[js code hints] Kill and restart the codehint worker thread every 30 hint requests. #4025
Conversation
This avoids some memory leaks that we were encountering. Still looking into those, but this should at least stabilize things and prevent the memory from growing so much it crashes brackets. To avoid a noticeable performance hit when we restart the worker, we instead start a new worker, and only switch to it after it has finished intializing (including priming the pump). So when we switch between the old and new worker it should be unoticeable to the user. Had to move around a lot of code in ScopeManager - there are not a lot of actual changes, but the diff is messy due to this.
|
@dloverin sorry about the messy diff - it is mostly code moving around - I will go through and try to highlight my actual changes. |
| numInitialFiles = 0, | ||
| numResolvedFiles = 0, | ||
| numAddedFiles = 0, | ||
| stopAddingFiles = false, |
There was a problem hiding this comment.
Most of the changes in this file are moving everything to a TernWorker object - namely all the promises around adding files, priming the pump, etc. By having those live in 1 Object, we can just create and init a new one when we want to kill the worker and start a new one.
|
This should address the crashing seen in Issue #3977. We should reset the worker before we run out of memory now. |
ghost
assigned 
|
Looks good, just one nit. |
|
I'm guessing if we get partial text updates working for Tern queries we can increase the max hints to a higher number? |
|
@pfarland yes - once we have file fragments we should be able to raise the limit a lot higher based on what we've seen in the tern demo. |
[js code hints] Kill and restart the codehint worker thread every 30 hint requests.
This avoids some memory leaks that we were encountering. Still looking into those, but this should at least stabilize things and prevent the memory from growing so much it crashes brackets.
To avoid a noticeable performance hit when we restart the worker, we instead start a new worker, and only switch to it after it has finished intializing (including priming the pump). So when we switch between the old and new worker it should be unoticeable to the user.
Had to move around a lot of code in ScopeManager - there are not a lot of actual changes, but the diff is messy due to this.