Debugging a failing GitLab pipeline job

Problem

A GitLab pipeline job failed, but the error is not reproducible outside the pipeline.

Steps

1. Add debugging code to the pipeline in a pipeline-definition MR that will dump the environment and sleep in the case of failure, e.g. add the following in pipeline/stages/build.yml:

   if ! rpmbuild --rebuild ...; then
     environment_file=$(mktemp)
     cki_echo_error "Build failed"
     cki_echo_notify "dumping environment to $environment_file"
     export > "$environment_file"
     cki_echo_notify "sleeping"
     sleep infinity
     exit 1
   fi
   

2. From the kernel repository MR, get the ID for the failed pipeline in the pipeline repositories. Then, in the pipeline-definition MR, retrigger that pipeline with the debug code via the bot with something like

   @cki-ci-bot, please test [baseline/12345678]
   

3. Wait until the pipeline hits the sleep. In the failing job log, determine the gitlab-runner and spawned EC2 machine from lines like

   Running with ...
   on ...-aws-internal-a-...
   ...
   Running on ... via runner-abcdef-arr-cki.prod.general.1234567-123abcdef...
   

4. Use the script ansible_ssh.sh from the deployment-all repository checkout to access the appropriate EC2 instance. Select arr-cki-prod-internal-runner-us-east-1a.infra.cki-project.org if the pipeline runs in ...-aws-internal-a-... or arr-cki-prod-internal-runner-us-east-1b.infra.cki-project.org if the pipeline runs in ...-aws-internal-b-....

5. Log into the runner

   sudo docker-machine ssh runner-abcdef-arr-cki.prod.general.1234567-123abcdef
   

6. The docker container can then be entered via

   sudo docker ps
   sudo docker exec -it container-id /bin/bash