We intentionally use an unsafe storage method for writing payloads to take advantage of a full connection pool for performance. However this means if a HA Sequencer Writer has lost its instance lock it may still attempt to write payloads while another Sequencer Writer is active with the same instance index. As we use this instance index to generate an (almost) conflict free payload id, in this circumstance there is a slim chance that we may attempt to write conflicting payloads with the same id. If we were using a simple idempotent write approach this could result in the active sequencer writing an event with a payload from the offline writer process (and not the payload it is expecting). This would be a terrible and difficult to diagnose corruption issue.

If this exception is raised we currently just halt the writer and run crash recovery. This is slightly suboptimal as in the above scenario we may crash the active writer (if they were second to write a conflicting payload id). However this will be safe. We could optimise this by checking the active lock status and only halting if this is found to be false.