I'm using Lettuce Redis client for JVM to build a queue backed by a Redis List. Ideally it would behave like an in-memory queue but since the network interaction is involved this is not possible.
There is a section on error handling in Lettuce docs and in my case I would like to retry failed commands on error. The problem is the exception hierarchy is not very fine grained and I'm not sure how to deal with the following issues:
- how to determine if the failed command can be retried? There are a number of reasons that command might fail indefinitely and retrying would lead to an infinite loop - current Redis version doesn't support the command syntax used, the key already exists and it is of a non-compatible type, etc.
- can I rely on Lettuce to always reconnect in case of unreliable network? Is there a possibility that in some cases I should not retry on some
RedisExceptionorNativeIoExceptionbut recreate the Redis client instance or even restart the whole application? - is there a way to know if the failed command was not actually executed by Redis and retrying will not lead to duplication or data loss (or at least this might be determined in some cases by the exception class)? This might be a fundamental issue of exactly-once delivery and as far as I know Redis doesn't provide any means to deal with this (unlike Kafka, for example) but maybe there is some established practice to deal with this problem?
This problem looks like something everybody has to deal with but there doesn't seem to be a lot of information regarding this.