Do Kubernetes liveness probes run in parallel with your application?

Question

I have a pod running on Kubernetes for which I am designing a liveness probe. My application reads from a queue (via a loop which continually searches for new messages and executes other functions if it finds one) and is not exposed via HTTP, so I need a command liveness probe. I am pondering whether a simple implementation would work:

livenessProbe:
      exec:
        command:
        - cat
        - /tmp/healthy

However, I'm unsure whether the cat would succeed even if the application was 'stuck' at some point in the loop - the file would still be there. This comes down to a fundamental lack of understanding of liveness probes which I was unable to find in the documentation - presumably they run somehow in series with your application so if your app is not running, the command cannot be executed? But I am not confident on this point.

If the command can be executed in parallel then I believe I will need some kind of timestamp check where I update a file on each loop and the liveness probe checks its timestamp. If the first way works it is simpler, but can anyone confirm if this is the case? Thanks.

Edit: my app code. I added in the sleep(60)s to try and test whether the liveness probe would fail if the file hadn't been updated in a minute, but they wouldn't be part of the normal app code.

INITIALISATION CODE

with open('loaded.txt','w') as f:          # readiness probe = check this file exists
    f.write('loaded')

current_backoff = 0
    max_backoff = 10
    while True:
        if current_backoff < max_backoff:
            current_backoff +=1
        with open('loaded.txt','w') as f:
            f.write('loaded')
            sleep(60)

        messages = input_queue_client.receive_messages(visibility_timeout=100)
        for message in messages:
            with open('loaded.txt','w') as f:
                f.write('loaded')
            sleep(60)
            current_backoff = 0
 
            CODE TO PROCESS MESSAGES

        sleep(current_backoff)

My liveness probe attempts:

1.

        livenessProbe:
          exec:
            command:
              - find
              - /var/app/loaded.txt
              - -mmin 
              - '+0.1'
          initialDelaySeconds: 10
          periodSeconds: 10

2. (command returns failure if anything is returned from find, otherwise cat the file)

        livenessProbe:
          exec:
            command:
              - find
              - /var/app/loaded.txt
              - -mmin 
              - '+0.1'
              - -exec
              - cat
              - '/var/app/loaded.txt{}'
              -  ;
          initialDelaySeconds: 10
          periodSeconds: 10

3. (command returns failure if anything is returned from find, otherwise return nothing)

        livenessProbe:
          exec:
            command:
              - find
              - /var/app/loaded.txt
              - -mmin 
              - '+0.1'
              - -exec
              - if[[{}]]
              - ;
          initialDelaySeconds: 10
          periodSeconds: 10

I have also tried all of these with - instead of +. The probe never fails despite the very short window (which will eventuallly be longer!) and the sleep command.

Answer 1

Liveness probing done by kubelet in each node. And yes, it runs in parallel with your application.

In you case, you could touch /tmp/healthy file each time you start new iteration in loop. And use command like find /tmp/health -mmin +0.5 in health check. This command returns nothing if file is older than half a minute. If health check command returns nothing it's assumed that check is passing.