Bug: Azure fails to load and run the Docker Container

[entrotech]

Overview

As of 2/4/23, the deployment to Azure in both the Development and Production accounts started failing, preventing release 0.2.43 and updates to the development environment.

Action Items

• GitHub deployment actions appear to be successful, i.e.,
  
• Docker image deployment looks good:
  
• Azure logs from Development Azure account show problem with ingesting the docker image:
  
• The detailed error messsage reads:

2023-02-05T03:42:33.200Z ERROR - failed to register layer: Error processing tar file(exit status 1): Container ID 718322462 cannot be mapped to a host ID

• Found these articles that seems relevant:
  • https://azureossd.github.io/2022/06/30/Docker-User-Namespace-remapping-issues/index.html
  • https://docs.docker.com/engine/security/userns-remap/#user-namespace-known-limitations

Resources/Instructions

REPLACE THIS TEXT -If there is a website which has documentation that helps with this issue provide the link(s) here.

[entrotech]

Started by following the instructions from the first article above:

Started debug session using docker on local machine:

$ docker build -t tdmcalc/tdmapp
...

$ docker run -p 5001:5001 --env-file server/.env tdmcalc/tdmapp
server running on port 5001

Then in a separate terminal window, start a linux shell on the running container:

$ docker ps
CONTAINER ID   IMAGE            COMMAND                  CREATED          STATUS          PORTS                    NAMES
6b2977675949   tdmcalc/tdmapp   "/usr/local/bin/node…"   33 seconds ago   Up 26 seconds   0.0.0.0:5001->5001/tcp   sad_edison

johnc@John-ThinkPad MINGW64 /c/git/hackforla/tdm-calculator (develop)
$ docker exec -it 6b2977 sh
/ # ls
app                package-lock.json
bin                package.json
client             proc
dev                root
etc                run
home               sbin
lib                server.js
media              srv
middleware         sys
mnt                tmp
node_modules       usr
opt                var
/ # 

Per the first article in the Issue Description above, look for files that belong to a user with a UID value that is too high:

find / \( -uid 1000000 \) -ls 2>/dev/null

(Didn't find any)

From the same article, look for an NPM-specific issue where it may install node-modules into the node-modules directory with a too high UID for the file owner/creator:

FILES=$(find /app/node_modules/ ! -user root)
ls -lrta $FILES

In this case, there was voluminous output, but among the results were this:

So the problem appears to be related to the ownership of NPM installed files. The suggested fix is to modify the Dockerfile to change ownership of the files in node_modules:

As a resolution, we can either downgrade NPM (current Docker Images with Node have NPM preinstalled which are an earlier version than 9) - or - change the ownership of node_modules, like the below:

RUN npm i -g [email protected] && \
    npm i && \
    find /app/node_modules/ ! -user root | xargs chown root:root

Next step is to try to figure out how to modify the Dockerfile...

[entrotech]

Changes to the Dockerfile in the PR seemed to work - deployed to DEV environment successfully.