frostfs-node/cmd/neofs-node/config.go
Alex Vanin 5180e467f8 [#1278] neofs-node: Use global cached netmap source in services
Signed-off-by: Alex Vanin <alexey@nspcc.ru>

(cherry picked from commit be6ae3c066)
Signed-off-by: Alex Vanin <alexey@nspcc.ru>
2022-03-30 14:34:34 +03:00

495 lines
13 KiB
Go

package main
import (
"context"
"net"
"path"
"sync"
"time"
"github.com/nspcc-dev/neo-go/pkg/crypto/keys"
neogoutil "github.com/nspcc-dev/neo-go/pkg/util"
netmapV2 "github.com/nspcc-dev/neofs-api-go/v2/netmap"
"github.com/nspcc-dev/neofs-node/cmd/neofs-node/config"
apiclientconfig "github.com/nspcc-dev/neofs-node/cmd/neofs-node/config/apiclient"
contractsconfig "github.com/nspcc-dev/neofs-node/cmd/neofs-node/config/contracts"
engineconfig "github.com/nspcc-dev/neofs-node/cmd/neofs-node/config/engine"
shardconfig "github.com/nspcc-dev/neofs-node/cmd/neofs-node/config/engine/shard"
loggerconfig "github.com/nspcc-dev/neofs-node/cmd/neofs-node/config/logger"
metricsconfig "github.com/nspcc-dev/neofs-node/cmd/neofs-node/config/metrics"
nodeconfig "github.com/nspcc-dev/neofs-node/cmd/neofs-node/config/node"
objectconfig "github.com/nspcc-dev/neofs-node/cmd/neofs-node/config/object"
"github.com/nspcc-dev/neofs-node/pkg/core/container"
netmapCore "github.com/nspcc-dev/neofs-node/pkg/core/netmap"
"github.com/nspcc-dev/neofs-node/pkg/local_object_storage/blobstor"
"github.com/nspcc-dev/neofs-node/pkg/local_object_storage/engine"
meta "github.com/nspcc-dev/neofs-node/pkg/local_object_storage/metabase"
"github.com/nspcc-dev/neofs-node/pkg/local_object_storage/shard"
"github.com/nspcc-dev/neofs-node/pkg/local_object_storage/writecache"
"github.com/nspcc-dev/neofs-node/pkg/metrics"
"github.com/nspcc-dev/neofs-node/pkg/morph/client"
nmwrapper "github.com/nspcc-dev/neofs-node/pkg/morph/client/netmap/wrapper"
"github.com/nspcc-dev/neofs-node/pkg/morph/event"
netmap2 "github.com/nspcc-dev/neofs-node/pkg/morph/event/netmap"
"github.com/nspcc-dev/neofs-node/pkg/morph/timer"
"github.com/nspcc-dev/neofs-node/pkg/network"
"github.com/nspcc-dev/neofs-node/pkg/network/cache"
"github.com/nspcc-dev/neofs-node/pkg/services/control"
"github.com/nspcc-dev/neofs-node/pkg/services/object/acl/eacl"
trustcontroller "github.com/nspcc-dev/neofs-node/pkg/services/reputation/local/controller"
truststorage "github.com/nspcc-dev/neofs-node/pkg/services/reputation/local/storage"
tokenStorage "github.com/nspcc-dev/neofs-node/pkg/services/session/storage"
"github.com/nspcc-dev/neofs-node/pkg/services/util/response"
"github.com/nspcc-dev/neofs-node/pkg/util"
"github.com/nspcc-dev/neofs-node/pkg/util/logger"
"github.com/nspcc-dev/neofs-node/pkg/util/state"
apiclient "github.com/nspcc-dev/neofs-sdk-go/client"
"github.com/nspcc-dev/neofs-sdk-go/netmap"
"github.com/nspcc-dev/neofs-sdk-go/owner"
"github.com/nspcc-dev/neofs-sdk-go/version"
"github.com/panjf2000/ants/v2"
"go.etcd.io/bbolt"
"go.uber.org/atomic"
"go.uber.org/zap"
"google.golang.org/grpc"
)
const addressSize = 72 // 32 bytes oid, 32 bytes cid, 8 bytes protobuf encoding
const maxMsgSize = 4 << 20 // transport msg limit 4 MiB
// capacity of the pools of the morph notification handlers
// for each contract listener.
const notificationHandlerPoolSize = 10
type cfg struct {
ctx context.Context
appCfg *config.Config
ctxCancel func()
internalErr chan error // channel for internal application errors at runtime
log *zap.Logger
wg *sync.WaitGroup
key *keys.PrivateKey
ownerIDFromKey *owner.ID // owner ID calculated from key
apiVersion *version.Version
cfgGRPC cfgGRPC
cfgMorph cfgMorph
cfgAccounting cfgAccounting
cfgContainer cfgContainer
cfgNetmap cfgNetmap
privateTokenStore *tokenStorage.TokenStore
cfgNodeInfo cfgNodeInfo
localAddr network.AddressGroup
cfgObject cfgObject
metricsCollector *metrics.StorageMetrics
workers []worker
respSvc *response.Service
cfgControlService cfgControlService
healthStatus *atomic.Int32
closers []func()
cfgReputation cfgReputation
clientCache *cache.ClientCache
persistate *state.PersistentStorage
netMapSource netmapCore.Source
}
type cfgGRPC struct {
listeners []net.Listener
servers []*grpc.Server
maxChunkSize uint64
maxAddrAmount uint64
}
type cfgMorph struct {
client *client.Client
notaryEnabled bool
disableCache bool
blockTimers []*timer.BlockTimer // all combined timers
eigenTrustTimer *timer.BlockTimer // timer for EigenTrust iterations
proxyScriptHash neogoutil.Uint160
}
type cfgAccounting struct {
scriptHash neogoutil.Uint160
}
type cfgContainer struct {
scriptHash neogoutil.Uint160
parsers map[event.Type]event.NotificationParser
subscribers map[event.Type][]event.Handler
workerPool util.WorkerPool // pool for asynchronous handlers
}
type cfgNetmap struct {
scriptHash neogoutil.Uint160
wrapper *nmwrapper.Wrapper
parsers map[event.Type]event.NotificationParser
subscribers map[event.Type][]event.Handler
workerPool util.WorkerPool // pool for asynchronous handlers
state *networkState
needBootstrap bool
reBoostrapTurnedOff *atomic.Bool // managed by control service in runtime
startEpoch uint64 // epoch number when application is started
}
type cfgNodeInfo struct {
// values from config
localInfo netmap.NodeInfo
}
type cfgObject struct {
cnrSource container.Source
eaclSource eacl.Source
pool cfgObjectRoutines
cfgLocalStorage cfgLocalStorage
}
type cfgLocalStorage struct {
localStorage *engine.StorageEngine
shardOpts [][]shard.Option
}
type cfgObjectRoutines struct {
putRemote *ants.Pool
putRemoteCapacity int
replication *ants.Pool
}
type cfgControlService struct {
server *grpc.Server
}
type cfgReputation struct {
workerPool util.WorkerPool // pool for EigenTrust algorithm's iterations
localTrustStorage *truststorage.Storage
localTrustCtrl *trustcontroller.Controller
scriptHash neogoutil.Uint160
}
var persistateSideChainLastBlockKey = []byte("side_chain_last_processed_block")
func initCfg(path string) *cfg {
var p config.Prm
appCfg := config.New(p,
config.WithConfigFile(path),
)
key := nodeconfig.Key(appCfg)
ownerIDFromKey := owner.NewIDFromPublicKey(&key.PrivateKey.PublicKey)
var logPrm logger.Prm
err := logPrm.SetLevelString(
loggerconfig.Level(appCfg),
)
fatalOnErr(err)
log, err := logger.NewLogger(logPrm)
fatalOnErr(err)
netAddr := nodeconfig.BootstrapAddresses(appCfg)
maxChunkSize := uint64(maxMsgSize) * 3 / 4 // 25% to meta, 75% to payload
maxAddrAmount := uint64(maxChunkSize) / addressSize // each address is about 72 bytes
netState := newNetworkState()
persistate, err := state.NewPersistentStorage(nodeconfig.PersistentState(appCfg).Path())
fatalOnErr(err)
containerWorkerPool, err := ants.NewPool(notificationHandlerPoolSize)
fatalOnErr(err)
netmapWorkerPool, err := ants.NewPool(notificationHandlerPoolSize)
fatalOnErr(err)
reputationWorkerPool, err := ants.NewPool(notificationHandlerPoolSize)
fatalOnErr(err)
relayOnly := nodeconfig.Relay(appCfg)
c := &cfg{
ctx: context.Background(),
appCfg: appCfg,
internalErr: make(chan error),
log: log,
wg: new(sync.WaitGroup),
key: key,
apiVersion: version.Current(),
cfgAccounting: cfgAccounting{
scriptHash: contractsconfig.Balance(appCfg),
},
cfgContainer: cfgContainer{
scriptHash: contractsconfig.Container(appCfg),
workerPool: containerWorkerPool,
},
cfgNetmap: cfgNetmap{
scriptHash: contractsconfig.Netmap(appCfg),
state: netState,
workerPool: netmapWorkerPool,
needBootstrap: !relayOnly,
reBoostrapTurnedOff: atomic.NewBool(relayOnly),
},
cfgGRPC: cfgGRPC{
maxChunkSize: maxChunkSize,
maxAddrAmount: maxAddrAmount,
},
cfgMorph: cfgMorph{
proxyScriptHash: contractsconfig.Proxy(appCfg),
},
localAddr: netAddr,
respSvc: response.NewService(
response.WithNetworkState(netState),
),
cfgObject: cfgObject{
pool: initObjectPool(appCfg),
},
healthStatus: atomic.NewInt32(int32(control.HealthStatus_HEALTH_STATUS_UNDEFINED)),
cfgReputation: cfgReputation{
scriptHash: contractsconfig.Reputation(appCfg),
workerPool: reputationWorkerPool,
},
clientCache: cache.NewSDKClientCache(
apiclient.WithDialTimeout(apiclientconfig.DialTimeout(appCfg)),
),
persistate: persistate,
ownerIDFromKey: ownerIDFromKey,
}
if metricsconfig.Address(c.appCfg) != "" {
c.metricsCollector = metrics.NewStorageMetrics()
netState.metrics = c.metricsCollector
}
c.onShutdown(c.clientCache.CloseAll) // clean up connections
c.onShutdown(func() { _ = c.persistate.Close() })
initLocalStorage(c)
return c
}
func (c *cfg) LocalAddress() network.AddressGroup {
return c.localAddr
}
func initLocalStorage(c *cfg) {
initShardOptions(c)
engineOpts := []engine.Option{
engine.WithLogger(c.log),
engine.WithShardPoolSize(engineconfig.ShardPoolSize(c.appCfg)),
}
if c.metricsCollector != nil {
engineOpts = append(engineOpts, engine.WithMetrics(c.metricsCollector))
}
ls := engine.New(engineOpts...)
for _, opts := range c.cfgObject.cfgLocalStorage.shardOpts {
id, err := ls.AddShard(opts...)
fatalOnErr(err)
c.log.Info("shard attached to engine",
zap.Stringer("id", id),
)
}
c.cfgObject.cfgLocalStorage.localStorage = ls
c.onShutdown(func() {
c.log.Info("closing components of the storage engine...")
err := ls.Close()
if err != nil {
c.log.Info("storage engine closing failure",
zap.String("error", err.Error()),
)
} else {
c.log.Info("all components of the storage engine closed successfully")
}
})
}
func initShardOptions(c *cfg) {
var opts [][]shard.Option
require := !nodeconfig.Relay(c.appCfg) // relay node does not require shards
engineconfig.IterateShards(c.appCfg, require, func(sc *shardconfig.Config) {
var writeCacheOpts []writecache.Option
writeCacheCfg := sc.WriteCache()
if writeCacheCfg.Enabled() {
writeCacheOpts = []writecache.Option{
writecache.WithPath(writeCacheCfg.Path()),
writecache.WithLogger(c.log),
writecache.WithMaxMemSize(writeCacheCfg.MemSize()),
writecache.WithMaxObjectSize(writeCacheCfg.MaxObjectSize()),
writecache.WithSmallObjectSize(writeCacheCfg.SmallObjectSize()),
writecache.WithFlushWorkersCount(writeCacheCfg.WorkersNumber()),
writecache.WithMaxCacheSize(writeCacheCfg.SizeLimit()),
}
}
blobStorCfg := sc.BlobStor()
blobovniczaCfg := blobStorCfg.Blobovnicza()
metabaseCfg := sc.Metabase()
gcCfg := sc.GC()
metaPath := metabaseCfg.Path()
metaPerm := metabaseCfg.Perm()
fatalOnErr(util.MkdirAllX(path.Dir(metaPath), metaPerm))
opts = append(opts, []shard.Option{
shard.WithLogger(c.log),
shard.WithRefillMetabase(sc.RefillMetabase()),
shard.WithMode(sc.Mode()),
shard.WithBlobStorOptions(
blobstor.WithRootPath(blobStorCfg.Path()),
blobstor.WithCompressObjects(blobStorCfg.Compress()),
blobstor.WithRootPerm(blobStorCfg.Perm()),
blobstor.WithShallowDepth(blobStorCfg.ShallowDepth()),
blobstor.WithSmallSizeLimit(blobStorCfg.SmallSizeLimit()),
blobstor.WithBlobovniczaSize(blobovniczaCfg.Size()),
blobstor.WithBlobovniczaShallowDepth(blobovniczaCfg.ShallowDepth()),
blobstor.WithBlobovniczaShallowWidth(blobovniczaCfg.ShallowWidth()),
blobstor.WithBlobovniczaOpenedCacheSize(blobovniczaCfg.OpenedCacheSize()),
blobstor.WithLogger(c.log),
),
shard.WithMetaBaseOptions(
meta.WithLogger(c.log),
meta.WithPath(metaPath),
meta.WithPermissions(metaPerm),
meta.WithBoltDBOptions(&bbolt.Options{
Timeout: 100 * time.Millisecond,
}),
),
shard.WithWriteCache(writeCacheCfg.Enabled()),
shard.WithWriteCacheOptions(writeCacheOpts...),
shard.WithRemoverBatchSize(gcCfg.RemoverBatchSize()),
shard.WithGCRemoverSleepInterval(gcCfg.RemoverSleepInterval()),
shard.WithGCWorkerPoolInitializer(func(sz int) util.WorkerPool {
pool, err := ants.NewPool(sz)
fatalOnErr(err)
return pool
}),
shard.WithGCEventChannelInitializer(func() <-chan shard.Event {
ch := make(chan shard.Event)
addNewEpochNotificationHandler(c, func(ev event.Event) {
ch <- shard.EventNewEpoch(ev.(netmap2.NewEpoch).EpochNumber())
})
return ch
}),
})
})
c.cfgObject.cfgLocalStorage.shardOpts = opts
}
func initObjectPool(cfg *config.Config) (pool cfgObjectRoutines) {
var err error
optNonBlocking := ants.WithNonblocking(true)
pool.putRemoteCapacity = objectconfig.Put(cfg).PoolSizeRemote()
pool.putRemote, err = ants.NewPool(pool.putRemoteCapacity, optNonBlocking)
fatalOnErr(err)
pool.replication, err = ants.NewPool(pool.putRemoteCapacity)
fatalOnErr(err)
return pool
}
func (c *cfg) LocalNodeInfo() (*netmapV2.NodeInfo, error) {
ni := c.cfgNetmap.state.getNodeInfo()
if ni != nil {
return ni.ToV2(), nil
}
return c.cfgNodeInfo.localInfo.ToV2(), nil
}
// handleLocalNodeInfo rewrites local node info from netmap
func (c *cfg) handleLocalNodeInfo(ni *netmap.NodeInfo) {
c.cfgNetmap.state.setNodeInfo(ni)
}
// bootstrap sets local node's netmap status to "online".
func (c *cfg) bootstrap() error {
ni := c.cfgNodeInfo.localInfo
ni.SetState(netmap.NodeStateOnline)
prm := nmwrapper.AddPeerPrm{}
prm.SetNodeInfo(&ni)
return c.cfgNetmap.wrapper.AddPeer(prm)
}
// needBootstrap checks if local node should be registered in network on bootup.
func (c *cfg) needBootstrap() bool {
return c.cfgNetmap.needBootstrap
}
// ObjectServiceLoad implements system loader interface for policer component.
// It is calculated as size/capacity ratio of "remote object put" worker.
// Returns float value between 0.0 and 1.0.
func (c *cfg) ObjectServiceLoad() float64 {
return float64(c.cfgObject.pool.putRemote.Running()) / float64(c.cfgObject.pool.putRemoteCapacity)
}