Active queue management in a multi-node computing environment

A system configured for processing computing jobs of a managed network that is associated with a computational instance of a remote network management platform, the computing system comprising:

• a database job queue disposed within the computational instance and configured for queuing the jobs of the managed network;
• a server device configured to receive requests from one or more computing devices of the managed network for processing of computing jobs, and enqueue the computing jobs in the database job queue;
• and
• a plurality of worker nodes of the computational instance, each being comprised in a computing device;
• wherein each of the plurality of worker nodes is configured to carry out respective operations of at least a respective scheduler thread and a respective pool of worker threads, the respective operations including:
• causing the respective scheduler thread to wake up from a respective current sleep state upon expiration of a respective sleep-interval timer, the respective sleep-interval timer having a respective duration;
• after waking up, the respective scheduler thread determining a respective current number of jobs in a respective in-memory job queue that are waiting for processing by a worker thread of the respective pool;
• based at least in part on the respective current number of jobs in the respective in-memory job queue, the respective scheduler thread computing, over a respective, current sliding time-window of a width including the respective current sleep state, a respective job-completion rate of jobs processed by the respective pool of worker threads;
• based on the computed respective job-completion rate, the respective scheduler thread performing at least one of four scheduling tasks, the four scheduling tasks being:
• (i) retrieving one or more jobs from the database job queue and adding the one or more retrieved jobs to the respective in-memory job queue, (ii) removing one or more jobs from the respective in-memory job queue and returning the one or more removed jobs to the database job queue, (iii) leaving the respective in-memory job queue unchanged, or (iv) adjusting the respective duration of the respective sleep-interval timer;
• and
• causing the respective scheduler thread to transition to a new, respective sleep state for the respective duration of the respective sleep interval.

A method for processing computing jobs of a managed network that is associated with a computational instance of a remote network management platform, the method carried out by a computing device of the computational instance, and comprising:

• implementing a scheduler thread and a pool worker threads;
• causing the scheduler thread to wake up from a current sleep state upon expiration of a sleep-interval timer, the sleep-interval timer having a duration;
• after waking up, the scheduler thread determining a current number of jobs in an in-memory job queue in the memory that are waiting for processing by a worker thread of the pool;
• based at least in part on the current number of jobs in the in-memory job queue, the scheduler thread computing, over a current sliding time-window of a width including the current sleep state, a job-completion rate of jobs processed by the pool of worker threads;
• based on the computed job-completion rate, the scheduler thread performing at least one of four scheduling tasks, the four scheduling tasks being:
• (i) retrieving one or more jobs from a database job queue of the computational instance and adding the one or more retrieved jobs to the in-memory job queue, (ii) removing one or more jobs from the in-memory job queue and returning the one or more removed jobs to the database job queue, (iii) leaving the in-memory job queue unchanged, or (iv) adjusting the duration of the sleep-interval timer;
• and
• causing the scheduler thread to transition to a new sleep state for the duration of the sleep interval timer.