feat: add unresponsive timeout for fred (#3369)

Cargo.lock

@@ -2610,9 +2610,9 @@ dependencies = [
 
 [[package]]
 name = "fred"
-version = "7.1.0"
+version = "7.1.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9282e65613822eea90c99872c51afa1de61542215cb11f91456a93f50a5a131a"
+checksum = "b99c2b48934cd02a81032dd7428b7ae831a27794275bc94eba367418db8a9e55"
 dependencies = [
  "arc-swap",
  "async-trait",

config/config.example.toml

@@ -55,7 +55,8 @@ stream_read_count = 1             # Default number of entries to read from strea
 auto_pipeline = true              # Whether or not the client should automatically pipeline commands across tasks when possible.
 disable_auto_backpressure = false # Whether or not to disable the automatic backpressure features when pipelining is enabled.
 max_in_flight_commands = 5000 # The maximum number of in-flight commands (per connection) before backpressure will be applied.
-default_command_timeout = 0       # An optional timeout to apply to all commands.
+default_command_timeout = 30  # An optional timeout to apply to all commands. In seconds
+unresponsive_timeout = 10 # An optional timeout for Unresponsive commands in seconds. This should be less than default_command_timeout.
 max_feed_count = 200       # The maximum number of frames that will be fed to a socket before flushing.
 
 # This section provides configs for currency conversion api

config/deployments/env_specific.toml

@@ -170,7 +170,8 @@ stream_read_count = 1                                                   # Defaul
 auto_pipeline = true                                                    # Whether or not the client should automatically pipeline commands across tasks when possible.
 disable_auto_backpressure = false                                       # Whether or not to disable the automatic backpressure features when pipelining is enabled.
 max_in_flight_commands = 5000 # The maximum number of in-flight commands (per connection) before backpressure will be applied.
-default_command_timeout = 0                                             # An optional timeout to apply to all commands.
+default_command_timeout = 30  # An optional timeout to apply to all commands. In seconds
+unresponsive_timeout = 10 # An optional timeout for Unresponsive commands in seconds. This should be less than default_command_timeout.
 max_feed_count = 200       # The maximum number of frames that will be fed to a socket before flushing.
 cluster_enabled = true                                                  # boolean
 cluster_urls = ["redis.cluster.uri-1:8080", "redis.cluster.uri-2:4115"] # List of redis cluster urls

config/development.toml

@@ -44,7 +44,8 @@ stream_read_count = 1
 auto_pipeline = true
 disable_auto_backpressure = false
 max_in_flight_commands = 5000
-default_command_timeout = 0
+default_command_timeout = 30
+unresponsive_timeout = 10
 max_feed_count = 200
 
 

config/docker_compose.toml

@@ -78,7 +78,8 @@ stream_read_count = 1
 auto_pipeline = true
 disable_auto_backpressure = false
 max_in_flight_commands = 5000
-default_command_timeout = 0
+default_command_timeout = 30
+unresponsive_timeout = 10
 max_feed_count = 200
 
 [cors]

crates/redis_interface/Cargo.toml

@@ -9,7 +9,7 @@ license.workspace = true
 
 [dependencies]
 error-stack = "0.3.1"
-fred = { version = "7.0.0", features = ["metrics", "partial-tracing", "subscriber-client"] }
+fred = { version = "7.1.2", features = ["metrics", "partial-tracing", "subscriber-client", "check-unresponsive"] }
 futures = "0.3"
 serde = { version = "1.0.193", features = ["derive"] }
 thiserror = "1.0.40"

crates/redis_interface/src/lib.rs

@@ -132,6 +132,11 @@ impl RedisConnectionPool {
             },
         };
 
+        let connection_config = fred::types::ConnectionConfig {
+            unresponsive_timeout: std::time::Duration::from_secs(conf.unresponsive_timeout),
+            ..fred::types::ConnectionConfig::default()
+        };
+
         if !conf.use_legacy_version {
             config.version = fred::types::RespVersion::RESP3;
         }
@@ -151,7 +156,7 @@ impl RedisConnectionPool {
         let pool = fred::prelude::RedisPool::new(
             config,
             Some(perf),
-            None,
+            Some(connection_config),
             Some(reconnect_policy),
             conf.pool_size,
         )
@@ -201,6 +206,15 @@ impl RedisConnectionPool {
             }
         }
     }
+
+    pub async fn on_unresponsive(&self) {
+        let _ = self.pool.clients().iter().map(|client| {
+            client.on_unresponsive(|server| {
+                logger::warn!(redis_server =?server.host, "Redis server is unresponsive");
+                Ok(())
+            })
+        });
+    }
 }
 
 struct RedisConfig {

crates/redis_interface/src/types.rs

@@ -57,6 +57,7 @@ pub struct RedisSettings {
     pub max_in_flight_commands: u64,
     pub default_command_timeout: u64,
     pub max_feed_count: u64,
+    pub unresponsive_timeout: u64,
 }
 
 impl RedisSettings {
@@ -76,7 +77,17 @@ impl RedisSettings {
                 "Redis `cluster_urls` must be specified if `cluster_enabled` is `true`".into(),
             ))
             .into_report()
-        })
+        })?;
+
+        when(
+            self.default_command_timeout < self.unresponsive_timeout,
+            || {
+                Err(errors::RedisError::InvalidConfiguration(
+                    "Unresponsive timeout cannot be greater than the command timeout".into(),
+                ))
+                .into_report()
+            },
+        )
     }
 }
 
@@ -97,8 +108,9 @@ impl Default for RedisSettings {
             auto_pipeline: true,
             disable_auto_backpressure: false,
             max_in_flight_commands: 5000,
-            default_command_timeout: 0,
+            default_command_timeout: 30,
             max_feed_count: 200,
+            unresponsive_timeout: 10,
         }
     }
 }