From 24a210ab2064638ca4be1d76896ecc1c110d3a4f Mon Sep 17 00:00:00 2001
From: Yicheng Qin <qycqycqycqycqyc@gmail.com>
Date: Thu, 5 Mar 2015 21:52:35 -0800
Subject: [PATCH] tools/etcd-tester: add kill one member tests

---
 .../functional-tester/etcd-tester/cluster.go  | 32 +++++++-
 .../functional-tester/etcd-tester/failure.go  | 76 ++++++++++++++++++-
 tools/functional-tester/etcd-tester/main.go   | 25 ++----
 .../functional-tester/etcd-tester/stresser.go |  3 +-
 4 files changed, 115 insertions(+), 21 deletions(-)

diff --git a/tools/functional-tester/etcd-tester/cluster.go b/tools/functional-tester/etcd-tester/cluster.go
index 25b851666..876d57e7c 100644
--- a/tools/functional-tester/etcd-tester/cluster.go
+++ b/tools/functional-tester/etcd-tester/cluster.go
@@ -32,6 +32,7 @@ type cluster struct {
 
 	Size       int
 	Agents     []client.Agent
+	Stressers  []Stresser
 	Names      []string
 	ClientURLs []string
 }
@@ -98,8 +99,19 @@ func (c *cluster) Bootstrap() error {
 		}
 	}
 
+	stressers := make([]Stresser, len(clientURLs))
+	for i, u := range clientURLs {
+		s := &stresser{
+			Endpoint: u,
+			N:        200,
+		}
+		go s.Stress()
+		stressers[i] = s
+	}
+
 	c.Size = size
 	c.Agents = agents
+	c.Stressers = stressers
 	c.Names = names
 	c.ClientURLs = clientURLs
 	return nil
@@ -117,19 +129,35 @@ func (c *cluster) WaitHealth() error {
 	return err
 }
 
+func (c *cluster) Report() (success, failure int) {
+	for _, stress := range c.Stressers {
+		s, f := stress.Report()
+		success += s
+		failure += f
+	}
+	return
+}
+
 func (c *cluster) Cleanup() error {
+	var lasterr error
 	for _, a := range c.Agents {
 		if err := a.Cleanup(); err != nil {
-			return err
+			lasterr = err
 		}
 	}
-	return nil
+	for _, s := range c.Stressers {
+		s.Cancel()
+	}
+	return lasterr
 }
 
 func (c *cluster) Terminate() {
 	for _, a := range c.Agents {
 		a.Terminate()
 	}
+	for _, s := range c.Stressers {
+		s.Cancel()
+	}
 }
 
 // setHealthKey sets health key on all given urls.
diff --git a/tools/functional-tester/etcd-tester/failure.go b/tools/functional-tester/etcd-tester/failure.go
index fc8aebe39..79efe4c1b 100644
--- a/tools/functional-tester/etcd-tester/failure.go
+++ b/tools/functional-tester/etcd-tester/failure.go
@@ -14,7 +14,13 @@
 
 package main
 
-import "math/rand"
+import (
+	"fmt"
+	"math/rand"
+	"time"
+)
+
+const snapshotCount = 10000
 
 type failure interface {
 	// Inject injeccts the failure into the testing cluster at the given
@@ -98,3 +104,71 @@ func getToKillMap(size int, seed int) map[int]bool {
 		}
 	}
 }
+
+type failureKillOne struct {
+	description
+}
+
+func newFailureKillOne() *failureKillOne {
+	return &failureKillOne{
+		description: "kill one random member",
+	}
+}
+
+func (f *failureKillOne) Inject(c *cluster, round int) error {
+	i := round % c.Size
+	return c.Agents[i].Stop()
+}
+
+func (f *failureKillOne) Recover(c *cluster, round int) error {
+	i := round % c.Size
+	if _, err := c.Agents[i].Restart(); err != nil {
+		return err
+	}
+	return c.WaitHealth()
+}
+
+// failureKillOneForLongTime kills one member for long time, and restart
+// after a snapshot is required.
+type failureKillOneForLongTime struct {
+	description
+}
+
+func newFailureKillOneForLongTime() *failureKillOneForLongTime {
+	return &failureKillOneForLongTime{
+		description: "kill one member for long time and expect it to recover from incoming snapshot",
+	}
+}
+
+func (f *failureKillOneForLongTime) Inject(c *cluster, round int) error {
+	i := round % c.Size
+	if err := c.Agents[i].Stop(); err != nil {
+		return err
+	}
+	if c.Size >= 3 {
+		start, _ := c.Report()
+		var end int
+		// Normal healthy cluster could accept 1000req/s at least.
+		// Give it 3-times time to create a new snapshot.
+		retry := snapshotCount / 1000 * 3
+		for j := 0; j < retry; j++ {
+			end, _ = c.Report()
+			// If the number of proposals committed is bigger than snapshot count,
+			// a new snapshot should have been created.
+			if end-start > snapshotCount {
+				return nil
+			}
+			time.Sleep(time.Second)
+		}
+		return fmt.Errorf("cluster too slow: only commit %d requests in %ds", end-start, retry)
+	}
+	return nil
+}
+
+func (f *failureKillOneForLongTime) Recover(c *cluster, round int) error {
+	i := round % c.Size
+	if _, err := c.Agents[i].Restart(); err != nil {
+		return err
+	}
+	return c.WaitHealth()
+}
diff --git a/tools/functional-tester/etcd-tester/main.go b/tools/functional-tester/etcd-tester/main.go
index d0d4c2b0d..6c8fe5dac 100644
--- a/tools/functional-tester/etcd-tester/main.go
+++ b/tools/functional-tester/etcd-tester/main.go
@@ -33,24 +33,15 @@ func main() {
 	}
 	defer c.Terminate()
 
-	stressers := make([]Stresser, len(c.ClientURLs))
-	for i, u := range c.ClientURLs {
-		s := &stresser{
-			Endpoint: u,
-			N:        200,
-		}
-		go s.Stress()
-		stressers[i] = s
-	}
-
 	t := &tester{
-		failures: []failure{newFailureKillAll(), newFailureKillMajority()},
-		cluster:  c,
-		limit:    *limit,
+		failures: []failure{
+			newFailureKillAll(),
+			newFailureKillMajority(),
+			newFailureKillOne(),
+			newFailureKillOneForLongTime(),
+		},
+		cluster: c,
+		limit:   *limit,
 	}
 	t.runLoop()
-
-	for _, s := range stressers {
-		s.Cancel()
-	}
 }
diff --git a/tools/functional-tester/etcd-tester/stresser.go b/tools/functional-tester/etcd-tester/stresser.go
index 680d47337..280c1e18f 100644
--- a/tools/functional-tester/etcd-tester/stresser.go
+++ b/tools/functional-tester/etcd-tester/stresser.go
@@ -65,8 +65,9 @@ func (s *stresser) Stress() error {
 				s.mu.Lock()
 				if err != nil {
 					s.failure++
+				} else {
+					s.success++
 				}
-				s.success++
 				s.mu.Unlock()
 			}
 		}()