Skip to content

SeaweedFS is a distributed storage system for object storage (S3), file systems, Iceberg table buckets, and Lance vector buckets, designed to handle billions of files with O(1) disk access and effortless horizontal scaling.

9.2 KB
Raw Blame
package weed_server

import (
	"fmt"
	"net/http"
	"sync"
	"sync/atomic"
	"time"

	"github.com/seaweedfs/seaweedfs/weed/pb"
	"github.com/seaweedfs/seaweedfs/weed/pb/volume_server_pb"
	"github.com/seaweedfs/seaweedfs/weed/storage/types"

	"google.golang.org/grpc"

	"github.com/seaweedfs/seaweedfs/weed/stats"
	"github.com/seaweedfs/seaweedfs/weed/util"

	"github.com/seaweedfs/seaweedfs/weed/glog"
	"github.com/seaweedfs/seaweedfs/weed/security"
	"github.com/seaweedfs/seaweedfs/weed/storage"
)

type VolumeServer struct {
	volume_server_pb.UnimplementedVolumeServerServer
	inFlightUploadDataSize        int64
	inFlightDownloadDataSize      int64
	concurrentUploadLimit         int64
	concurrentDownloadLimit       int64
	inFlightUploadDataLimitCond   *sync.Cond
	inFlightDownloadDataLimitCond *sync.Cond
	inflightUploadDataTimeout     time.Duration
	inflightDownloadDataTimeout   time.Duration
	hasSlowRead                   bool
	readBufferSizeMB              int

	SeedMasterNodes []pb.ServerAddress
	// seedMasterSet mirrors SeedMasterNodes keyed by the canonical http
	// form. It is computed once in NewVolumeServer so admission paths can
	// answer is-this-a-seed-master in O(1).
	seedMasterSet     map[string]struct{}
	whiteList         []string
	currentMaster     pb.ServerAddress
	currentMasterLock sync.RWMutex
	// confirmedLeaderOnce records that a heartbeat stream was ever answered
	// by a leader; before first registration the retry loop reconnects
	// promptly instead of pacing a full pulse.
	confirmedLeaderOnce atomic.Bool
	pulsePeriod         time.Duration
	dataCenter          string
	rack                string
	store               *storage.Store
	guard               *security.Guard
	grpcDialOption      grpc.DialOption

	// leaderLostAt is when a stream a leader had answered last ended, in Unix
	// nanoseconds. For a pulse after that the loop also reconnects promptly:
	// a master that just dropped the stream is usually restarting.
	leaderLostAt atomic.Int64

	needleMapKind                 storage.NeedleMapKind
	ldbTimout                     int64
	FixJpgOrientation             bool
	ReadMode                      string
	AllowUntrustedRemoteEndpoints bool
	compactionBytePerSecond       int64
	maintenanceBytePerSecond      int64
	vacuumCommitSlots             chan struct{}
	vacuumCommitsWaiting          atomic.Int32
	metricsAddress                string
	metricsIntervalSec            int
	fileSizeLimitBytes            int64
	isHeartbeating                bool
	stopChan                      chan bool

	// Volumes with an EC decode in flight. A cancelled request does not
	// stop the handler; this keeps a retry from racing it on the same
	// volume files.
	ecDecodesInFlight sync.Map
	// A copy claims its volume's guard, so mounts and configures either
	// register before the copy's registration recheck (and abort it) or see
	// the claim and refuse. Entries are never removed: a guard handed out
	// while one side waits must stay the same mutex.
	volumeCopyGuards sync.Map
}

func NewVolumeServer(adminMux, publicMux *http.ServeMux, ip string,
	port int, grpcPort int, publicUrl string, id string,
	folders []string, maxCounts []int32, minFreeSpaces []util.MinFreeSpace, diskTypes []types.DiskType, diskTags [][]string,
	idxFolder string,
	needleMapKind storage.NeedleMapKind,
	masterNodes []pb.ServerAddress, pulsePeriod time.Duration,
	dataCenter string, rack string,
	whiteList []string,
	fixJpgOrientation bool,
	readMode string,
	compactionMBPerSecond int,
	maintenanceMBPerSecond int,
	maxParallelVacuumCommits int,
	fileSizeLimitMB int,
	concurrentUploadLimit int64,
	concurrentDownloadLimit int64,
	inflightUploadDataTimeout time.Duration,
	inflightDownloadDataTimeout time.Duration,
	hasSlowRead bool,
	readBufferSizeMB int,
	ldbTimeout int64,
	allowUntrustedRemoteEndpoints bool,
	diskProbeConfig stats.DiskIOProbeConfig,
	syncBatch storage.SyncBatchOption,
) *VolumeServer {

	v := util.GetViper()
	signingKey := v.GetString("jwt.signing.key")
	v.SetDefault("jwt.signing.expires_after_seconds", 10)
	expiresAfterSec := v.GetInt("jwt.signing.expires_after_seconds")
	enableUiAccess := v.GetBool("access.ui")

	readSigningKey := v.GetString("jwt.signing.read.key")
	v.SetDefault("jwt.signing.read.expires_after_seconds", 60)
	readExpiresAfterSec := v.GetInt("jwt.signing.read.expires_after_seconds")

	vs := &VolumeServer{
		pulsePeriod:                   pulsePeriod,
		dataCenter:                    dataCenter,
		rack:                          rack,
		needleMapKind:                 needleMapKind,
		FixJpgOrientation:             fixJpgOrientation,
		ReadMode:                      readMode,
		grpcDialOption:                security.LoadClientTLS(util.GetViper(), "grpc.volume"),
		compactionBytePerSecond:       int64(compactionMBPerSecond) * 1024 * 1024,
		maintenanceBytePerSecond:      int64(maintenanceMBPerSecond) * 1024 * 1024,
		fileSizeLimitBytes:            int64(fileSizeLimitMB) * 1024 * 1024,
		isHeartbeating:                true,
		stopChan:                      make(chan bool),
		inFlightUploadDataLimitCond:   sync.NewCond(new(sync.Mutex)),
		inFlightDownloadDataLimitCond: sync.NewCond(new(sync.Mutex)),
		concurrentUploadLimit:         concurrentUploadLimit,
		concurrentDownloadLimit:       concurrentDownloadLimit,
		inflightUploadDataTimeout:     inflightUploadDataTimeout,
		inflightDownloadDataTimeout:   inflightDownloadDataTimeout,
		hasSlowRead:                   hasSlowRead,
		readBufferSizeMB:              readBufferSizeMB,
		ldbTimout:                     ldbTimeout,
		whiteList:                     whiteList,
		AllowUntrustedRemoteEndpoints: allowUntrustedRemoteEndpoints,
	}
	if maxParallelVacuumCommits > 0 {
		vs.vacuumCommitSlots = make(chan struct{}, maxParallelVacuumCommits)
	}

	whiteList = append(whiteList, util.StringSplit(v.GetString("guard.white_list"), ",")...)
	// Copy the caller's slice so subsequent external mutation cannot desync
	// SeedMasterNodes from the frozen lookup set built below.
	seedMasters := make([]pb.ServerAddress, len(masterNodes))
	copy(seedMasters, masterNodes)
	vs.SeedMasterNodes = seedMasters
	vs.seedMasterSet = make(map[string]struct{}, len(seedMasters))
	for _, m := range seedMasters {
		vs.seedMasterSet[m.ToHttpAddress()] = struct{}{}
	}

	vs.checkWithMaster()

	vs.store = storage.NewStore(vs.grpcDialOption, ip, port, grpcPort, publicUrl, id, folders, maxCounts, minFreeSpaces, idxFolder, vs.needleMapKind, diskTypes, diskTags, ldbTimeout, diskProbeConfig)
	vs.store.SetVolumeLoadGuard(vs.volumeLoadGuard)
	vs.store.MasterFn = vs.getCurrentMaster
	vs.store.SetSyncBatchOption(syncBatch)
	vs.guard = security.NewGuard(whiteList, signingKey, expiresAfterSec, readSigningKey, readExpiresAfterSec)

	handleStaticResources(adminMux)
	adminMux.HandleFunc("/status", requestIDMiddleware(vs.statusHandler))
	adminMux.HandleFunc("/healthz", requestIDMiddleware(vs.healthzHandler))
	adminMux.HandleFunc("/readyz", requestIDMiddleware(vs.healthzHandler))
	if signingKey == "" || enableUiAccess {
		// only expose the volume server details for safe environments
		adminMux.HandleFunc("/ui/index.html", requestIDMiddleware(vs.uiStatusHandler))
		/*
			adminMux.HandleFunc("/stats/counter", vs.guard.WhiteList(statsCounterHandler))
			adminMux.HandleFunc("/stats/memory", vs.guard.WhiteList(statsMemoryHandler))
			adminMux.HandleFunc("/stats/disk", vs.guard.WhiteList(vs.statsDiskHandler))
		*/
	}
	adminMux.HandleFunc("/", requestIDMiddleware(vs.privateStoreHandler))
	if publicMux != adminMux {
		// separated admin and public port
		handleStaticResources(publicMux)
		publicMux.HandleFunc("/", requestIDMiddleware(vs.publicReadOnlyHandler))
	}

	stats.VolumeServerConcurrentDownloadLimit.Set(float64(vs.concurrentDownloadLimit))
	stats.VolumeServerConcurrentUploadLimit.Set(float64(vs.concurrentUploadLimit))
	stats.VolumeServerStartTimeSeconds.Set(float64(time.Now().Unix()))

	go vs.heartbeat()
	go stats.LoopPushingMetric("volumeServer", util.JoinHostPort(ip, port), vs.metricsAddress, vs.metricsIntervalSec)

	return vs
}

func (vs *VolumeServer) SetStopping() {
	glog.V(0).Infoln("Stopping volume server...")
	vs.store.SetStopping()
}

func (vs *VolumeServer) LoadNewVolumes() {
	glog.V(0).Infoln(" Loading new volume ids ...")
	vs.store.LoadNewVolumes()
}

func (vs *VolumeServer) Shutdown() {
	glog.V(0).Infoln("Shutting down volume server...")
	vs.store.Close()
	glog.V(0).Infoln("Shut down successfully!")
}

func (vs *VolumeServer) Reload() {
	glog.V(0).Infoln("Reload volume server...")

	util.LoadConfiguration("security", false)
	v := util.GetViper()
	vs.guard.UpdateWhiteList(append(vs.whiteList, util.StringSplit(v.GetString("guard.white_list"), ",")...))
	vs.guard.UpdateSigningKeys(
		v.GetString("jwt.signing.key"),
		v.GetInt("jwt.signing.expires_after_seconds"),
		v.GetString("jwt.signing.read.key"),
		v.GetInt("jwt.signing.read.expires_after_seconds"),
	)
}

// Returns whether a volume server is in maintenance (i.e. read-only) mode.
func (vs *VolumeServer) MaintenanceMode() bool {
	if vs.store == nil {
		return false
	}
	return vs.store.State.Proto().GetMaintenance()
}

// Checks if a volume server is in maintenance mode, and returns an error explaining why.
func (vs *VolumeServer) CheckMaintenanceMode() error {
	if !vs.MaintenanceMode() {
		return nil
	}
	return fmt.Errorf("volume server %s is in maintenance mode", vs.store.Id)
}