Modifying distributed application based on cloud diagnostic data

A system comprising non-transitory media encoded with code that, when executed by a processor, implements a process including:

• identifying plural public-cloud virtual machines of a public cloud that are hosting application instances of a distributed application;
• setting each of the virtual machines of a public cloud to a diagnostic mode of operation in which per-virtual-machine diagnostic metrics for that virtual machine are collected periodically using a first average collection period, such that the per-virtual-machine diagnostic metrics collected during each collection period are stored by the public cloud temporarily for a storage duration, each of the virtual machines having an alternative normal mode of operation in which such per-virtual-machine diagnostic metrics are not collected periodically and are not stored temporarily for the storage duration;
• extracting the per-virtual-machine diagnostic metrics from the public cloud periodically over an extraction duration longer than the storage duration;
• aggregating the extracted per-virtual-machine diagnostic metrics to generate distributed-application utilization metrics for the distributed application or for a multi-virtual-machine component of the distributed application;
• and
• reconfiguring the distributed application based at least in part on the distributed-application utilization metrics.