feat(http): add metrics on panics & query

src/query/service/src/servers/http/http_services.rs

@@ -37,6 +37,7 @@ use poem::Route;
 use super::v1::upload_to_stage;
 use crate::auth::AuthMgr;
 use crate::servers::http::middleware::HTTPSessionMiddleware;
+use crate::servers::http::middleware::PanicHandler;
 use crate::servers::http::v1::clickhouse_router;
 use crate::servers::http::v1::list_suggestions;
 use crate::servers::http::v1::query_route;
@@ -122,7 +123,7 @@ impl HttpHandler {
                 .nest("/health", ep_health),
         };
         ep.with(NormalizePath::new(TrailingSlash::Trim))
-            .with(CatchPanic::new())
+            .with(CatchPanic::new().with_handler(PanicHandler::new()))
             .boxed()
     }
 

src/query/service/src/servers/http/metrics.rs

@@ -23,3 +23,12 @@ pub fn metrics_incr_http_slow_request_count(method: String, api: String, status:
     let labels = [("method", method), ("api", api), ("status", status)];
     counter!("query_http_slow_requests_count", 1, &labels);
 }
+
+pub fn metrics_incr_http_response_errors_count(err: String, code: u16) {
+    let labels = [("err", err), ("code", code.to_string())];
+    counter!("query_http_response_errors_count", 1, &labels);
+}
+
+pub fn metrics_incr_http_response_panics_count() {
+    counter!("query_http_response_panics_count", 1);
+}

src/query/service/src/servers/http/middleware.rs

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::any::Any;
 use std::collections::HashMap;
 use std::sync::Arc;
 use std::time::Instant;
@@ -41,6 +42,7 @@ use super::v1::HttpQueryContext;
 use crate::auth::AuthMgr;
 use crate::auth::Credential;
 use crate::servers::http::metrics::metrics_incr_http_request_count;
+use crate::servers::http::metrics::metrics_incr_http_response_panics_count;
 use crate::servers::http::metrics::metrics_incr_http_slow_request_count;
 use crate::servers::HttpHandlerKind;
 use crate::sessions::SessionManager;
@@ -314,3 +316,21 @@ impl<E: Endpoint> Endpoint for MetricsMiddlewareEndpoint<E> {
         Ok(resp)
     }
 }
+
+#[derive(Clone, Debug)]
+pub(crate) struct PanicHandler {}
+
+impl PanicHandler {
+    pub fn new() -> Self {
+        Self {}
+    }
+}
+
+impl poem::middleware::PanicHandler for PanicHandler {
+    type Response = (StatusCode, &'static str);
+
+    fn get_response(&self, _err: Box<dyn Any + Send + 'static>) -> Self::Response {
+        metrics_incr_http_response_panics_count();
+        (StatusCode::INTERNAL_SERVER_ERROR, "internal server error")
+    }
+}

src/query/service/src/servers/http/v1/http_query_handlers.rs

@@ -33,6 +33,7 @@ use serde_json::Value as JsonValue;
 use super::query::ExecuteStateKind;
 use super::query::HttpQueryRequest;
 use super::query::HttpQueryResponseInternal;
+use crate::servers::http::metrics::metrics_incr_http_response_errors_count;
 use crate::servers::http::middleware::MetricsMiddleware;
 use crate::servers::http::v1::query::Progresses;
 use crate::servers::http::v1::HttpQueryContext;
@@ -156,13 +157,18 @@ impl QueryResponse {
             }
         };
 
+        if let Some(err) = &r.state.error {
+            metrics_incr_http_response_errors_count(err.name(), err.code());
+        }
+
         let schema = data.schema().clone();
         let session_id = r.session_id.clone();
         let stats = QueryStats {
             progresses: state.progresses.clone(),
             running_time_ms: state.running_time_ms,
         };
         let rows = data.data.len();
+
         Json(QueryResponse {
             data: data.into(),
             state: state.state,
@@ -184,6 +190,7 @@ impl QueryResponse {
     }
 
     pub(crate) fn fail_to_start_sql(err: &ErrorCode) -> impl IntoResponse {
+        metrics_incr_http_response_errors_count(err.name(), err.code());
         Json(QueryResponse {
             id: "".to_string(),
             stats: QueryStats::default(),