Question

When we build web software releases, we prepare the release in a temporary directory and then replace the release directory with the temporary one just prepared:

# Move and replace existing release directory.
mv /path/to/httpdocs /path/to/httpdocs.before
mv /path/to/$newReleaseName /path/to/httpdocs

Under this scheme, it happens that with about 1 in every 15 releases, a user was using a file in the original release directory exactly at the time the commands above are run, and a fatal error occurs for that user.

I am wondering if using symlinking like follows would be significantly faster, in terms of processing time, thereby helping to lessen the likelihood of this problem:

# Remove and replace existing release symlink.
ln -sf /path/to/$newReleaseName path/to/httpdocs

Answer 1

The main advantage of using the symlink is that you can do it atomically. Forcing a new symlink is not atomic, but you can create a new symlink and then rename the it to the existing one. The rename is an atomic operation, so you should have no race condition there.

ln -s /path/to/newrelease /path/to/httpdocs.tmp && mv -Tf /path/to/httpdocs.tmp /path/to/httpdocs