e2e: add TestCtlV3Corrupt

e2e/cluster_test.go

@@ -20,6 +20,7 @@ import (
 	"net/url"
 	"os"
 	"strings"
+	"time"
 
 	"github.com/coreos/etcd/etcdserver"
 )
@@ -115,6 +116,7 @@ type etcdProcessClusterConfig struct {
 	forceNewCluster   bool
 	initialToken      string
 	quotaBackendBytes int64
+	corruptCheckTime  time.Duration
 	noStrictReconfig  bool
 }
 
@@ -221,6 +223,11 @@ func (cfg *etcdProcessClusterConfig) etcdServerProcessConfigs() []*etcdServerPro
 				"--quota-backend-bytes", fmt.Sprintf("%d", cfg.quotaBackendBytes),
 			)
 		}
+		if cfg.corruptCheckTime > 0 {
+			args = append(args,
+				"--experimental-corrupt-check-time", cfg.corruptCheckTime.String(),
+			)
+		}
 		if cfg.noStrictReconfig {
 			args = append(args, "--strict-reconfig-check=false")
 		}

e2e/ctl_v3_alarm_test.go

@@ -16,12 +16,16 @@ package e2e
 
 import (
 	"context"
+	"fmt"
 	"os"
+	"path/filepath"
 	"strings"
 	"testing"
 	"time"
 
 	"github.com/coreos/etcd/clientv3"
+	"github.com/coreos/etcd/mvcc"
+	"github.com/coreos/etcd/mvcc/backend"
 )
 
 func TestCtlV3Alarm(t *testing.T) {
@@ -99,6 +103,60 @@ func alarmTest(cx ctlCtx) {
 	}
 }
 
+func TestCtlV3Corrupt(t *testing.T) {
+	testCtl(t, corruptTest, withCfg(configNoTLS), withQuorum(), withCorruptCheckTime(2*time.Second))
+}
+
+type fakeConsistentIndex struct{ rev uint64 }
+
+func (f *fakeConsistentIndex) ConsistentIndex() uint64 { return f.rev }
+
+func corruptTest(cx ctlCtx) {
+	for i := 0; i < 10; i++ {
+		if err := ctlV3Put(cx, "k", "v", ""); err != nil {
+			if cx.dialTimeout > 0 && !isGRPCTimedout(err) {
+				cx.t.Fatalf("putTest ctlV3Put error (%v)", err)
+			}
+		}
+	}
+
+	eps := cx.epc.EndpointsV3()
+	cli1, err := clientv3.New(clientv3.Config{Endpoints: []string{eps[1]}, DialTimeout: 3 * time.Second})
+	if err != nil {
+		cx.t.Fatal(err)
+	}
+	defer cli1.Close()
+
+	sresp, err := cli1.Status(context.TODO(), eps[0])
+	if err != nil {
+		cx.t.Fatal(err)
+	}
+	id0 := sresp.Header.GetMemberId()
+
+	cx.epc.procs[0].Stop()
+
+	// Corrupt member 0 by modifying backend offline.
+	fp := filepath.Join(cx.epc.procs[0].Config().dataDirPath, "member", "snap", "db")
+	be := backend.NewDefaultBackend(fp)
+	s := mvcc.NewStore(be, nil, &fakeConsistentIndex{13})
+	s.Put([]byte("abc"), []byte("def"), 0)
+	s.Put([]byte("xyz"), []byte("123"), 0)
+	s.Compact(5)
+	s.Commit()
+	s.Close()
+	be.Close()
+
+	ep := cx.epc.procs[0]
+	proc, err := spawnCmd(append([]string{ep.Config().execPath}, ep.Config().args...))
+	if err != nil {
+		cx.t.Fatal(err)
+	}
+	defer proc.Stop()
+
+	// restarting corrupted member should fail
+	waitReadyExpectProc(proc, []string{fmt.Sprintf("etcdserver: %016x is corrupted", id0)})
+}
+
 func ctlV3Alarm(cx ctlCtx, cmd string, as ...string) error {
 	cmdArgs := append(cx.PrefixArgs(), "alarm", cmd)
 	return spawnWithExpects(cmdArgs, as...)

e2e/ctl_v3_test.go

@@ -55,6 +55,7 @@ type ctlCtx struct {
 	t                 *testing.T
 	cfg               etcdProcessClusterConfig
 	quotaBackendBytes int64
+	corruptCheckTime  time.Duration
 	noStrictReconfig  bool
 
 	epc *etcdProcessCluster
@@ -101,6 +102,10 @@ func withQuota(b int64) ctlOption {
 	return func(cx *ctlCtx) { cx.quotaBackendBytes = b }
 }
 
+func withCorruptCheckTime(d time.Duration) ctlOption {
+	return func(cx *ctlCtx) { cx.corruptCheckTime = d }
+}
+
 func withCompactPhysical() ctlOption {
 	return func(cx *ctlCtx) { cx.compactPhysical = true }
 }
@@ -130,6 +135,9 @@ func testCtl(t *testing.T, testFunc func(ctlCtx), opts ...ctlOption) {
 	if ret.quotaBackendBytes > 0 {
 		ret.cfg.quotaBackendBytes = ret.quotaBackendBytes
 	}
+	if ret.corruptCheckTime > 0 {
+		ret.cfg.corruptCheckTime = ret.corruptCheckTime
+	}
 	ret.cfg.noStrictReconfig = ret.noStrictReconfig
 
 	epc, err := newEtcdProcessCluster(&ret.cfg)