Software that Learns from its Own Failures

All non-trivial software systems suffer from unanticipated production failures. However, those systems are passive with respect to failures and do not take advantage of them in order to improve their future behavior: they simply wait for them to happen and trigger hard-coded failure recovery strategies. Instead, I propose a new paradigm in which software systems learn from their own failures. By using an advanced monitoring system they have a constant awareness of their own state and health. They are designed in order to automatically explore alternative recovery strategies inferred from past successful and failed executions. Their recovery capabilities are assessed by self-injection of controlled failures; this process produces knowledge in prevision of future unanticipated failures.