Home > Software design >  kubernetes pod crashloopbackoff when python script execute in pod
kubernetes pod crashloopbackoff when python script execute in pod

Time:09-06

I would like to have a Python script that simply creates a file run at intervals in Kubernetes. First, I used a docker image to run a python script on a one-time basis, but I got an error.

create_txt.py

import numpy as np
import datetime

res = np.random.rand(1)[0]
res = np.round(res,3) * 1000

with open(f'/home/sjw/kube/{str(int(res))}.txt','w') as f:
    txt = datetime.datetime.now().strftime("%H:%M:%S")
    f.write(txt)

Dockerfile

FROM python:3
WORKDIR /home/sjw/kube
COPY create_txt.py ./
RUN pip install numpy
CMD ["python","./create_txt.py"]

First, I uploaded the image to Docker Hub.

below is manifest

apiVersion: v1
kind: Pod
metadata:
  name: createinterval
spec:
  containers:
  - name: createinterval
    image: idioluck/kube_create:v01
    command: ["/bin/sh"]
    args: ["python create_txt.py"]
    volumeMounts:
    - mountPath: /home/sjw/kube
      name: testvol
  volumes:
  - name: testvol
    hostPath:
      path: /home/sjw/kube
      type: DirectoryOrCreate

pod status is

NAME              READY   STATUS             RESTARTS     AGE
createinterval    0/1     CrashLoopBackOff   7 (90s ago)   12m

I want to finally use cronjob to run a python script that creates a file at regular intervals and saves the result to a connected local storage.

** attach output of kubectl describe pod createinterval

sjw@DESKTOP-O6E7MND:~/kube/docker_sample/kube_create_txt_interval$ kubectl describe pod createinterval
Name:         createinterval
Namespace:    default
Priority:     0
Node:         minikube/192.168.49.2
Start Time:   Mon, 05 Sep 2022 16:42:24  0900
Labels:       <none>
Annotations:  <none>
Status:       Running
IP:           172.17.0.4
IPs:
  IP:  172.17.0.4
Containers:
  createinterval:
    Container ID:   docker://89d2fd4597e445bfd11dace1e06ab325572d2e3072d14df9892b31ebbc7fa7d1
    Image:          idioluck/kube_create:v02
    Image ID:       docker-pullable://idioluck/kube_create@sha256:0868e3dc569c88641a3db05adbf2be9387609f9a0d184869ac939e80b93af5bb
    Port:           <none>
    Host Port:      <none>
    State:          Waiting
      Reason:       CrashLoopBackOff
    Last State:     Terminated
      Reason:       Completed
      Exit Code:    0
      Started:      Mon, 05 Sep 2022 17:08:35  0900
      Finished:     Mon, 05 Sep 2022 17:08:35  0900
    Ready:          False
    Restart Count:  10
    Environment:    <none>
    Mounts:
      /home/sjw/kube from testvol (rw)
      /var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-5n28q (ro)
Conditions:
  Type              Status
  Initialized       True
  Ready             False
  ContainersReady   False
  PodScheduled      True
Volumes:
  testvol:
    Type:          HostPath (bare host directory volume)
    Path:          /home/sjw/kube
    HostPathType:  DirectoryOrCreate
  kube-api-access-5n28q:
    Type:                    Projected (a volume that contains injected data from multiple sources)
    TokenExpirationSeconds:  3607
    ConfigMapName:           kube-root-ca.crt
    ConfigMapOptional:       <nil>
    DownwardAPI:             true
QoS Class:                   BestEffort
Node-Selectors:              <none>
Tolerations:                 node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
                             node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
Events:
  Type     Reason     Age                   From               Message
  ----     ------     ----                  ----               -------
  Normal   Scheduled  27m                   default-scheduler  Successfully assigned default/createinterval to minikube
  Normal   Pulling    27m                   kubelet            Pulling image "idioluck/kube_create:v02"
  Normal   Pulled     26m                   kubelet            Successfully pulled image "idioluck/kube_create:v02" in 4.633892322s
  Normal   Created    25m (x5 over 26m)     kubelet            Created container createinterval
  Normal   Started    25m (x5 over 26m)     kubelet            Started container createinterval
  Normal   Pulled     25m (x4 over 26m)     kubelet            Container image "idioluck/kube_create:v02" already present on machine
  Warning  BackOff    109s (x117 over 26m)  kubelet            Back-off restarting failed container

CodePudding user response:

  1. You don't need to pass command and args in your Pod's manifest. You've already defined these in your Dockerfile.

  2. For running CronJobs in K8s, check out this official doc

  3. Please share logs of your Pod by running:

kubectl logs createinterval

CodePudding user response:

There are several significant problems in your Kubernetes setup; I'll walk through those. You ultimately want to get into a setup like

apiVersion: batch/v1
kind: CronJob
metadata:
  name: createinterval
spec:
  schedule: '* * * * *'
  template:
    metadata:
      name: createinterval
    spec:
      containers:
        - name: createinterval
          image: idioluck/kube_create:v01
      restartPolicy: OnFailure

In your original file:

apiVersion: v1
kind: Pod

You almost never want to create a bare Pod. There are some operational problems with them (you can't change one at all once they're created, for example) and if a node becomes overcommitted a Pod can move into Evicted status and not be replaced. For your use case you want a CronJob; for a more typical long-running server application you'll typically want a Deployment.

    command: ["/bin/sh"]
    args: ["python create_txt.py"]

These overwrite the ENTRYPOINT and CMD in your Dockerfile, respectively, and I'd just delete them. This particular construction is broken and probably the actual cause of your error (though as always double-check the kubectl logs of your pod): it looks for a file named python create_txt.py including a space in the filename, then tries to execute that file as a shell script. If you had to overwrite it then command: [python, create_txt.py] would be the simplest thing that worked.

    volumeMounts:
    - mountPath: /home/sjw/kube
      name: testvol

This directory is also the WORKDIR of your image, which means the volume mount is hiding all of the code in the image. You may be used to a feature of Docker named volumes where image content is copied into a volume on first use; this does not happen on Kubernetes (or for that matter in Docker if the image is updated, or with Docker bind mounts, or ...) and I'd avoid relying on this capability. You should delete this mount.

  volumes:
  - name: testvol
    hostPath: { ... }

A hostPath volume picks up the named directory on whatever node the pod happens to be running on. If the pod is recreated on a different node, the hostPath mount will get a different directory, and the original volume content will be...maybe not "lost" per se, but "misplaced". Again, you almost never want to use hostPath volumes.

... creates a file run at intervals in Kubernetes.

You may want to reconsider this setup. "Files" turn out to be surprisingly hard to manage in Kubernetes. If you look at the list of Types of Volumes and more specifically the table in the PersistentVolume Volume Mode documentation, you'll notice that notice of the volume types that are easier to get only support ReadWriteOnce access; this generally means you can't use the same volume with multiple replicas of your application pod or this cron job and your application at the same time. (Technically it can require all of the replicas to be scheduled on the same node, but they might not fit there and you often want to protect against single-node failure.)

Restructuring this workflow to "create a record in a database" or "make an HTTP call to a special backend endpoint" won't have this problem.

  • Related