Opportunistic offlining for faulty devices in datacenters

A method performed by one or more computing devices comprising processing hardware and storage hardware, the method for determining whether to offline a resource distribution unit (RDU) in a datacenter, the datacenter comprised of RDUs providing a resource to respective sets of datacenter hosts that service components of datacenter tenants, the method comprising:

• receiving an indication that the RDU has experienced a fault;
• collecting, and storing in the storage hardware, capacity information indicating net available capacity of a computing resource of the datacenter;
• collecting, and storing in the storage hardware, tenant availability information indicating presence of tenant components on one or more hosts receiving the resource from the RDU;
• and
• based on the indication that the RDU has experienced a fault, determining, by the processing hardware, whether to offline the RDU, the determining comprising:
• (i) according to the capacity information in the storage hardware, computing a net available capacity of the computing resource that would result from offlining the RDU, and
• (ii) according to the tenant availability information in the storage hardware, computing a tenant availability measure that would result from offlining the RDU.

A computing device comprising:

• processing hardware;
• storage hardware storing information configured to cause the processing hardware to perform a process, the process comprising:
• determining that a resource distribution unit (RDU) of a datacenter is in a faulty state, the RDU supplying a resource to computer hosts of the datacenter, and based on the determining that the RDU is in a faulty state:
• determining whether to make the RDU unavailable to the computer hosts supplied thereby by predicting whether making the RDU unavailable would cause an insufficient availability of tenant virtual machines (VMs) and by predicting whether making the RDU unavailable would cause an insufficient amount of available capacity of a computing resource of the datacenter.