In the Handling faults recipe, we saw how stream processors delegate hard errors by publishing fault events with all the data pertaining to the unit of work that failed. In this recipe, a fault monitor consumes these fault events and stores them in an S3 bucket. This enables the team to review the specific fault to help determine the root cause of the problem. The fault contains the specific exception that was caught, the event that failed, and all of the contextual information that was attached to the unit of work.
Once the root cause has been addressed, the original event can be submitted back to the stream processor that published the fault. This is possible because the fault event contains the original Kinesis record ( ...
