nestRemoting causes huge performance changes

[Undrium]

Our project has Users which need access to models further away than one step from them in relationship. A User is part of a group which in its turn has other items.

/Users/uid/Groups/gid/Items/iid

The above URI shouldn't be a problem, right? The users are connected to the groups through another model, otherwise this uses normal relations. Since we only got access to /Users/uid/Groups/gid by default we had to use nestRemoting().

app.models.User.nestRemoting('groups');
app.models.Group.nestRemoting('items');

This worked, but also added 6 seconds of loading time for our application (one request to /Users/uid/Groups/gid/Items/). Now we have pageloads at 7-8 seconds instead of 1-2. Is there any known issues here? We are using the absolutely most simple type of relationship here. Only thing that might cause a problem is the relationship between User and Group which is a simple through model.

[Undrium]

After some deep diving into the wonderful world of debugging of NodeJS I found the perpertrator to be another Boot-script (getCurrentContext-quickfix) used from here: https://gist.github.com/ebarault/fdc39c5d07cc40f08664256f9e00905a

I know alot of people use this as a patch, so might be helpful in the future. Also, I hope this is not patched into the official version (#1495) as it really slows down Loopback.

[davidcheung]

@Undrium interesting and thanks for bringing this up, it would be great help if you could try to upgrade to the latest LB with 4de3aa7, (the implementation is different from the one offered from the gist) and see if there are any performance degradation, or a sandbox so we could have a reference point.

[Undrium]

It seems fine since it does not go through every available model-function injecting the context but lets you do it manually. What should be tested is when this is done to a larger amount of model functions or measure the difference on a single function getting it assigned. This occured to us when we started using the nestRemoting() so should probably be a model which is indirectly related as a child to another from a parent-middlechild-child perspective.

At first glance this looks fine, not entirely sure how much more juice it demands, though.