Merge pull request #1843 from dongluochen/ConnectionLeak-fix

Do not recycle connections. Let net/http handles connections.

api/handlers.go

@@ -477,7 +477,6 @@ func getContainerJSON(c *context, w http.ResponseWriter, r *http.Request) {
 
 	// cleanup
 	defer resp.Body.Close()
-	defer closeIdleConnections(client)
 
 	data, err := ioutil.ReadAll(resp.Body)
 	if err != nil {
@@ -786,7 +785,6 @@ func postContainersExec(c *context, w http.ResponseWriter, r *http.Request) {
 
 	// cleanup
 	defer resp.Body.Close()
-	defer closeIdleConnections(client)
 
 	// check status code
 	if resp.StatusCode < 200 || resp.StatusCode >= 400 {

api/utils.go

@@ -89,13 +89,6 @@ func copyHeader(dst, src http.Header) {
 	}
 }
 
-// prevents leak with https
-func closeIdleConnections(client *http.Client) {
-	if tr, ok := client.Transport.(*http.Transport); ok {
-		tr.CloseIdleConnections()
-	}
-}
-
 func proxyAsync(engine *cluster.Engine, w http.ResponseWriter, r *http.Request, callback func(*http.Response)) error {
 	// RequestURI may not be sent to client
 	r.RequestURI = ""
@@ -121,7 +114,6 @@ func proxyAsync(engine *cluster.Engine, w http.ResponseWriter, r *http.Request,
 
 	// cleanup
 	resp.Body.Close()
-	closeIdleConnections(client)
 
 	return nil
 }

test/integration/nodemanagement/connection.bats

@@ -0,0 +1,41 @@
+#!/usr/bin/env bats
+
+load ../helpers
+
+function teardown() {
+	swarm_manage_cleanup
+	stop_docker
+}
+
+@test "Swarm not leak tcp connections" {
+	# Start engine with busybox image
+	start_docker_with_busybox 2
+	# Start swarm and check it can reach the node
+	swarm_manage --engine-refresh-min-interval "20s" --engine-refresh-max-interval "20s" --engine-failure-retry 20 "${HOSTS[0]},${HOSTS[1]}"
+	eval "docker_swarm info | grep -q -i 'Nodes: 2'"
+
+	# create busybox with host network so that we can get netstat
+	run docker_swarm run -itd --name=busybox0 --net=host -e constraint:node==node-0 busybox sh
+	[ "$status" -eq 0 ]
+	run docker_swarm run -itd --name=busybox1 --net=host -e constraint:node==node-1 busybox sh
+	[ "$status" -eq 0 ]
+
+	# run most common container operations
+	for((i=0; i<30; i++)); do
+		# test postContainerCreate
+		docker_swarm run --name="hello$i" hello-world
+		# test getContainerJSON
+		docker_swarm inspect "hello$i"
+		# test proxyContainer
+		docker_swarm logs "hello$i"
+		# test proxyContainerAndForceRefresh
+		docker_swarm stop "hello$i"
+	done
+
+	# get connection count
+	count0=$(docker_swarm exec busybox0 netstat -an | grep "${HOSTS[0]}" | grep -i "ESTABLISHED" | wc -l)
+	count1=$(docker_swarm exec busybox1 netstat -an | grep "${HOSTS[1]}" | grep -i "ESTABLISHED" | wc -l)
+	[[ "$count0" -le 10 ]]
+	[[ "$count1" -le 10 ]]
+}
+

test/integration/nodemanagement/nodehealth.bats

@@ -8,6 +8,13 @@ function teardown() {
 }
 
 @test "scheduler avoids failing node" {
+	# Docker issue #14203 in runC causing this test to fail.
+	# Issue fixed in Docker 1.10
+	run docker --version
+	if [[ "${output}" == "Docker version 1.9"* ]]; then
+		skip
+	fi
+
 	# Start 1 engine and register it in the file.
 	start_docker 2
 	# Start swarm and check it can reach the node