Some potential data racing issues

[ihciah]

1. On the same instance, we operate on the same git index concurrently, which may cuase the results be interdependent. The solution I can think about is to hold a mutex when write the git repo.
2. Between different instances, git repo is not "refresh"ed. I noticed the refresh call is commented. However, what I want to remind is, even if you refresh or do some git pull --rebase, there could still exist some conflicts that can not be fixed automatically. Maybe you should retry{ modify files?; git add and push?; break; }.

And there's another problem when using git command: The response text and status are not used which causes ignorence of git push failure. For more robustness, maybe a reset --hard can be used on push failure to make sure the failure won't last.