The service communicates what workload this is; Oracle/network/Clusterware can decide where it currently runs. This makes the same name useful for Resource Manager mapping, service statistics, RAC placement and Data Guard role routing.

2. Free lab: create a PDB-local service

BEGIN  BEGIN DBMS_SERVICE.STOP_SERVICE('servicehub_ch21_ops'); EXCEPTION WHEN OTHERS THEN NULL; END;  BEGIN DBMS_SERVICE.DELETE_SERVICE('servicehub_ch21_ops'); EXCEPTION WHEN OTHERS THEN NULL; END;  DBMS_SERVICE.CREATE_SERVICE(    service_name => 'servicehub_ch21_ops',    network_name => 'servicehub_ch21_ops'  );  DBMS_SERVICE.START_SERVICE('servicehub_ch21_ops');END;/SELECT name,network_name,pdbFROM v$servicesWHERE name='servicehub_ch21_ops';

When DBMS_SERVICE.CREATE_SERVICE runs in a PDB, the service is associated with that PDB. In a database managed by Oracle Restart/Clusterware, Oracle recommends managing services with SRVCTL instead.

3. Verify the client really used the service

sql servicehub_app@//localhost:1521/servicehub_ch21_opsBEGIN  DBMS_APPLICATION_INFO.SET_MODULE('SERVICEHUB_OPS','HEALTH_CHECK');END;/SELECT  SYS_CONTEXT('USERENV','SERVICE_NAME') AS service_name,  SYS_CONTEXT('USERENV','CON_NAME') AS con_name,  SYS_CONTEXT('USERENV','INSTANCE_NAME') AS instance_nameFROM dual;

SELECT  service_name,  module,  action,  COUNT(*) AS sessionsFROM v$sessionWHERE type='USER'GROUP BY service_name,module,actionORDER BY service_name,module,action;

4. Deliberately wrong: hard-code an RAC node hostname/SID

If the client targets only racnode1, the database can remain available on node 2 while the application still cannot connect. RAC best practice is a service discovered through SCAN and managed by Clusterware. Planned maintenance relocates/drains the service before stopping an instance.

srvctl add service   -db servicehub_rac   -service servicehub_oltp   -pdb FREEPDB1   -preferred shrac1   -available shrac2   -policy AUTOMATIC   -notification TRUEsrvctl status service   -db servicehub_rac   -service servicehub_oltp   -verbose

5. Planned draining is different from failure recovery

srvctl relocate service   -db servicehub_rac   -service servicehub_oltp   -oldinst shrac1   -newinst shrac2   -drain_timeout 120   -stopoption TRANSACTIONAL

The drain timeout is workload-specific. During planned draining, new work should be directed away while eligible in-flight requests finish. An unplanned instance crash cannot grant that same drain window.

6. Data Guard services follow database roles

A Data Guard role transition changes which database is primary. Broker/startup triggers/Clusterware role-based services can ensure a writable service runs only on PRIMARY, while a reporting service may run on a read-capable standby when licensed. This is separate from redo transport/apply itself.

SELECT db_unique_name,database_role,open_modeFROM v$database;SELECT name,pdb,network_nameFROM v$servicesWHERE name LIKE 'servicehub%';

Clients still need valid connect descriptors/SCAN/listener/DNS endpoints and retry/failover behavior. Data Guard failover does not teleport an existing TCP session.

7. Connection failover versus request replay

Fast Application Notification (FAN)-aware pools can discard dead connections and create new ones on healthy service members. Application Continuity can replay eligible requests, but it has RAC/Active Data Guard/RAC One Node entitlement and driver/service requirements. Do not promise replay for arbitrary external side effects or unknown commit outcomes.

8. RESET_STATE improves stateless pooled-session hygiene in 26ai

As of 26ai, the service attribute RESET_STATE works independently of Application Continuity. For stateless applications, it can clear session state between requests and reduce leakage of contexts/package/session settings to the next pool borrower. Validate the exact driver/pool/service combination and do not enable it for applications that intentionally persist database session state between requests.

9. Services become Resource Manager mapping inputs on entitled systems

BEGIN  DBMS_RESOURCE_MANAGER.CREATE_PENDING_AREA();  DBMS_RESOURCE_MANAGER.SET_CONSUMER_GROUP_MAPPING(    DBMS_RESOURCE_MANAGER.SERVICE_NAME,    'servicehub_oltp',    'SH21_OLTP');  DBMS_RESOURCE_MANAGER.SET_CONSUMER_GROUP_MAPPING(    DBMS_RESOURCE_MANAGER.SERVICE_NAME,    'servicehub_batch',    'SH21_BATCH');  DBMS_RESOURCE_MANAGER.VALIDATE_PENDING_AREA();  DBMS_RESOURCE_MANAGER.SUBMIT_PENDING_AREA();END;/

This creates one consistent vocabulary across connection routing, monitoring and resource governance.

10. Service SLO card

11. Cleanup

BEGIN  DBMS_SERVICE.STOP_SERVICE('servicehub_ch21_ops');  DBMS_SERVICE.DELETE_SERVICE('servicehub_ch21_ops');END;/

12. Production judgment

Name services by workload contract, not server name. Instrument every pooled request with module/action/client ID, test planned drain and unplanned failure separately, and verify the database role/service before accepting writes after Data Guard transitions. Keep connect timeouts/retries bounded and driver-specific.

DBMS_SERVICE local service creation is Free-compatible. RAC SRVCTL, FAN/SCAN and role placement require their topology/licensing; Application Continuity has explicit option prerequisites on EE/EE-ES. Lesson 5 consolidates all chapter mechanisms into one governance policy with emergency override and rollback.

Authoritative references

• Managing Connections to a PDB — DBMS_SERVICE/SRVCTL PDB service scope
• DBMS_SERVICE — service creation/attributes
• RAC Workload Management with Services — preferred/available placement
• Ensuring Application Continuity — draining/FAN/replay/RESET_STATE
• Data Guard Broker — role transitions and service operations