Invalidating half the require.cache could create memory leak.

[thepont]

Im attempting to use express-graphql with express-graphqlexpress-graphql and serverless ofline and am running into issues with graphql schemas already being defined since the require cache for my local files and therefor graphql schemas are invalidated, because the node_modules are not invaldated graphql still knows about my schema, this therefore gives me the error

'Error: Schema must contain unique named types but contains multiple types named "User".

running with the --skipCacheInvalidation resolves this issue for me.

I would also be concerned that this could cause memory leaks ( in situations similar to mine where you are adding references to types you create outside of node_modules to libaries that are persisting and not invalidated inside of node_modules.

Code in question in functionHelper.js

createHandler(funOptions, options) {
  if (!options.skipCacheInvalidation) {
    debugLog('Invalidating cache...');

    for (const key in require.cache) {
      // Require cache invalidation, brutal and fragile.
      // Might cause errors, if so please submit an issue.
      if (!key.match('node_modules')) delete require.cache[key];
    }
  }

See example of creating a memory leak due to the invaldiation of require cache here:

https://github.com/thepont/offline-memory-leak

[dherault]

Hi @thepont, thanks for reporting this.

I've run your repo, and seen the leak.

• Are memory leaks important in dev env ?
• How could we account for changes without invalidating the cache: Nodemon ?

[thepont]

Thanks for the response,

I think as well as the potential leaks there is an issue with the amount of unintended concerquences of this approach, such as my issues with the stale GraphQL schemas.

Currently any node_module that refrences objects made outside of node_modules could accumulate those refrences, this could cause it to serve up stale versions of the object, complain about redefining refences as per my issue or simply just leak memory, these issues could prove quite difficult for users to debug.

Nodemon seems a good alternative, could start the offline processes under nodemon then it could be restarted on each file change.