Why does kubectl logs --follow show duplicate output when used in a loop to watch Kubernetes Job logs?

I'm running a Kubernetes Job from a Jenkins pipeline and want to stream its logs until completion.

I currently use this pattern in a Bash script:

job_status_cmd_complete="kubectl get job ${kube_job_name} --namespace=${kube_namespace} -o jsonpath={..status.conditions[?(@.type=='Complete')].status}"
job_status_cmd_failed="kubectl get job ${kube_job_name} --namespace=${kube_namespace} -o jsonpath={..status.conditions[?(@.type=='Failed')].status}"

function jobIsFinished {
  if [[ $($job_status_cmd_complete) == 'True' || $($job_status_cmd_failed) == 'True' ]]; then
    echo 'True'
  else
    echo 'False'
  fi
}

# First log stream — immediately after pod is ready
kubectl logs --tail=-1 --follow --selector=job-name=$job_name --namespace=$namespace -c $job_name

# Loop to check for completion
while [[ $(jobIsFinished) != 'True' ]]; do
  # Second log stream to catch more output in case the job is still running
  kubectl logs --tail=100 --follow --selector=job-name=$job_name --namespace=$namespace -c $job_name
done

Why I do this:

1. The first kubectl logs --follow is to get logs as soon as the pod is ready.
2. The second one in the loop is a safeguard to continue watching if the job is long-running or restarts.
3. This was intended to ensure I don't miss logs if the first stream gets disconnected or if the pod is slow to start logging.

Issue:

1. This approach sometimes causes duplicate logs, especially in Jenkins output.

2. It's not clear whether --follow maintains log offset or restarts from the beginning in each call.

Is kubectl logs --follow stateless and prone to duplication if used multiple times?