Remove blocking IO #1803
Remove blocking IO #1803
Conversation
|
|
||
| impl Default for RustReleasePost { | ||
| fn default() -> Self { | ||
| Self(Default::default(), Instant::now()) |
There was a problem hiding this comment.
I think this is consistent with the previous behavior: If no version (or release post) can be fetched, an empty string is used instead. Not sure how much sense that makes though.
| teams: RustTeams::get(teams_cache) | ||
| .await | ||
| .0 | ||
| .ok_or_else(|| Box::<dyn Error + Send + Sync>::from("failed to load teams"))?, |
There was a problem hiding this comment.
The teams data does currently cause an error if it cannot be fetched (unlike rust version and release post) so I used an option in the cache type to indicate there might not be any teams data.
| let rust_version = RustVersion::fetch().await.unwrap_or_default(); | ||
| let rust_release_post = RustReleasePost::fetch().await.unwrap_or_default(); | ||
| let teams = RustTeams::fetch().await.unwrap_or_default(); |
There was a problem hiding this comment.
This is a change in behavior, the caches will now already be prefilled with data at startup. Previously, I think the first request using a cache would actually block until some data could be retrieved.
There was a problem hiding this comment.
I think that's a good call, yeah.
src/main.rs
Outdated
| async fn index() -> Template { | ||
| render_index(ENGLISH.into()).await | ||
| async fn index( | ||
| version_cache: &State<Arc<RwLock<RustVersion>>>, |
There was a problem hiding this comment.
nit: can we use some kind of typedef for this?
There was a problem hiding this comment.
trait Cached + type Cache<T> = ... would probably be fine. Open to other opinions
There was a problem hiding this comment.
Yeah that's cleaner 👍
| let rust_version = RustVersion::fetch().await.unwrap_or_default(); | ||
| let rust_release_post = RustReleasePost::fetch().await.unwrap_or_default(); | ||
| let teams = RustTeams::fetch().await.unwrap_or_default(); |
There was a problem hiding this comment.
I think that's a good call, yeah.
I had another go at removing the remaining blocking IO in
cache.rs. The main problem I had previously encountered was that passing function pointers toasyncfunctions is very messy. I usedasync_traitto not have to do that in the first place. In the process, I switched over to Rocket's managed state. This reduces complexity incache.rsand theoretically even improves performance a bit, since there are now three separate caches, with separateRwLocks, which should reduce lock contention.