Incomplete tail output from the same file in different containers in Kubernetes emptyDir volume

Question

To test if a file in a emptyDir Volume is synchronized between containers I used tail to observe the same file in two containers, and I stumbled upon the following behavior:

Pod definition:

apiVersion: v1
kind: Pod
metadata:
  name: fortune
spec:
  containers:
    - image: luksa/fortune
      name: html-generator
      volumeMounts:
        - name: html
          mountPath: /var/htdocs
    - image: nginx:alpine
      name: web-server
      volumeMounts:
        - name: html
          mountPath: /usr/share/nginx/html
          readOnly: true
      ports:
        - containerPort: 80
          protocol: TCP
  volumes:
    - name: html
      emptyDir: {}

Example was taken from the book Kubernetes in Action by Marko Luksa. The luksa/fortune image just writes a fortune text to the file /var/htdocs/index.html inside the html-generator container. Every 10 seconds a new file is written in which the content is the output of fortune.

Tailing the same file in both containers outputs sometimes an incomplete response by the web-server container.

Part of the html-generator container output:

kubectl exec -c html-generator -it fortune -- tail -f /var/htdocs/index.html

The very ink with which all history is written is merely fluid prejudice.
                -- Mark Twain

Part of the web-server container output

kubectl exec -c web-server -it fortune -- tail -f /usr/share/nginx/html/index.html

h all history is written is merely fluid prejudice.
                -- Mark Twain

Question: is this caused by

1. tail
2. slow IO speed of the node disk
3. Kubernetes volume sync logic
4. something else?

PS.: I also noted that cURLing the web-service pod port while the index.html is being written to causes nginx to return an empty response body.

Answer 1

The issue with incomplete output from the container is caused by nginx alpine used in pod definition. When you change the image from nginx:alpine to nginx the issue disappears because of different tail binaries used in those images.

Kubernetes volumes sync seems unlikely to cause the issue- as written in emptyDir documentation

By default, emptyDir volumes are stored on whatever medium is backing the node - that might be disk or SSD or network storage, depending on your environment.

The partition created by emptyDir is ephemeral and applications cannot expect any performance SLAs (Disk IOPS for example) from this partition so "2. slow IO speed of the node disk" can also cause such issue but based on reproduction and changing image (what seemed to solve the issue) might be excluded.