Using Redis with a Guidewire cluster
In general practice, Guidewire recommends that you set up three redundant Redis servers (without Sentinel) on different physical hosts to avoid a single point of failure. Even in the unlikely event that all Redis servers failed simultaneously, the PolicyCenter cluster will continue to work. Only broadcast messages (including cache evictions) will stop working with all the Redis servers down, which can cause a rise in the frequency of CDCEs (Concurrent Data Change Exceptions) and other potential problems, for example, product model reloading. However, as long as at least one of the redundant Redis servers is working and reachable by all cluster nodes, Guidewire applications will continue working with very minimal impact.
Guidewire recommendations
Guidewire makes the following general recommendations in working with a Redis server.
- Redis server
- Ensure that you turn off the Redis persistent settings.
RedisBroadcastTransportFactoryplugin- Set the
batchWriteIntervalplugin parameter to be between 100 to 200 (ms). Start at the upper value and gradually reduce the value until the number of CDCEs approaches zero. - Application log
- Monitor the log entries closely during testing. Multiple occurrences of the following
log entry indicate that a Redis server is down:
- redis.clients.jedis.exceptions.JedisConnectionException