Merge pull request etcd-io#14691 from serathius/lazyfs

tests: Add support for lazyfs

.github/workflows/robustness-nightly.yaml

@@ -12,22 +12,22 @@ jobs:
     uses: ./.github/workflows/robustness-template.yaml
     with:
       etcdBranch: main
-      count: 100
+      count: 80
       testTimeout: 200m
       artifactName: main
   main-arm64:
     uses: ./.github/workflows/robustness-template-arm64.yaml
     with:
       etcdBranch: main
-      count: 100
+      count: 80
       testTimeout: 200m
       artifactName: main-arm64
       runs-on: "['self-hosted', 'Linux', 'ARM64']"
   release-35:
     uses: ./.github/workflows/robustness-template.yaml
     with:
       etcdBranch: release-3.5
-      count: 100
+      count: 80
       testTimeout: 200m
       artifactName: release-35
   release-35-arm64:
@@ -41,6 +41,6 @@ jobs:
     uses: ./.github/workflows/robustness-template.yaml
     with:
       etcdBranch: release-3.4
-      count: 100
+      count: 80
       testTimeout: 200m
       artifactName: release-34

.github/workflows/robustness-template.yaml

@@ -39,6 +39,10 @@ jobs:
           set -euo pipefail
           go clean -testcache
 
+          # Build LazyFS
+          sudo apt-get -y install cmake libfuse3-dev libfuse3-3 fuse3
+          make install-lazyfs
+
           # Use --failfast to avoid overriding report generated by failed test
           GO_TEST_FLAGS="-v --count ${{ inputs.count }} --timeout ${{ inputs.testTimeout }} --failfast --run TestRobustness"
           case "${ETCD_BRANCH}" in

.github/workflows/robustness.yaml

@@ -7,6 +7,6 @@ jobs:
     uses: ./.github/workflows/robustness-template.yaml
     with:
       etcdBranch: main
-      count: 15
+      count: 12
       testTimeout: 30m
       artifactName: main

Makefile

@@ -149,13 +149,27 @@ ifeq (, $(shell which yamlfmt))
 endif
 	yamlfmt -conf tools/.yamlfmt .
 
+# Tools
+
+.PHONY: install-lazyfs
+install-lazyfs: bin/lazyfs
+
+bin/lazyfs:
+	rm /tmp/lazyfs -rf
+	git clone --depth 1 --branch 0.2.0 https://github.com/dsrhaslab/lazyfs /tmp/lazyfs
+	cd /tmp/lazyfs/libs/libpcache; ./build.sh
+	cd /tmp/lazyfs/lazyfs; ./build.sh
+	mkdir -p ./bin
+	cp /tmp/lazyfs/lazyfs/build/lazyfs ./bin/lazyfs
+
 # Cleanup
 
 clean:
 	rm -f ./codecov
 	rm -rf ./covdir
 	rm -f ./bin/Dockerfile-release
 	rm -rf ./bin/etcd*
+	rm -rf ./bin/lazyfs
 	rm -rf ./default.etcd
 	rm -rf ./tests/e2e/default.etcd
 	rm -rf ./release

tests/framework/e2e/cluster.go

@@ -185,6 +185,7 @@ type EtcdProcessClusterConfig struct {
 	CompactHashCheckEnabled bool
 	CompactHashCheckTime    time.Duration
 	GoFailEnabled           bool
+	LazyFSEnabled           bool
 	CompactionBatchLimit    int
 	CompactionSleepInterval time.Duration
 
@@ -344,6 +345,10 @@ func WithGoFailEnabled(enabled bool) EPClusterOption {
 	return func(c *EtcdProcessClusterConfig) { c.GoFailEnabled = enabled }
 }
 
+func WithLazyFSEnabled(enabled bool) EPClusterOption {
+	return func(c *EtcdProcessClusterConfig) { c.LazyFSEnabled = enabled }
+}
+
 func WithWarningUnaryRequestDuration(time time.Duration) EPClusterOption {
 	return func(c *EtcdProcessClusterConfig) { c.WarningUnaryRequestDuration = time }
 }
@@ -407,7 +412,7 @@ func InitEtcdProcessCluster(t testing.TB, cfg *EtcdProcessClusterConfig) (*EtcdP
 
 	// launch etcd processes
 	for i := range etcdCfgs {
-		proc, err := NewEtcdProcess(etcdCfgs[i])
+		proc, err := NewEtcdProcess(t, etcdCfgs[i])
 		if err != nil {
 			epc.Close()
 			return nil, fmt.Errorf("cannot configure: %v", err)
@@ -659,6 +664,7 @@ func (cfg *EtcdProcessClusterConfig) EtcdServerProcessConfig(tb testing.TB, i in
 		InitialToken:  cfg.InitialToken,
 		GoFailPort:    gofailPort,
 		Proxy:         proxyCfg,
+		LazyFSEnabled: cfg.LazyFSEnabled,
 	}
 }
 
@@ -826,7 +832,7 @@ func (epc *EtcdProcessCluster) StartNewProc(ctx context.Context, cfg *EtcdProces
 
 	// Then start process
 	tb.Log("start new member")
-	proc, err := NewEtcdProcess(serverCfg)
+	proc, err := NewEtcdProcess(tb, serverCfg)
 	if err != nil {
 		epc.Close()
 		return 0, fmt.Errorf("cannot configure: %v", err)
@@ -855,7 +861,7 @@ func (epc *EtcdProcessCluster) UpdateProcOptions(i int, tb testing.TB, opts ...E
 	}
 	epc.Cfg.SetInitialOrDiscovery(serverCfg, initialCluster, "new")
 
-	proc, err := NewEtcdProcess(serverCfg)
+	proc, err := NewEtcdProcess(tb, serverCfg)
 	if err != nil {
 		return err
 	}

tests/framework/e2e/cluster_direct.go

@@ -16,6 +16,8 @@
 
 package e2e
 
-func NewEtcdProcess(cfg *EtcdServerProcessConfig) (EtcdProcess, error) {
-	return NewEtcdServerProcess(cfg)
+import "testing"
+
+func NewEtcdProcess(t testing.TB, cfg *EtcdServerProcessConfig) (EtcdProcess, error) {
+	return NewEtcdServerProcess(t, cfg)
 }

tests/framework/e2e/cluster_proxy.go

@@ -24,6 +24,7 @@ import (
 	"path"
 	"strconv"
 	"strings"
+	"testing"
 
 	"go.uber.org/zap"
 
@@ -38,12 +39,12 @@ type proxyEtcdProcess struct {
 	proxyV3 *proxyV3Proc
 }
 
-func NewEtcdProcess(cfg *EtcdServerProcessConfig) (EtcdProcess, error) {
-	return NewProxyEtcdProcess(cfg)
+func NewEtcdProcess(t testing.TB, cfg *EtcdServerProcessConfig) (EtcdProcess, error) {
+	return NewProxyEtcdProcess(t, cfg)
 }
 
-func NewProxyEtcdProcess(cfg *EtcdServerProcessConfig) (*proxyEtcdProcess, error) {
-	ep, err := NewEtcdServerProcess(cfg)
+func NewProxyEtcdProcess(t testing.TB, cfg *EtcdServerProcessConfig) (*proxyEtcdProcess, error) {
+	ep, err := NewEtcdServerProcess(t, cfg)
 	if err != nil {
 		return nil, err
 	}

tests/framework/e2e/etcd_process.go

@@ -56,6 +56,7 @@ type EtcdProcess interface {
 	Config() *EtcdServerProcessConfig
 	PeerProxy() proxy.Server
 	Failpoints() *BinaryFailpoints
+	LazyFS() *LazyFS
 	Logs() LogsExpect
 	Kill() error
 }
@@ -70,6 +71,7 @@ type EtcdServerProcess struct {
 	cfg        *EtcdServerProcessConfig
 	proc       *expect.ExpectProcess
 	proxy      proxy.Server
+	lazyfs     *LazyFS
 	failpoints *BinaryFailpoints
 	donec      chan struct{} // closed when Interact() terminates
 }
@@ -96,22 +98,29 @@ type EtcdServerProcessConfig struct {
 	InitialCluster string
 	GoFailPort     int
 
-	Proxy *proxy.ServerConfig
+	LazyFSEnabled bool
+	Proxy         *proxy.ServerConfig
 }
 
-func NewEtcdServerProcess(cfg *EtcdServerProcessConfig) (*EtcdServerProcess, error) {
+func NewEtcdServerProcess(t testing.TB, cfg *EtcdServerProcessConfig) (*EtcdServerProcess, error) {
 	if !fileutil.Exist(cfg.ExecPath) {
 		return nil, fmt.Errorf("could not find etcd binary: %s", cfg.ExecPath)
 	}
 	if !cfg.KeepDataDir {
 		if err := os.RemoveAll(cfg.DataDirPath); err != nil {
 			return nil, err
 		}
+		if err := os.Mkdir(cfg.DataDirPath, 0700); err != nil {
+			return nil, err
+		}
 	}
 	ep := &EtcdServerProcess{cfg: cfg, donec: make(chan struct{})}
 	if cfg.GoFailPort != 0 {
 		ep.failpoints = &BinaryFailpoints{member: ep}
 	}
+	if cfg.LazyFSEnabled {
+		ep.lazyfs = newLazyFS(cfg.lg, cfg.DataDirPath, t)
+	}
 	return ep, nil
 }
 
@@ -146,6 +155,14 @@ func (ep *EtcdServerProcess) Start(ctx context.Context) error {
 			return err
 		}
 	}
+	if ep.lazyfs != nil {
+		ep.cfg.lg.Info("starting lazyfs...", zap.String("name", ep.cfg.Name))
+		err := ep.lazyfs.Start(ctx)
+		if err != nil {
+			return err
+		}
+	}
+
 	ep.cfg.lg.Info("starting server...", zap.String("name", ep.cfg.Name))
 	proc, err := SpawnCmdWithLogger(ep.cfg.lg, append([]string{ep.cfg.ExecPath}, ep.cfg.Args...), ep.cfg.EnvVars, ep.cfg.Name)
 	if err != nil {
@@ -205,6 +222,14 @@ func (ep *EtcdServerProcess) Stop() (err error) {
 			return err
 		}
 	}
+	if ep.lazyfs != nil {
+		ep.cfg.lg.Info("stopping lazyfs...", zap.String("name", ep.cfg.Name))
+		err = ep.lazyfs.Stop()
+		ep.lazyfs = nil
+		if err != nil {
+			return err
+		}
+	}
 	return nil
 }
 
@@ -298,6 +323,10 @@ func (ep *EtcdServerProcess) PeerProxy() proxy.Server {
 	return ep.proxy
 }
 
+func (ep *EtcdServerProcess) LazyFS() *LazyFS {
+	return ep.lazyfs
+}
+
 func (ep *EtcdServerProcess) Failpoints() *BinaryFailpoints {
 	return ep.failpoints
 }

tests/framework/e2e/flags.go

@@ -48,6 +48,18 @@ type binPath struct {
 	EtcdLastRelease string
 	Etcdctl         string
 	Etcdutl         string
+	LazyFS          string
+}
+
+func (bp *binPath) LazyFSAvailable() bool {
+	_, err := os.Stat(bp.LazyFS)
+	if err != nil {
+		if !os.IsNotExist(err) {
+			panic(err)
+		}
+		return false
+	}
+	return true
 }
 
 func InitFlags() {
@@ -65,6 +77,7 @@ func InitFlags() {
 		EtcdLastRelease: *binDir + "/etcd-last-release",
 		Etcdctl:         *binDir + "/etcdctl",
 		Etcdutl:         *binDir + "/etcdutl",
+		LazyFS:          *binDir + "/lazyfs",
 	}
 	CertPath = CertDir + "/server.crt"
 	PrivateKeyPath = CertDir + "/server.key.insecure"