fix: configure HTTP/2 keep-alive for heartbeat client to detect network failures faster (#7344)

* fix: configure HTTP/2 keep-alive for heartbeat client to detect network failures faster

Signed-off-by: WenyXu <wenymedia@gmail.com>

* chore: apply suggestions from CR

Signed-off-by: WenyXu <wenymedia@gmail.com>

---------

Signed-off-by: WenyXu <wenymedia@gmail.com>

config/config.md

@@ -294,7 +294,6 @@
 | `meta_client` | -- | -- | The metasrv client options. |
 | `meta_client.metasrv_addrs` | Array | -- | The addresses of the metasrv. |
 | `meta_client.timeout` | String | `3s` | Operation timeout. |
-| `meta_client.heartbeat_timeout` | String | `500ms` | Heartbeat timeout. |
 | `meta_client.ddl_timeout` | String | `10s` | DDL timeout. |
 | `meta_client.connect_timeout` | String | `1s` | Connect server timeout. |
 | `meta_client.tcp_nodelay` | Bool | `true` | `TCP_NODELAY` option for accepted connections. |
@@ -457,7 +456,6 @@
 | `meta_client` | -- | -- | The metasrv client options. |
 | `meta_client.metasrv_addrs` | Array | -- | The addresses of the metasrv. |
 | `meta_client.timeout` | String | `3s` | Operation timeout. |
-| `meta_client.heartbeat_timeout` | String | `500ms` | Heartbeat timeout. |
 | `meta_client.ddl_timeout` | String | `10s` | DDL timeout. |
 | `meta_client.connect_timeout` | String | `1s` | Connect server timeout. |
 | `meta_client.tcp_nodelay` | Bool | `true` | `TCP_NODELAY` option for accepted connections. |
@@ -629,7 +627,6 @@
 | `meta_client` | -- | -- | The metasrv client options. |
 | `meta_client.metasrv_addrs` | Array | -- | The addresses of the metasrv. |
 | `meta_client.timeout` | String | `3s` | Operation timeout. |
-| `meta_client.heartbeat_timeout` | String | `500ms` | Heartbeat timeout. |
 | `meta_client.ddl_timeout` | String | `10s` | DDL timeout. |
 | `meta_client.connect_timeout` | String | `1s` | Connect server timeout. |
 | `meta_client.tcp_nodelay` | Bool | `true` | `TCP_NODELAY` option for accepted connections. |

config/datanode.example.toml

@@ -99,9 +99,6 @@ metasrv_addrs = ["127.0.0.1:3002"]
 ## Operation timeout.
 timeout = "3s"
 
-## Heartbeat timeout.
-heartbeat_timeout = "500ms"
-
 ## DDL timeout.
 ddl_timeout = "10s"
 

config/flownode.example.toml

@@ -78,9 +78,6 @@ metasrv_addrs = ["127.0.0.1:3002"]
 ## Operation timeout.
 timeout = "3s"
 
-## Heartbeat timeout.
-heartbeat_timeout = "500ms"
-
 ## DDL timeout.
 ddl_timeout = "10s"
 

config/frontend.example.toml

@@ -226,9 +226,6 @@ metasrv_addrs = ["127.0.0.1:3002"]
 ## Operation timeout.
 timeout = "3s"
 
-## Heartbeat timeout.
-heartbeat_timeout = "500ms"
-
 ## DDL timeout.
 ddl_timeout = "10s"
 

src/cmd/tests/load_config_test.rs

@@ -52,7 +52,6 @@ fn test_load_datanode_example_config() {
             meta_client: Some(MetaClientOptions {
                 metasrv_addrs: vec!["127.0.0.1:3002".to_string()],
                 timeout: Duration::from_secs(3),
-                heartbeat_timeout: Duration::from_millis(500),
                 ddl_timeout: Duration::from_secs(10),
                 connect_timeout: Duration::from_secs(1),
                 tcp_nodelay: true,
@@ -118,7 +117,6 @@ fn test_load_frontend_example_config() {
             meta_client: Some(MetaClientOptions {
                 metasrv_addrs: vec!["127.0.0.1:3002".to_string()],
                 timeout: Duration::from_secs(3),
-                heartbeat_timeout: Duration::from_millis(500),
                 ddl_timeout: Duration::from_secs(10),
                 connect_timeout: Duration::from_secs(1),
                 tcp_nodelay: true,
@@ -241,7 +239,6 @@ fn test_load_flownode_example_config() {
             meta_client: Some(MetaClientOptions {
                 metasrv_addrs: vec!["127.0.0.1:3002".to_string()],
                 timeout: Duration::from_secs(3),
-                heartbeat_timeout: Duration::from_millis(500),
                 ddl_timeout: Duration::from_secs(10),
                 connect_timeout: Duration::from_secs(1),
                 tcp_nodelay: true,

src/common/meta/src/distributed_time_constants.rs

@@ -41,6 +41,17 @@ pub const POSTGRES_KEEP_ALIVE_SECS: u64 = 30;
 /// In a lease, there are two opportunities for renewal.
 pub const META_KEEP_ALIVE_INTERVAL_SECS: u64 = META_LEASE_SECS / 2;
 
+/// The timeout of the heartbeat request.
+pub const HEARTBEAT_TIMEOUT: Duration = Duration::from_secs(META_KEEP_ALIVE_INTERVAL_SECS + 1);
+
+/// The keep-alive interval of the heartbeat channel.
+pub const HEARTBEAT_CHANNEL_KEEP_ALIVE_INTERVAL_SECS: Duration =
+    Duration::from_secs(META_KEEP_ALIVE_INTERVAL_SECS + 1);
+
+/// The keep-alive timeout of the heartbeat channel.
+pub const HEARTBEAT_CHANNEL_KEEP_ALIVE_TIMEOUT_SECS: Duration =
+    Duration::from_secs(META_KEEP_ALIVE_INTERVAL_SECS + 1);
+
 /// The default mailbox round-trip timeout.
 pub const MAILBOX_RTT_SECS: u64 = 1;
 

src/meta-client/src/client.rs

@@ -189,6 +189,9 @@ impl MetaClientBuilder {
         let mgr = client.channel_manager.clone();
 
         if self.enable_heartbeat {
+            if self.heartbeat_channel_manager.is_some() {
+                info!("Enable heartbeat channel using the heartbeat channel manager.");
+            }
             let mgr = self.heartbeat_channel_manager.unwrap_or(mgr.clone());
             client.heartbeat = Some(HeartbeatClient::new(
                 self.id,

src/meta-client/src/lib.rs

@@ -18,6 +18,10 @@ use std::time::Duration;
 use client::RegionFollowerClientRef;
 use common_base::Plugins;
 use common_grpc::channel_manager::{ChannelConfig, ChannelManager};
+use common_meta::distributed_time_constants::{
+    HEARTBEAT_CHANNEL_KEEP_ALIVE_INTERVAL_SECS, HEARTBEAT_CHANNEL_KEEP_ALIVE_TIMEOUT_SECS,
+    HEARTBEAT_TIMEOUT,
+};
 use common_telemetry::{debug, info};
 use serde::{Deserialize, Serialize};
 
@@ -34,8 +38,6 @@ pub struct MetaClientOptions {
     #[serde(with = "humantime_serde")]
     pub timeout: Duration,
     #[serde(with = "humantime_serde")]
-    pub heartbeat_timeout: Duration,
-    #[serde(with = "humantime_serde")]
     pub ddl_timeout: Duration,
     #[serde(with = "humantime_serde")]
     pub connect_timeout: Duration,
@@ -52,7 +54,6 @@ impl Default for MetaClientOptions {
         Self {
             metasrv_addrs: vec!["127.0.0.1:3002".to_string()],
             timeout: Duration::from_millis(3_000u64),
-            heartbeat_timeout: Duration::from_millis(500u64),
             ddl_timeout: Duration::from_millis(10_000u64),
             connect_timeout: Duration::from_millis(1_000u64),
             tcp_nodelay: true,
@@ -97,7 +98,11 @@ pub async fn create_meta_client(
         .timeout(meta_client_options.timeout)
         .connect_timeout(meta_client_options.connect_timeout)
         .tcp_nodelay(meta_client_options.tcp_nodelay);
-    let heartbeat_config = base_config.clone();
+    let heartbeat_config = base_config
+        .clone()
+        .timeout(HEARTBEAT_TIMEOUT)
+        .http2_keep_alive_interval(HEARTBEAT_CHANNEL_KEEP_ALIVE_INTERVAL_SECS)
+        .http2_keep_alive_timeout(HEARTBEAT_CHANNEL_KEEP_ALIVE_TIMEOUT_SECS);
 
     if let MetaClientType::Frontend = client_type {
         let ddl_config = base_config.clone().timeout(meta_client_options.ddl_timeout);