Is it safe to update a symlink to a jar while the process is running?

So I am thinking of setting up my deployment scripts for my processes to point to a a symlink which points to the latest version of the jar for my application.

So:

symlink x -> /path/to/jar/version_1.0.0

Now the script for my process will load the jar be referencing symlink x in it's start script.

Now my question is:

Given the process is still running...can I safely update the symlink x to be:

symlink x -> /path/to/jar/version_1.2.0

I won't be removing /path/to/jar/version_1.0.0 so that will still remain on the host.

My concern is that the process may start loading jars that are not in memory yet from the new symlink path rather than the /path/to/jar/version_1.0.0 path the script first ran with.

It all depends when you update the symlink and when you load your jar.
When you create a symlink, a new inode is allocated, which means your symlink is a different entity than your original file.

When you update (unlink + link again) your symlink, it basically creates a new inode. So, if your first symlink was already called and used, the new one will not be used anymore, unless is called again. Although they have the same path, they are different resources at the file system level.

In general, I think your concern is correct.
JVM loads classes dynamically, when they’re really needed.

So its possible that when the process runs some classes are not being loaded yet (say, you “import” some classes for a very specific business flow that has not been triggered yet since the start of the current run).
So if you’ll change the link its possible that some things will break.

So if you absolutely need to provide an upgrade “without restarts” you better do it in a different way:

• Start two JVMs simultaneously: one will run the old process and one will run the newer process
• Start when the new version is “ready” start routing the traffic to it (ok, it really depends on what your application really does)
• When you see it works as expected - stop the old app.

Now this really depends on how does your application really look like, you might need to sync the state of DBs, execute the requests with two versions of the application at the same time, and so forth, so in a general case this can be pretty complicated thing to do.

Some modern advanced tools (like kubernetes) can help in DevOps area here but still you’ll have to understand what exactly happens in the application to achieve the best results.