Abstract: Embodiments relate to diagnosis and recovery of cloud based systems. From an incident ticket, a dynamic diagnostics graph is generated visualizing a hierarchy (ancestor, child) of diagnostic jobs investigating the functioning cloud system. By indicating and checking job statuses, child jobs dependent on a skipped or failed job can be skipped according to a dynamic pruning technique—thereby trimming an entire branch. And, by running separate groups of diagnostic jobs in parallel across different nodes, the diagnostic process can be finished rapidly and efficiently. A diagnostic report includes the dynamic diagnostics graph. For system recovery, the dynamic diagnostic graph is analyzed to automatically provide one or more appropriate Recommended Actions (RAs) resolving cloud system problem(s) revealed by diagnostic efforts. Those appropriate RAs may be provided by performing machine learning (e.g., referencing a neural network) with a model trained from historical cloud diagnostic and recovery activity.