Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Summary: ## This stack Track and down methods that are holding up the reactor and optimize them. ## This diff This code is loading both sides of a diff from blobstore concurrently is the right thing. In practice, most of the cost comes from the Thrift deserialization, which is synchronous. Worse, `try_join` does not yield back to Tokio. The combination means blocking the reactor for 150-200ms or even longer. With this diff we'll actually run them in parallel, saving both max poll time and quite likely wall clock time too. Reviewed By: markbt Differential Revision: D66761433 fbshipit-source-id: 5d44506a453697d7a45afe3bd811152ba0e21982
- Loading branch information