Strange cyclic 1-100ms delay in start processing jobs added by upsertJobScheduler

[echoi92]

As mentioned in title, jobs added by upsertJobScheduler seems to have cyclic delay between 1-100ms every ~20s window where the delay gradually increases until it drops to ~0 again.

The above image is showing processDelay value of the code below.

const queue = new Queue('test-queue', {
  defaultJobOptions: {
    removeOnComplete: 10,
    removeOnFail: 10,
  },
  connection: jobRedis,
})

const worker = new Worker(
    queue.name,
    async (job) => {
      const name = job.name
      const startTime = Date.now()
      const scheduledOn =
        _.get(job, 'timestamp') + _.get(job, ['opts', 'delay'], 0)
      const info = {
        jobId: job.id,
        registeredOn: scheduledOn,
        start: startTime,
        processDelay: startTime - scheduledOn,
        name: job.name,
        queueName: job.queueName,
      }
      logger.info('bullmq test', info)
    },
    {
      connection: jobRedis,
      concurrency: 10,
    },
  )
  
  queue.upsertJobScheduler(
    '1s-test',
    {
      every: 1_000,
    },
    {
      name: '1s-test',
    },
  )

if the job is added directly and not through upsertJobScheduler, processDelay is kept under 20ms (99th percentile)

I'm curious if bullmq job scheduler has cyclic variability for better distribution of jobs? I could not find any clues to this in the doc.