tree 4c43fe8933cff5c9331ad1f3c76699f971bbd092
parent 21a6fe41014db2b9b5647d70e751b268718157d2
author Ari Aye <ariaye@google.com> 1565789357 +0000
committer Alexander Thomas <athom@google.com> 1565877893 +0200

Use a small pool of code completion model isolates

This change makes it so that model can parallelize load over
a handful of isolates when a user is typing very quickly
instead of relying on a single isolate. I played around with
this manually while looking at the diagnostics graphs and found
that model response latency grows with consecutive requests using
a single isolate. Four isolates kept latency constant even as I
typed very quickly.

Change-Id: Id03afc67cb7d6a9e13605393a8658835bd88e4b5
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/113000
Reviewed-by: Brian Wilkerson <brianwilkerson@google.com>