Mechanism for continuously and unobtrusively varying stress on a computer application while processing real user workloads

A system comprising:

• a computing device having a memory to store instructions relating to a mechanism, and a processing device to execute the instructions such that the mechanism performs one or more operations, wherein the mechanism is operable to:
• configure application resources for a recovery configuration whose service levels are satisfactory;
• configure the application resources for stress configurations to affect service levels;
• transition the application resources from the recovery configuration to a stress configuration for a time duration, wherein the application resources of the stress configuration are transitioned back to the recovery configuration;
• acquiring service level and stress data for the stress configuration where the latency to acquire the data when used as the time duration of the stress configuration causes unsatisfactory service levels;
• determine a combination of a next stress configuration and the time duration to vary stress such that user service levels are unobtrusively affected by limiting the time duration in inverse relation to an uncertainty in predicting the service level impact of the stress configuration;
• and
• determining a stress level metric for the stress configuration, wherein the stress level metric comprises a load per a resource which is one of calculated as a total workload divided by a number of resources or calculated as the workload on a specific resource which receives a variable fraction of the total workload, and wherein the maximum stress limit is calculated as a lower of a currently defined maximum stress limit and the stress level metric determined when service levels impact a response time service impact limit.

A non-transitory machine-readable storage medium including data that, when accessed by a machine, cause the machine to perform operations comprising:

• configuring application resources for a recovery configuration whose service levels are satisfactory;
• configuring the application resources for stress configurations to affect service levels;
• transitioning the application resources from the recovery configuration to a stress configuration for a time duration, wherein the application resources of the stress configuration are transitioned back to the recovery configuration;
• acquiring service level and stress data for the stress configuration where the latency to acquire the data when used as the time duration of the stress configuration causes unsatisfactory service levels;
• determining a combination of a next stress configuration and the time duration to vary stress such that user service levels are unobtrusively affected by limiting the time duration in inverse relation to an uncertainty in predicting the service level impact of the stress configuration;
• and
• determining a stress level metric for the stress configuration, wherein the stress level metric comprises a load per a resource which is one of calculated as a total workload divided by a number of resources or calculated as the workload on a specific resource which receives a variable fraction of the total workload, and wherein the maximum stress limit is calculated as a lower of a currently defined maximum stress limit and the stress level metric determined when service levels impact a response time service impact limit.