cherry-pick #19433 to 1.34 (#19473)

## Description 

cherry pick of #19433

crates/sui-node/src/lib.rs

@@ -2038,7 +2038,7 @@ pub async fn build_http_server(
 
  if config.enable_experimental_rest_api {
  let mut rest_service = sui_rest_api::RestService::new(
- Arc::new(RestReadStore::new(state, store)),
+ Arc::new(RestReadStore::new(state.clone(), store)),
  software_version,
  );
 
@@ -2050,6 +2050,11 @@ pub async fn build_http_server(
 
  router = router.merge(rest_service.into_router());
  }
+ // TODO: Remove this health check when experimental REST API becomes default
+ // This is a copy of the health check in crates/sui-rest-api/src/health.rs
+ router = router
+ .route("/health", axum::routing::get(health_check_handler))
+ .route_layer(axum::Extension(state));
 
  let listener = tokio::net::TcpListener::bind(&config.json_rpc_address)
  .await
@@ -2072,6 +2077,51 @@ pub async fn build_http_server(
  Ok(Some(handle))
 }
 
+#[derive(Debug, serde::Serialize, serde::Deserialize)]
+pub struct Threshold {
+ pub threshold_seconds: Option<u32>,
+}
+
+async fn health_check_handler(
+ axum::extract::Query(Threshold { threshold_seconds }): axum::extract::Query<Threshold>,
+ axum::Extension(state): axum::Extension<Arc<AuthorityState>>,
+) -> impl axum::response::IntoResponse {
+ if let Some(threshold_seconds) = threshold_seconds {
+ // Attempt to get the latest checkpoint
+ let summary = match state
+ .get_checkpoint_store()
+ .get_highest_executed_checkpoint()
+ {
+ Ok(Some(summary)) => summary,
+ Ok(None) => {
+ warn!("Highest executed checkpoint not found");
+ return (axum::http::StatusCode::SERVICE_UNAVAILABLE, "down");
+ }
+ Err(err) => {
+ warn!("Failed to retrieve highest executed checkpoint: {:?}", err);
+ return (axum::http::StatusCode::SERVICE_UNAVAILABLE, "down");
+ }
+ };
+
+ // Calculate the threshold time based on the provided threshold_seconds
+ let latest_chain_time = summary.timestamp();
+ let threshold =
+ std::time::SystemTime::now() - Duration::from_secs(threshold_seconds as u64);
+
+ // Check if the latest checkpoint is within the threshold
+ if latest_chain_time < threshold {
+ warn!(
+ ?latest_chain_time,
+ ?threshold,
+ "failing healthcheck due to checkpoint lag"
+ );
+ return (axum::http::StatusCode::SERVICE_UNAVAILABLE, "down");
+ }
+ }
+ // if health endpoint is responding and no threshold is given, respond success
+ (axum::http::StatusCode::OK, "up")
+}
+
 #[cfg(not(test))]
 fn max_tx_per_checkpoint(protocol_config: &ProtocolConfig) -> usize {
  protocol_config.max_transactions_per_checkpoint() as usize