Reuse connections in engine health probes

services/nvpair-engine-manager/lifecycle.go

@@ -18,6 +18,8 @@ import (
 	"sync"
 	"syscall"
 	"time"
+
+	"nvpair-shared/httpcon"
 )
 
 const (
@@ -816,7 +818,7 @@ func (e *Executor) probe(ctx context.Context, p *Probe, port int) bool {
 		if err != nil {
 			return false
 		}
-		resp.Body.Close()
+		httpcon.DrainAndClose(resp.Body)
 		want := p.Status
 		if want == 0 {
 			want = 200

services/nvpair-ui-broker/advertiser.go

@@ -10,6 +10,7 @@ import (
 	"net/http"
 	"time"
 
+	"nvpair-shared/httpcon"
 	"nvpair-shared/noderec"
 )
 
@@ -271,7 +272,7 @@ func checkOllamaHealth(client *http.Client, port int) bool {
 	if err != nil {
 		return false
 	}
-	resp.Body.Close()
+	httpcon.DrainAndClose(resp.Body)
 	return resp.StatusCode == http.StatusOK
 }
 
@@ -285,7 +286,7 @@ func checkLMStudioHealth(client *http.Client, port int) bool {
 	if err != nil {
 		return false
 	}
-	resp.Body.Close()
+	httpcon.DrainAndClose(resp.Body)
 	return resp.StatusCode == http.StatusOK
 }
 
@@ -334,7 +335,7 @@ func (b *Broker) reconcileAdvertiseOMLX(client *http.Client) {
 func checkOMLXHealth(client *http.Client, port int) bool {
 	resp, err := client.Get(fmt.Sprintf("http://localhost:%d/health", port))
 	if err == nil {
-		resp.Body.Close()
+		httpcon.DrainAndClose(resp.Body)
 		if resp.StatusCode == http.StatusOK {
 			return true
 		}
@@ -343,6 +344,6 @@ func checkOMLXHealth(client *http.Client, port int) bool {
 	if err != nil {
 		return false
 	}
-	resp.Body.Close()
+	httpcon.DrainAndClose(resp.Body)
 	return resp.StatusCode == http.StatusOK || resp.StatusCode == http.StatusUnauthorized
 }

services/shared/httpcon/httpcon.go

@@ -0,0 +1,19 @@
+// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+// Package httpcon provides helpers for HTTP connections.
+package httpcon
+
+import "io"
+
+const maxDrainBytes = 1 << 20
+
+// DrainAndClose discards a bounded response body before closing it so normal
+// HTTP/1 responses reach EOF and their connections can return to the transport
+// pool. Without this, connections cannot be reused and this can lead to socket
+// exhaustion. The caller is responsible for bounding stalled reads with a
+// request context or client timeout.
+func DrainAndClose(body io.ReadCloser) {
+	_, _ = io.Copy(io.Discard, io.LimitReader(body, maxDrainBytes))
+	_ = body.Close()
+}

services/shared/httpcon/httpcon_test.go

@@ -0,0 +1,47 @@
+// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+package httpcon
+
+import (
+	"io"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+)
+
+// Closing a body unread keeps its connection out of the pool, so a poller
+// opens one connection per poll; draining it first lets one connection serve
+// every poll.
+func TestDrainAndCloseReusesTheConnection(t *testing.T) {
+	for _, chunked := range []bool{false, true} {
+		var conns atomic.Int32
+		srv := httptest.NewUnstartedServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+			if chunked {
+				w.(http.Flusher).Flush()
+			}
+			_, _ = io.WriteString(w, strings.Repeat("x", 4096))
+		}))
+		srv.Config.ConnState = func(_ net.Conn, state http.ConnState) {
+			if state == http.StateNew {
+				conns.Add(1)
+			}
+		}
+		srv.Start()
+		client := srv.Client()
+		for i := 0; i < 16; i++ {
+			resp, err := client.Get(srv.URL)
+			if err != nil {
+				t.Fatal(err)
+			}
+			DrainAndClose(resp.Body)
+		}
+		srv.Close()
+		if got := conns.Load(); got != 1 {
+			t.Errorf("chunked=%v: %d connections for 16 polls, want 1", chunked, got)
+		}
+	}
+}