Configuring Core Dumps

Challenges

Linux allows you to create a core dump file if an application crashes, which contains the data the application had in memory at the time of the crash. You can analyze the file to locate the fault.

Generally, when a service application crashes, its container exits and is reclaimed and destroyed. Therefore, container core files need to be permanently stored on the host or cloud storage. This topic describes how to configure container core dumps.

Enabling Core Dump on a Node

Log in to the node, run the following command to enable core dump, and set the path and format for storing core files:

echo "/tmp/cores/core.%h.%e.%p.%t" > /proc/sys/kernel/core_pattern

Parameters:

• %h: host name (or pod name). You are advised to configure this parameter.
• %e: program file name. You are advised to configure this parameter.
• %p: (optional) process ID.
• %t: (optional) time of the core dump.

You can also configure a pre-installation or post-installation script to automatically run this command when creating a node.

Permanently Storing Core Dumps

apiVersion: v1
kind: Pod
metadata:
  name: coredump
spec:
  volumes:
  - name: coredump-path
    hostPath:
      path: /home/coredump
  containers:
  - name: ubuntu
    image: ubuntu:12.04
    command: ["/bin/sleep","3600"]
    volumeMounts:
    - mountPath: /tmp/cores
      name: coredump-path

Create a pod using kubectl.

kubectl create -f pod.yaml

Verification

After the pod is created, access the container and trigger a segmentation fault of the current shell terminal.

$ kubectl get pod
NAME                          READY   STATUS    RESTARTS   AGE
coredump                      1/1     Running   0          56s
$ kubectl exec -it coredump -- /bin/bash
root@coredump:/# kill -s SIGSEGV $$
command terminated with exit code 139

Log in to the node and check whether a core file is generated in the /home/coredump directory. The following example indicates that a core file is generated.

# ls /home/coredump
core.coredump.bash.18.1650438992