6.2.1 (#130)

* fix: add timeouts to cluster sync commands and backchannel disconnect

CHANGELOG.md

@@ -1,3 +1,7 @@
+## 6.2.1
+
+* Fix cluster failover with paused nodes. 
+
 ## 6.2.0
 
 * Add `Pipeline::try_all`

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "fred"
-version = "6.2.0"
+version = "6.2.1"
 authors = ["Alec Embke <[email protected]>"]
 edition = "2021"
 description = "An async Redis client built on Tokio."

src/protocol/connection.rs

@@ -644,8 +644,13 @@ impl RedisTransport {
 
   /// Send `QUIT` and close the connection.
   pub async fn disconnect(&mut self, inner: &Arc<RedisClientInner>) -> Result<(), RedisError> {
+    let timeout = globals().default_connection_timeout_ms();
     let command: RedisCommand = RedisCommandKind::Quit.into();
-    let _ = self.request_response(command, inner.is_resp3()).await?;
+    let quit_ft = self.request_response(command, inner.is_resp3());
+
+    if let Err(e) = client_utils::apply_timeout(quit_ft, timeout).await {
+      _warn!(inner, "Error calling QUIT on backchannel: {:?}", e);
+    }
     let _ = self.transport.close().await;
 
     Ok(())

src/router/clustered.rs

@@ -1,5 +1,6 @@
 use crate::{
   error::{RedisError, RedisErrorKind},
+  globals::globals,
   interfaces,
   interfaces::Resp3Frame,
   modules::inner::RedisClientInner,
@@ -13,6 +14,7 @@ use crate::{
   },
   router::{responses, types::ClusterChange, utils, Connections, Written},
   types::ClusterStateChange,
+  utils as client_utils,
 };
 use std::{
   collections::{BTreeSet, HashMap},
@@ -552,6 +554,8 @@ pub async fn cluster_slots_backchannel(
   inner: &Arc<RedisClientInner>,
   cache: Option<&ClusterRouting>,
 ) -> Result<ClusterRouting, RedisError> {
+  let timeout = globals().default_connection_timeout_ms();
+
   let (response, host) = {
     let command: RedisCommand = RedisCommandKind::ClusterSlots.into();
 
@@ -561,8 +565,8 @@ pub async fn cluster_slots_backchannel(
       if let Some(ref mut transport) = backchannel.transport {
         let default_host = transport.default_host.clone();
 
-        transport
-          .request_response(command, inner.is_resp3())
+        _trace!(inner, "Sending backchannel CLUSTER SLOTS to {}", transport.server);
+        client_utils::apply_timeout(transport.request_response(command, inner.is_resp3()), timeout)
           .await
           .ok()
           .map(|frame| (frame, default_host))
@@ -582,7 +586,8 @@ pub async fn cluster_slots_backchannel(
       if frame.is_error() {
         // try connecting to any of the nodes, then try again
         let mut transport = connect_any(inner, old_cache).await?;
-        let frame = transport.request_response(command, inner.is_resp3()).await?;
+        let frame =
+          client_utils::apply_timeout(transport.request_response(command, inner.is_resp3()), timeout).await?;
         let host = transport.default_host.clone();
         inner.update_backchannel(transport).await;
 
@@ -594,7 +599,7 @@ pub async fn cluster_slots_backchannel(
     } else {
       // try connecting to any of the nodes, then try again
       let mut transport = connect_any(inner, old_cache).await?;
-      let frame = transport.request_response(command, inner.is_resp3()).await?;
+      let frame = client_utils::apply_timeout(transport.request_response(command, inner.is_resp3()), timeout).await?;
       let host = transport.default_host.clone();
       inner.update_backchannel(transport).await;
 

src/router/responses.rs

@@ -310,7 +310,7 @@ pub fn check_special_errors(inner: &Arc<RedisClientInner>, frame: &Resp3Frame) -
 
 /// Handle an error in the reader task that should end the connection.
 pub fn broadcast_reader_error(inner: &Arc<RedisClientInner>, server: &Server, error: Option<RedisError>) {
-  _debug!(inner, "Ending reader task from {} due to {:?}", server, error);
+  _warn!(inner, "Ending reader task from {} due to {:?}", server, error);
 
   if inner.should_reconnect() {
     inner.send_reconnect(Some(server.clone()), false, None);
@@ -329,7 +329,7 @@ pub fn broadcast_replica_error(inner: &Arc<RedisClientInner>, server: &Server, e
 
 #[cfg(feature = "replicas")]
 pub fn broadcast_replica_error(inner: &Arc<RedisClientInner>, server: &Server, error: Option<RedisError>) {
-  _debug!(inner, "Ending replica reader task from {} due to {:?}", server, error);
+  _warn!(inner, "Ending replica reader task from {} due to {:?}", server, error);
 
   if inner.should_reconnect() {
     inner.send_replica_reconnect(server);