From bef00dffefd425d3e4eab81001d4ec56a6d882d2 Mon Sep 17 00:00:00 2001
From: ginuerzh <ginuerzh@gmail.com>
Date: Thu, 16 Jul 2026 16:44:48 +0800
Subject: [PATCH] e2e: add node liveness probe tests

TCP probe + FailFilter: dead node pre-marked, all requests succeed.
LowestLatency strategy: two probed nodes, selector picks fastest.
---
 tests/e2e/probe_test.go                     | 90 +++++++++++++++++++++
 tests/e2e/testdata/probe/lowestlatency.yaml | 49 +++++++++++
 tests/e2e/testdata/probe/tcp.yaml           | 44 ++++++++++
 3 files changed, 183 insertions(+)
 create mode 100644 tests/e2e/probe_test.go
 create mode 100644 tests/e2e/testdata/probe/lowestlatency.yaml
 create mode 100644 tests/e2e/testdata/probe/tcp.yaml

diff --git a/tests/e2e/probe_test.go b/tests/e2e/probe_test.go
new file mode 100644
index 0000000..7db11bf
--- /dev/null
+++ b/tests/e2e/probe_test.go
@@ -0,0 +1,90 @@
+package e2e
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/suite"
+	"github.com/testcontainers/testcontainers-go"
+)
+
+type ProbeSuite struct {
+	suite.Suite
+	ctx   context.Context
+	echoC testcontainers.Container
+	echoIP string
+}
+
+func (s *ProbeSuite) SetupSuite() {
+	s.ctx = context.Background()
+
+	echoC, err := RunEchoContainer(s.ctx, SharedNetworkName)
+	s.Require().NoError(err)
+	s.echoC = echoC
+
+	echoIP, err := echoC.ContainerIP(s.ctx)
+	s.Require().NoError(err)
+	s.echoIP = echoIP
+}
+
+func (s *ProbeSuite) TearDownSuite() {
+	if s.echoC != nil {
+		s.echoC.Terminate(s.ctx)
+	}
+}
+
+func (s *ProbeSuite) proxyRequest(gostC testcontainers.Container, port string) (int, string) {
+	cmd := []string{
+		"curl", "-s",
+		"-x", fmt.Sprintf("http://127.0.0.1:%s", port),
+		fmt.Sprintf("http://%s:5678", s.echoIP),
+	}
+	code, out, err := gostC.Exec(s.ctx, cmd)
+	s.Require().NoError(err)
+
+	body, err := io.ReadAll(out)
+	s.Require().NoError(err)
+	return code, string(body)
+}
+
+// TestTCPProbeFailover verifies that the TCP probe detects a dead node and
+// marks it before any real traffic, so the FailFilter excludes it. With the
+// dead node pre-marked, every request succeeds immediately.
+func (s *ProbeSuite) TestTCPProbeFailover() {
+	gostC, err := RunGostContainerWithPorts(s.ctx, SharedNetworkName, "testdata/probe/tcp.yaml", "8080/tcp")
+	s.Require().NoError(err)
+	defer gostC.Terminate(s.ctx)
+
+	// The probe fires immediately at startup, so the dead node is already
+	// marked by the time we send requests.
+	for range 10 {
+		code, body := s.proxyRequest(gostC, "8080")
+		s.Require().Equal(0, code, "every request must succeed; dead node pre-marked by probe")
+		s.Require().Contains(body, "hello-gost")
+	}
+}
+
+// TestLowestLatencyProbe verifies that the lowestlatency strategy works with
+// probed nodes. Both nodes are live; the strategy selects the one with the
+// lowest measured latency.
+func (s *ProbeSuite) TestLowestLatencyProbe() {
+	gostC, err := RunGostContainerWithPorts(s.ctx, SharedNetworkName, "testdata/probe/lowestlatency.yaml", "8080/tcp")
+	s.Require().NoError(err)
+	defer gostC.Terminate(s.ctx)
+
+	// Give the initial probe a moment to fire so both nodes show healthy.
+	time.Sleep(200 * time.Millisecond)
+
+	for range 10 {
+		code, body := s.proxyRequest(gostC, "8080")
+		s.Require().Equal(0, code, "all requests must succeed with lowestlatency strategy")
+		s.Require().Contains(body, "hello-gost")
+	}
+}
+
+func TestProbeSuite(t *testing.T) {
+	suite.Run(t, new(ProbeSuite))
+}
diff --git a/tests/e2e/testdata/probe/lowestlatency.yaml b/tests/e2e/testdata/probe/lowestlatency.yaml
new file mode 100644
index 0000000..c3c71c8
--- /dev/null
+++ b/tests/e2e/testdata/probe/lowestlatency.yaml
@@ -0,0 +1,49 @@
+services:
+- name: proxy
+  addr: :8080
+  handler:
+    type: http
+    chain: my-chain
+  listener:
+    type: tcp
+
+# Two live relays — both forward to the echo server.
+- name: relay-a
+  addr: 127.0.0.1:18081
+  handler:
+    type: http
+  listener:
+    type: tcp
+
+- name: relay-b
+  addr: 127.0.0.1:18082
+  handler:
+    type: http
+  listener:
+    type: tcp
+
+chains:
+- name: my-chain
+  hops:
+  - name: hop-1
+    selector:
+      strategy: lowestlatency
+      maxFails: 1
+      failTimeout: 10s
+    nodes:
+    - name: node-a
+      addr: 127.0.0.1:18081
+      connector:
+        type: http
+      probe:
+        type: tcp
+        addr: 127.0.0.1:18081
+        interval: 5s
+    - name: node-b
+      addr: 127.0.0.1:18082
+      connector:
+        type: http
+      probe:
+        type: tcp
+        addr: 127.0.0.1:18082
+        interval: 5s
diff --git a/tests/e2e/testdata/probe/tcp.yaml b/tests/e2e/testdata/probe/tcp.yaml
new file mode 100644
index 0000000..19be3b6
--- /dev/null
+++ b/tests/e2e/testdata/probe/tcp.yaml
@@ -0,0 +1,44 @@
+services:
+- name: proxy
+  addr: :8080
+  handler:
+    type: http
+    chain: my-chain
+  listener:
+    type: tcp
+
+# Live relay: forwards to the echo server.
+- name: relay
+  addr: 127.0.0.1:18081
+  handler:
+    type: http
+  listener:
+    type: tcp
+
+chains:
+- name: my-chain
+  hops:
+  - name: hop-1
+    selector:
+      strategy: round
+      maxFails: 1
+      failTimeout: 10s
+    nodes:
+    - name: node-live
+      addr: 127.0.0.1:18081
+      connector:
+        type: http
+      probe:
+        type: tcp
+        addr: 127.0.0.1:18081
+        interval: 5s
+    # dead node: probed, marked as failed by the probe goroutine,
+    # then excluded by FailFilter before any real traffic tries it.
+    - name: node-dead
+      addr: 127.0.0.1:18082
+      connector:
+        type: http
+      probe:
+        type: tcp
+        addr: 127.0.0.1:18082
+        interval: 5s