Features

A task runs as an OS process
Tasks queue with priority and process limit per node are supported
Fully database driven (a workers and tasks are controlled through the table rows)
A task can be cancelled result in the OS process killed (see task lifecycle)
A task can be scheduled at a time or with interval
Uses postgres NOTIFY feature in opposite to pooling => instant changes discovery, less requests
Clustering is supported
Failover is supported

Class diagramm

Create new task in the Draft state

INSERT INTO long_task.task(group_id, state_id, priority, command)
VALUES (0, 'DR', 0, '{ping,-n,1,127.0.0.1}')
RETURNING id;

Queue draft task for ASAP executing by any worker in the group

UPDATE long_task.task
SET state_id='AW', worker_id=null
WHERE id = <id>

Re-queue a completed task for ASAP executing by any worker in the group

UPDATE long_task.task
SET state_id='AW', worker_id=null
WHERE id = <id> AND state_id like 'C%'

Cancel a queued or executing task

UPDATE long_task.task
SET state_id='AC'
WHERE id = <id> AND state_id like 'A%'

Schedule existing task for each 2 hours from now (start will be skipped if state_id is not like 'C%')

UPDATE long_task.task
SET 
	next_start=now(),
	shed_period_id='HOU',
	shed_period_count=2,
	shed_clone=false
WHERE id = <id>

Schedule existing task for each 10 seconds that will be cloned and queued each time

UPDATE long_task.task
SET 
	next_start=now(),
	shed_period_id='SEC',
	shed_period_count=10,
	shed_clone=false
WHERE id = <id>

git clone https://github.com/ivanovrvl/pg_tasks.git
pip install -r requirements.txt
Apply liquibase script db/changelog.xml: use update_db.bat for help (backup stored procedures customized before)
Restore stored procedures customized, adotp them if needed
Change db_config.json, config.py if needed
Run Nodes with appropriate command line parameters (see config.py):
python node.py [<worker_id> [<group_id> [<max_task_count> [<node_name>]]]]