I was running through the cycle of the periodic runner threads (abort oldest under cache pressure, and abort expired) in gdb and noticed that for inactive transactions, we are not reporting the error code when we abort, however they still abort successfully and the counter for kills done by the threads are incremented.

For active transactions, this works as designed and we report the temporarily unavailable error code that are passed down and also the log message, "Transaction aborted due to cache pressure." More details in the comment thread below.

These are my observations to investigate further.