fix(nats): explicit consumer config, ack timeouts, unknown-event acking, delivery_count

crates/adapters/nats/src/lib.rs

@@ -10,6 +10,13 @@ const STREAM_SUBJECT: &str = "thoughts-events.>";
 const CONSUMER_NAME: &str = "worker";
 const MAX_MESSAGES: i64 = 100_000;
 
+/// Maximum NATS delivery attempts before a message is considered exhausted.
+pub const CONSUMER_MAX_DELIVER: i64 = 5;
+/// How long NATS waits for an ack before redelivering.
+const CONSUMER_ACK_WAIT_SECS: u64 = 30;
+/// Timeout for spawned ack/nack async tasks.
+const ACK_TASK_TIMEOUT_SECS: u64 = 5;
+
 fn stream_config() -> StreamConfig {
     StreamConfig {
         name: STREAM_NAME.to_string(),
@@ -121,6 +128,10 @@ impl MessageSource for NatsMessageSource {
                     CONSUMER_NAME,
                     jetstream::consumer::pull::Config {
                         durable_name: Some(CONSUMER_NAME.to_string()),
+                        deliver_policy: jetstream::consumer::DeliverPolicy::New,
+                        ack_policy: jetstream::consumer::AckPolicy::Explicit,
+                        ack_wait: std::time::Duration::from_secs(CONSUMER_ACK_WAIT_SECS),
+                        max_deliver: CONSUMER_MAX_DELIVER,
                         // No filter_subject — consume everything from the stream.
                         // filter_subject matching the stream's own wildcard can be
                         // inconsistent across NATS server versions.
@@ -164,25 +175,48 @@ impl MessageSource for NatsMessageSource {
 
                     let subject = msg.subject.to_string();
                     let payload = msg.payload.to_vec();
+                    let delivery_count = msg
+                        .info()
+                        .map(|info| info.delivered.max(0) as u64)
+                        .unwrap_or(1);
                     let msg = Arc::new(msg);
                     let msg_nack = Arc::clone(&msg);
 
                     let raw = RawMessage {
                         subject,
                         payload,
+                        delivery_count,
                         ack: Box::new(move || {
                             let m = Arc::clone(&msg);
                             tokio::spawn(async move {
-                                if let Err(e) = m.ack().await {
-                                    tracing::warn!("NATS ack failed: {e}");
+                                let result = tokio::time::timeout(
+                                    std::time::Duration::from_secs(ACK_TASK_TIMEOUT_SECS),
+                                    m.ack(),
+                                )
+                                .await;
+                                match result {
+                                    Ok(Ok(())) => {}
+                                    Ok(Err(e)) => tracing::warn!("NATS ack failed: {e}"),
+                                    Err(_) => tracing::warn!(
+                                        "NATS ack timed out after {ACK_TASK_TIMEOUT_SECS}s"
+                                    ),
                                 }
                             });
                         }),
                         nack: Box::new(move || {
                             let m = Arc::clone(&msg_nack);
                             tokio::spawn(async move {
-                                if let Err(e) = m.ack_with(AckKind::Nak(None)).await {
-                                    tracing::warn!("NATS nak failed: {e}");
+                                let result = tokio::time::timeout(
+                                    std::time::Duration::from_secs(ACK_TASK_TIMEOUT_SECS),
+                                    m.ack_with(AckKind::Nak(None)),
+                                )
+                                .await;
+                                match result {
+                                    Ok(Ok(())) => {}
+                                    Ok(Err(e)) => tracing::warn!("NATS nack failed: {e}"),
+                                    Err(_) => tracing::warn!(
+                                        "NATS nack timed out after {ACK_TASK_TIMEOUT_SECS}s"
+                                    ),
                                 }
                             });
                         }),