job: retry on failure, general refactor

src/job/mod.rs

@@ -1,47 +1,88 @@
 use actix_rt::Arbiter;
+use actix_web::web;
 use async_trait::async_trait;
-use sqlx::Execute;
-use std::{cell::RefCell, fmt, sync::Mutex};
-use tokio::sync::mpsc::{channel, Receiver, Sender};
+use std::sync::atomic::{AtomicUsize, Ordering};
+use std::sync::Weak;
+use std::time::Duration;
+use serde::{Deserialize, Serialize};
+use tokio::time::sleep;
 
 use crate::core::*;
-use crate::state::AppState;
+use crate::state::{AppState, State};
 
+const MAX_RETRIES: u32 = 3;
+const ERROR_DELAY: Duration = Duration::from_secs(5 * 60);
+
+/// A background job to be executed asynchronously.
+///
+/// Jobs run in a separate execution context, detached from the one that
+/// enqueued it (which is usually a request handler).  This mechanism allows
+/// for fast response times to requests and automatic retries on failures
+/// such as network errors when a remote instance is currently offline.
 #[async_trait]
-pub trait Job: Send + Sync + fmt::Debug {
-    async fn perform(&self) -> Result<()>;
+pub trait Job: Send + Sync + Serialize + Deserialize<'static> {
+    /// Execute the errand.
+    ///
+    /// This will be called by the scheduler in a background thread.
+    /// If the result is an error, the scheduler may retry it again later.
+    async fn perform(&self, state: &AppState) -> Result<()>;
+
+    /// Return this job's name.
+    ///
+    /// Currently only used for logging.
     fn name(&self) -> &'static str;
 }
 
 pub struct Scheduler {
     arbiter: Arbiter,
-    sender: Sender<Box<dyn Job>>,
+    counter: AtomicUsize,
+    state: Weak<State>,
 }
 
 impl Scheduler {
-    pub fn new() -> Scheduler {
-        let (tx, rx) = channel(32);
-        let sched = Scheduler {
+    pub fn new(state: Weak<State>) -> Scheduler {
+        Scheduler {
             arbiter: Arbiter::new(),
-            sender: tx,
-        };
-
-        if !sched.arbiter.spawn(worker(rx)) {
-            panic!("Job scheduler initialization failed");
+            counter: AtomicUsize::new(1),
+            state,
         }
-        sched
     }
 
-    pub async fn schedule(&self, job: Box<dyn Job>) {
-        self.sender.send(job).await.expect("Job send failed");
-    }
-}
-
-async fn worker(mut receiver: Receiver<Box<dyn Job>>) {
-    while let Some(job) = receiver.recv().await {
-        debug!(target: "job", "Starting job {}", job.name());
-        if let Err(e) = job.perform().await {
-            error!(target: "job", "Service {} failed: {}", job.name(), e);
+    pub fn schedule<T>(&self, job: T)
+    where
+        T: 'static + Job,
+    {
+        let id = self.counter.fetch_add(1, Ordering::Relaxed);
+        let state: AppState = web::Data::from(
+            self.state
+                .upgrade()
+                .expect("Zombie scheduler (this is supposed to be impossible)"),
+        );
+        if !self.arbiter.spawn(worker(job, id, state)) {
+            panic!("Arbiter has died");
+        }
+    }
+}
+
+async fn worker<T>(job: T, id: usize, state: AppState)
+where
+    T: Job,
+{
+    let name = job.name();
+    for n in 1..=MAX_RETRIES {
+        debug!(target: "job", "Starting job {name}#{id}");
+
+        if let Err(e) = job.perform(&state).await {
+            warn!(target: "job", "Job {name}#{id} failed (try {n}/{MAX_RETRIES}): {e}");
+
+            if n == MAX_RETRIES {
+                error!(target: "job", "Giving up on job {name}#{id} (retry limit reached)");
+            } else {
+                sleep(ERROR_DELAY).await;
+            }
+        } else {
+            debug!(target: "job", "Job {name}#{id} succeeded");
+            break;
         }
     }
 }

src/state.rs

@@ -12,13 +12,12 @@ pub struct State {
     pub sched: Scheduler,
 }
 
-pub type AppStateRaw = Arc<State>;
-pub type AppState = web::Data<AppStateRaw>;
+pub type AppState = web::Data<State>;
 
 pub fn new(config: Config, db_pool: PgPool) -> AppState {
-    web::Data::new(Arc::new(State {
+    web::Data::from(Arc::new_cyclic(|ptr| State {
         config,
         repo: AppRepo::new(db_pool),
-        sched: Scheduler::new(),
+        sched: Scheduler::new(ptr.clone()),
     }))
 }