This is a reproduction showing an issue that happens in Vite Dev server SSR environments when dependencies get optimized while a request is being handled.
The problem is that when new dependencies are optimized a fullReload() is triggered (which invalidates the environmen's whole module graph). Although this is the generally the expected behavior, this might be problematic when dependencies are discovered while a request is being handled. In such case the state of the environment will be different before and after the reload introducing the risk of inconsistent state, potentially causing problems in the application.
To run the minimal reproduction simply run:
pnpm i
pnpm dev
and navigate to dev server's URL, you should see an error such as this:
Refreshing the page will cause the error to go away, running
pnpm dev
makes the error reappear (since a predev script deletes the node-modules/.vite directories)
running
pnpm vite dev
afterwords does not trigger the error (since this only happens then dependencies are getting optimized).
The minimal reproduction presented here might make it look like this is not a realistic issue to run into, but
I think that it might actually be pretty common. The ssr-react variant produces this issue just by using react-server/dom
in a very common and basic way.
In the reproduction this error is triggered by not providing any entrypoints in optimizeDeps.entries, meaning that there is no initial crawling and dependencies are optimized as they are encountered.
The issue can however be triggered even when there is an initial crawling process, simply by using virtual modules since those are not part of the initial crawling process, you can see an example of this in the virtual-modules variant.
Note
This is actually the main issue here since many fullstack frameworks rely on virtual modules to control how the user's application is run