Maintenance Manual

Chapter 2 Fault Tolerance
Central Controller
2-12
Cisco ICM Enterprise Edition Administrator Guide Release 6.0(0)
of the Central Controller lose synchronization with the other side. The CallRouter,
Logger, and Database Manager must all be resynchronized before normal
duplexed operation can resume.
For a single-customer ICM, the recovery process begins when the Node Manager
notices the failure of a CallRouter process and automatically restarts it. Other
processes are not impacted. In a network service provider (NSP) environment, if
several ICM instances are running on the same machine, the Node Manager
cannot restart the machine. In such NSP environments, manual intervention is
required to restart the failed CallRouter process.
The restarted CallRouter immediately initiates a state transfer from its currently
executing peer. Each CallRouter sends a message to its Logger. The Loggers then
perform their own state transfer.
When the state transfer is completed, all processes are now synchronized. The
newly on-line Central Controller sends an in-service status to all local Agents. It
then begins processing input messages. After the state transfer, both sides of the
Central Controller see exactly the same sequence of input messages. At this point
the ICM system is returned to full duplexed operation.
Logger and Database Manager Recovery
Logger recovery is closely linked with central database recovery. In central
database recovery, the SQL Server component of the central database is accessed
directly through its client interface rather than through proprietary ICM
interfaces. Therefore, in addition to synchronization and state transfer, the
following database recovery procedures must be performed before the Logger can
return to full duplexed operation:
• The Database Manager must run SQL Server automatic recovery.
• The state transfer process may need to update configuration data in the central
database.
• The Database Manager may need to run historical recovery to recover
historical data gathered during the off-line period.
When a single Logger process fails, ICM software continues to operate with the
Logger on the other side. The remaining Logger ensures that output messages
continue to reach PGs and Admin Workstations. The ICM’s Message Delivery
Service detects the failure of the Logger and notifies the PGs and Admin