Handling workload interruptions in cloud computing platforms

A method for handling interrupts, the method comprising:

• receiving, using a controller process of a plurality of controller processes within a cloud computing system, an interrupt indicator from a worker process of a plurality of worker processes, wherein the interrupt indicator signals a potential interruption to the worker process, wherein each worker process of the plurality of worker processes is associated with a corresponding controller process of the plurality of controller processes, and wherein each controller process manages a corresponding worker process;
• determining, based on the interrupt indicator, whether the worker process is expected to be halted and removed by the cloud computing system within a predetermined time period, wherein the cloud computing system periodically halts and removes processes to manage assigned computing resources;
• based on determining that the worker process is expected to be halted and removed by the cloud computing system, transmitting a status request to the worker process;
• in response to the status request, receiving, from the worker process, a first set of computational results for a first set of tasks that have been completed by the worker process;
• determining, based on the first set of computational results, a second set of tasks that have not been completed the worker process, wherein the first set of tasks and the second set of tasks have been assigned to the worker process;
• generating, for the worker process, a checkpoint data structure that indicates that the first set of tasks has been completed and that the second set of tasks has not been completed;
• transmitting, to be recorded within the cloud computing system, a computational payload comprising the first set of computational results and the checkpoint data structure;
• receiving, using the controller process, a task completion message indicating that a subsequent task of the second set of tasks has been completed by the worker process, wherein the task completion message comprises a corresponding computational result;
• updating the checkpoint data structure to remove the subsequent task from the second set of tasks and add the subsequent task to the first set of tasks with the corresponding computational result to generate an updated checkpoint data structure;
• and
• transmitting the updated checkpoint data structure to be recorded within the cloud computing system.

One or more non-transitory, computer-readable media storing instructions thereon that when executed by one or more processors perform operations comprising:

• receiving, using a controller process of a plurality of controller processes within a cloud computing system, an interrupt indicator from a worker process of a plurality of worker processes, wherein the interrupt indicator signals a potential interruption to the worker process, wherein each worker process of the plurality of worker processes is associated with a corresponding controller process of the plurality of controller processes, and wherein each controller process manages a corresponding worker process;
• determining, based on the interrupt indicator, whether the worker process is expected to be halted and removed by the cloud computing system within a predetermined time period, wherein the cloud computing system periodically halts and removes processes to manage assigned computing resources;
• based on determining that the worker process is expected to be halted and removed by the cloud computing system, transmitting a status request to the worker process;
• in response to the status request, receiving, from the worker process, a first set of computational results for a first set of tasks that have been completed by the worker process;
• determining, based on the first set of computational results, a second set of tasks that have not been completed the worker process, wherein the first set of tasks and the second set of tasks have been assigned to the worker process;
• generating, for the worker process, a checkpoint data structure that indicates that the first set of tasks has been completed and that the second set of tasks has not been completed;
• transmitting, to be recorded within the cloud computing system, a computational payload comprising the first set of computational results and the checkpoint data structure;
• initiating, within the cloud computing system, a new controller process and a new worker process;
• assigning a new task list to the worker process via the new controller process;
• assigning a first removal priority to the new worker process, the cloud computing system halts and removes worker processes and controller processes based on an assigned removal priority;
• subsequently to assigning the first removal priority, determining that the new worker process has been executing for a given time period;
• and
• based on determining that the new worker process has been executing for the given time period, raising the first removal priority to indicate that the new worker process has become more important than when the first removal priority has been assigned.