Semaphore Repository Workspace Corruption During Parallel Task Execution

[HTGEK]

Hi Team,

We are observing repository workspace corruption when enabling “Allow Parallel Tasks” for a Semaphore task template executing Ansible + Python workflows from a GitHub repository.

Environment:

• Semaphore Community Edition
• GitHub HTTPS repository
• Parallel GitLab-triggered API requests (~20–30 concurrent)
• Semaphore task template configured with “Allow Parallel Tasks”
• Python scripts executed from cloned repository workspace

Current behavior:
When parallel execution is enabled, tasks intermittently fail during repository update/checkout phase before Ansible/Python execution starts.

Observed errors:

error: cannot open '.git/FETCH_HEAD': No such file or directory
fatal: Unable to read current working directory: No such file or directory
BUG: refs/files-backend.c:3175: initial ref transaction called with existing refs
Failed updating repository: signal: aborted (core dumped)

The issue appears related to concurrent mutation/update of the same repository workspace under:

/opt/semaphore/tmp/project_x/repository_x_template_xx

Questions:

1. What is the recommended Semaphore-supported approach for safe parallel execution against the same Git repository/template?
2. Is there a way to force isolated/fresh repository checkout per parallel task execution?
3. Is there a repository cache/workspace locking configuration recommended for high concurrency?
4. Are there known Git workspace corruption issues in Semaphore Community Edition under parallel task execution?
5. Is there a recommended runner/repository configuration for 20–30 simultaneous API-triggered task executions?

Important context:

• Sequential execution works fine
• Parallelism is a hard requirement for the workflow
• Existing shell-script-based workflow handled parallel requests successfully
• Failure occurs before Ansible/Python logic starts
• The issue appears isolated to Semaphore Git workspace/repository handling under concurrency

Any guidance/recommended production-grade configuration would be greatly appreciated.

Thanks.