We have a GitLab runner running on one of our Windows 2016 boxes. For the most part it runs fine. However, a problem I encounter a lot when I'm trying to get a new pipeline up and running is that the runner seems to forget that to user PowerShell about half the time.
The GitLab Runner's config.toml looks like this:
concurrent = 1
check_interval = 0
log_level = "debug"
[session_server]
session_timeout = 1800
[[runners]]
name = "Application Template"
url = "https://my-gitlab-instance.com"
token = "mYt0k3n"
executor = "shell"
shell = "powershell"
[runners.custom_build_dir]
[runners.cache]
[runners.cache.s3]
[runners.cache.gcs]
[runners.cache.azure]
Each time I add a new runner I have to manually change the shell = "pwsh" line to shell = "powershell" otherwise it doesn't work at all.
I'm currently working on a new pipeline which includes this deploy job:
deploy-job:
stage: deploy
dependencies:
- build-job
environment: production
script:
- 'Remove-Item -Path "$env:DEPLOY_PATH\*.*"'
- 'xcopy /y /s ".\public\*.*" "$env:DEPLOY_PATH"'
When the pipeline runs, at least 50% of the time this is the result:
See there on line 24 how it says Remove-Item: command not found? That's the problem I keep getting. Thing is, if I rerun the job a few times, it'll eventually work without me changing anything.
I suspect the problem is related to the fact that line 24 begins with bash, indicating that the runner is trying to use Bash (which isn't installed on the server) to execute the script. When the job runs successfully the output is missing bash:
But why is the runner sometimes using Bash to execute the script? I have explicitly told it in the config.toml to use PowerShell. And why would it sometimes use Bash and sometimes not?