Proactive operating system memory unit replacement based on namespace health

A system comprising:

• a memory comprising computer readable instructions;
• and
• a processing device for executing the computer readable instructions, the computer readable instructions controlling the processing device to perform operations comprising:
• receiving diagnostic data about a plurality of operating system (OS) memory units of a distributed computing system, the diagnostic data comprises key-value pairs of metrics;
• converting, for each of the key-value pairs of metrics, a value of a key-value pair into an array of numbers, quantizing the array into a vector using k-means clustering, classifying the vector, and storing, when classified within an unhealthy cluster, a distance to a centroid of the unhealthy cluster within the value of a key-value store;
• defining a failure state for predicting whether at least one of the plurality of OS memory units will experience a failure;
• predicting whether the at least one of the plurality of OS memory units will experience the failure based at least in part on the failure state and the diagnostic data about the plurality of OS memory units, wherein the predicting comprises determining that the at least one of the plurality of OS memory units is likely to fail responsive to the distance to the centroid of the unhealthy cluster being below a threshold distance;
• and
• causing the at least one of the plurality of OS memory units predicted to experience the failure to be replaced with at least one replacement OS memory unit without impacting a namespace availability of the distributed computing system.

A computer program product comprising:

• a set of one or more computer-readable storage media;
• program instructions, collectively stored in the set of one or more storage media, for causing a processor set to perform the following computer operations:
• receiving diagnostic data about a plurality of operating system (OS) memory units of a distributed computing system, the diagnostic data comprises key-value pairs of metrics;
• converting, for each of the key-value pairs of metrics, a value of a key-value pair into an array of numbers, quantizing the array into a vector using k-means clustering, classifying the vector, and storing, when classified within an unhealthy cluster, a distance to a centroid of the unhealthy cluster within the value of a key-value store;
• defining a failure state for predicting whether at least one of the plurality of OS memory units will experience a failure;
• predicting whether the at least one of the plurality of OS memory units will experience the failure based at least in part on the failure state and the diagnostic data about the plurality of OS memory units, wherein the predicting comprises determining that the at least one of the plurality of OS memory units is likely to fail responsive to the distance to the centroid of the unhealthy cluster being below a threshold distance;
• and
• causing the at least one of the plurality of OS memory units predicted to experience the failure to be replaced with at least one replacement OS memory unit without impacting a namespace availability of the distributed computing system.