0064.yml

version: 64
description: Add `queue_worker_seen` and `queue_task_queue_seen` functions
methods:
  create_task_queue: {deprecated: true}
  update_task_queue: {deprecated: true}
  task_queue_seen:
    description: |-
      Recognize that a task queue has been seen, creating it if necessary, updating
      its properties if not null, and in any case bumping its last seen time time. 
      The expiration time is not allowed to move backward.
      
      This function always writes to the DB, so calls should be suitably rate-limited at the
      client side.
    mode: write
    serviceName: queue
    args: task_queue_id_in text, expires_in timestamptz, description_in text, stability_in text
    returns: void
    body: |-
      begin
        insert
          into task_queues (task_queue_id, expires, last_date_active, description, stability)
          values (
            task_queue_id_in,
            expires_in,
            now(),
            coalesce(description_in, ''),
            coalesce(stability_in, 'experimental')
          )
          on conflict (task_queue_id) do update
          set
            expires = greatest(coalesce(expires_in, task_queues.expires), task_queues.expires),
            last_date_active = now(),
            description = coalesce(description_in, task_queues.description),
            stability = coalesce(stability_in, task_queues.stability)
          where task_queues.task_queue_id = task_queue_id_in;
      end
  update_queue_worker_tqid: {deprecated: true}
  create_queue_worker_tqid: {deprecated: true}
  queue_worker_seen:
    description: |-
      Recognize that a worker has been seen by the queue, creating it if necessary.  This is called
      when workers claim or re-claim work.  The expiration time is not allowed to move backward.
      
      This function always writes to the DB, so calls should be suitably rate-limited at the
      client side.
    mode: write
    serviceName: queue
    args: task_queue_id_in text, worker_group_in text, worker_id_in text, expires_in timestamptz
    returns: void
    body: |-
      begin
        insert
          into queue_workers (task_queue_id, worker_group, worker_id, quarantine_until, expires, first_claim, recent_tasks)
          values (
            task_queue_id_in,
            worker_group_in,
            worker_id_in,
            now() - interval '10 years',
            expires_in,
            now(),
            jsonb_build_array()
          )
          on conflict (task_queue_id, worker_group, worker_id) do update
          set
            expires = greatest(coalesce(expires_in, queue_workers.expires), queue_workers.expires)
          where
            queue_workers.task_queue_id = task_queue_id_in and
            queue_workers.worker_group = worker_group_in and
            queue_workers.worker_id = worker_id_in;
      end
  quarantine_queue_worker:
    description: |-
      Update the quarantine_until date for a worker.  The Queue service interprets a date in the past
      as "not quarantined".  This function also "bumps" the expiration of the worker so that un-quarantined
      workers do not immediately expire.  Returns the worker row just as get_queue_worker would, or no rows if
      no such worker exists.
    mode: write
    serviceName: queue
    args: task_queue_id_in text, worker_group_in text, worker_id_in text, quarantine_until_in timestamptz
    returns: table(task_queue_id text, worker_group text, worker_id text, quarantine_until timestamptz, expires timestamptz, first_claim timestamptz, recent_tasks jsonb)
    body: |-
      begin
        return query update queue_workers
        set
          quarantine_until = quarantine_until_in,
          expires = greatest(queue_workers.expires, now() + interval '1 day')
        where
          queue_workers.task_queue_id = task_queue_id_in and
          queue_workers.worker_group = worker_group_in and
          queue_workers.worker_id = worker_id_in
        returning
          queue_workers.task_queue_id,
          queue_workers.worker_group,
          queue_workers.worker_id,
          queue_workers.quarantine_until,
          queue_workers.expires,
          queue_workers.first_claim,
          queue_workers.recent_tasks;
      end
  queue_worker_task_seen:
    description: |-
      Update the worker record to indicate that this task run was seen there.  The
      task run should be a JSON object with keys `taskId` and `runId`.  This will
      add the task to `recent_tasks`, keeping the most recent 20 tasks. This
      will do nothing, but not fail, if the worker does not exist, as it is
      unusual for a nonexistent worker to claim work.
    mode: write
    serviceName: queue
    args: task_queue_id_in text, worker_group_in text, worker_id_in text, task_run_in jsonb
    returns: void
    body: |-
      begin
        update queue_workers
        set
          -- append without increasing size over 20
          recent_tasks = case
            when jsonb_array_length(recent_tasks) > 19 then (recent_tasks - 0)
            else recent_tasks
          end || jsonb_build_array(task_run_in)
        where
          queue_workers.task_queue_id = task_queue_id_in and
          queue_workers.worker_group = worker_group_in and
          queue_workers.worker_id = worker_id_in;
      end