Merge pull request #10444 from WIZARD-CXY/nnboltdb

*: add flag to let etcd use the new boltdb freelistType feature

Documentation/op-guide/configuration.md

@@ -408,6 +408,11 @@ Follow the instructions when using these flags.
 
 ## Experimental flags
 
+### --experimental-backend-bbolt-freelist-type
++ The freelist type that etcd backend(bboltdb) uses (array and map are supported types).
++ default: array
++ env variable: ETCD_EXPERIMENTAL_BACKEND_BBOLT_FREELIST_TYPE
+
 ### --experimental-corrupt-check-time
 + Duration of time between cluster corruption check passes
 + default: 0s

embed/config.go

@@ -37,6 +37,7 @@ import (
 	"go.etcd.io/etcd/pkg/types"
 
 	"github.com/ghodss/yaml"
+	bolt "go.etcd.io/bbolt"
 	"go.uber.org/zap"
 	"go.uber.org/zap/zapcore"
 	"golang.org/x/crypto/bcrypt"
@@ -75,6 +76,8 @@ const (
 	// maxElectionMs specifies the maximum value of election timeout.
 	// More details are listed in ../Documentation/tuning.md#time-parameters.
 	maxElectionMs = 50000
+	// backend freelist map type
+	freelistMapType = "map"
 )
 
 var (
@@ -273,6 +276,8 @@ type Config struct {
 	ExperimentalInitialCorruptCheck bool          `json:"experimental-initial-corrupt-check"`
 	ExperimentalCorruptCheckTime    time.Duration `json:"experimental-corrupt-check-time"`
 	ExperimentalEnableV2V3          string        `json:"experimental-enable-v2v3"`
+	// ExperimentalBackendFreelistType specifies the type of freelist that boltdb backend uses (array and map are supported types).
+	ExperimentalBackendFreelistType string `json:"experimental-backend-bbolt-freelist-type"`
 
 	// ForceNewCluster starts a new cluster even if previously started; unsafe.
 	ForceNewCluster bool `json:"force-new-cluster"`
@@ -896,3 +901,11 @@ func (cfg *Config) getMetricsURLs() (ss []string) {
 	}
 	return ss
 }
+
+func parseBackendFreelistType(freelistType string) bolt.FreelistType {
+	if freelistType == freelistMapType {
+		return bolt.FreelistMapType
+	}
+
+	return bolt.FreelistArrayType
+}

embed/etcd.go

@@ -159,6 +159,8 @@ func StartEtcd(inCfg *Config) (e *Etcd, err error) {
 		return e, err
 	}
 
+	backendFreelistType := parseBackendFreelistType(cfg.ExperimentalBackendFreelistType)
+
 	srvcfg := etcdserver.ServerConfig{
 		Name:                       cfg.Name,
 		ClientURLs:                 cfg.ACUrls,
@@ -182,6 +184,7 @@ func StartEtcd(inCfg *Config) (e *Etcd, err error) {
 		AutoCompactionMode:         cfg.AutoCompactionMode,
 		QuotaBackendBytes:          cfg.QuotaBackendBytes,
 		BackendBatchLimit:          cfg.BackendBatchLimit,
+		BackendFreelistType:        backendFreelistType,
 		BackendBatchInterval:       cfg.BackendBatchInterval,
 		MaxTxnOps:                  cfg.MaxTxnOps,
 		MaxRequestBytes:            cfg.MaxRequestBytes,

etcdmain/config.go

@@ -248,6 +248,7 @@ func newConfig() *config {
 	fs.BoolVar(&cfg.ec.ExperimentalInitialCorruptCheck, "experimental-initial-corrupt-check", cfg.ec.ExperimentalInitialCorruptCheck, "Enable to check data corruption before serving any client/peer traffic.")
 	fs.DurationVar(&cfg.ec.ExperimentalCorruptCheckTime, "experimental-corrupt-check-time", cfg.ec.ExperimentalCorruptCheckTime, "Duration of time between cluster corruption check passes.")
 	fs.StringVar(&cfg.ec.ExperimentalEnableV2V3, "experimental-enable-v2v3", cfg.ec.ExperimentalEnableV2V3, "v3 prefix for serving emulated v2 state.")
+	fs.StringVar(&cfg.ec.ExperimentalBackendFreelistType, "experimental-backend-bbolt-freelist-type", cfg.ec.ExperimentalBackendFreelistType, "ExperimentalBackendFreelistType specifies the type of freelist that boltdb backend uses(array and map are supported types)")
 
 	// unsafe
 	fs.BoolVar(&cfg.ec.ForceNewCluster, "force-new-cluster", false, "Force to create a new one member cluster.")

etcdmain/help.go

@@ -200,6 +200,8 @@ Experimental feature:
     Duration of time between cluster corruption check passes.
   --experimental-enable-v2v3 ''
     Serve v2 requests through the v3 backend under a given prefix.
+  --experimental-backend-bbolt-freelist-type
+    ExperimentalBackendFreelistType specifies the type of freelist that boltdb backend uses(array and map are supported types).
 
 Unsafe feature:
   --force-new-cluster 'false'

etcdserver/backend.go

@@ -43,6 +43,7 @@ func newBackend(cfg ServerConfig) backend.Backend {
 			cfg.Logger.Info("setting backend batch interval", zap.Duration("batch interval", cfg.BackendBatchInterval))
 		}
 	}
+	bcfg.BackendFreelistType = cfg.BackendFreelistType
 	bcfg.Logger = cfg.Logger
 	if cfg.QuotaBackendBytes > 0 && cfg.QuotaBackendBytes != DefaultQuotaBytes {
 		// permit 10% excess over quota for disarm

etcdserver/config.go

@@ -26,6 +26,7 @@ import (
 	"go.etcd.io/etcd/pkg/transport"
 	"go.etcd.io/etcd/pkg/types"
 
+	bolt "go.etcd.io/bbolt"
 	"go.uber.org/zap"
 	"go.uber.org/zap/zapcore"
 )
@@ -60,6 +61,9 @@ type ServerConfig struct {
 	// BackendBatchLimit is the maximum operations before commit the backend transaction.
 	BackendBatchLimit int
 
+	// BackendFreelistType is the type of the backend boltdb freelist.
+	BackendFreelistType bolt.FreelistType
+
 	InitialPeerURLsMap  types.URLsMap
 	InitialClusterToken string
 	NewCluster          bool

mvcc/backend/backend.go

@@ -110,6 +110,8 @@ type BackendConfig struct {
 	BatchInterval time.Duration
 	// BatchLimit is the maximum puts before flushing the BatchTx.
 	BatchLimit int
+	// BackendFreelistType is the backend boltdb's freelist type.
+	BackendFreelistType bolt.FreelistType
 	// MmapSize is the number of bytes to mmap for the backend.
 	MmapSize uint64
 	// Logger logs backend-side operations.
@@ -140,6 +142,7 @@ func newBackend(bcfg BackendConfig) *backend {
 		*bopts = *boltOpenOptions
 	}
 	bopts.InitialMmapSize = bcfg.mmapSize()
+	bopts.FreelistType = bcfg.BackendFreelistType
 
 	db, err := bolt.Open(bcfg.Path, 0600, bopts)
 	if err != nil {