2016-06-03 09:09:14 +08:00
package command
2013-12-04 15:22:26 +08:00
import (
"net/http"
"os"
"runtime"
2014-05-08 01:17:06 +08:00
"runtime/pprof"
2013-12-04 15:22:26 +08:00
"strconv"
"strings"
2014-03-16 14:03:49 +08:00
"sync"
2013-12-04 15:22:26 +08:00
"time"
2014-10-27 02:34:55 +08:00
2016-06-03 09:09:14 +08:00
"github.com/chrislusf/seaweedfs/weed/glog"
2018-05-10 14:11:54 +08:00
"github.com/chrislusf/seaweedfs/weed/pb/master_pb"
2016-06-03 09:09:14 +08:00
"github.com/chrislusf/seaweedfs/weed/server"
"github.com/chrislusf/seaweedfs/weed/storage"
"github.com/chrislusf/seaweedfs/weed/util"
2014-10-27 02:34:55 +08:00
"github.com/gorilla/mux"
2017-01-10 17:01:12 +08:00
"github.com/soheilhy/cmux"
"google.golang.org/grpc/reflection"
2013-12-04 15:22:26 +08:00
)
2014-05-08 01:17:06 +08:00
type ServerOptions struct {
cpuprofile * string
}
2014-03-31 11:57:25 +08:00
var (
2014-05-08 01:17:06 +08:00
serverOptions ServerOptions
2014-05-13 15:03:10 +08:00
filerOptions FilerOptions
2014-03-31 11:57:25 +08:00
)
2013-12-04 15:22:26 +08:00
func init ( ) {
cmdServer . Run = runServer // break init cycle
}
var cmdServer = & Command {
2014-05-27 08:34:54 +08:00
UsageLine : "server -port=8080 -dir=/tmp -volume.max=5 -ip=server_name" ,
2013-12-04 15:22:26 +08:00
Short : "start a server, including volume server, and automatically elect a master server" ,
2014-11-29 08:34:03 +08:00
Long : ` start both a volume server to provide storage spaces
2013-12-04 15:22:26 +08:00
and a master server to provide volume = > location mapping service and sequence number of file ids
2014-11-29 08:34:03 +08:00
2013-12-04 15:22:26 +08:00
This is provided as a convenient way to start both volume server and master server .
2014-04-26 13:09:42 +08:00
The servers are exactly the same as starting them separately .
2013-12-04 15:22:26 +08:00
So other volume servers can use this embedded master server also .
2014-11-29 08:34:03 +08:00
2014-04-26 13:09:42 +08:00
Optionally , one filer server can be started . Logically , filer servers should not be in a cluster .
They run with meta data on disk , not shared . So each filer server is different .
2014-11-29 08:34:03 +08:00
2013-12-04 15:22:26 +08:00
` ,
}
var (
2015-02-12 13:04:43 +08:00
serverIp = cmdServer . Flag . String ( "ip" , "localhost" , "ip or server name" )
2014-09-21 14:34:13 +08:00
serverBindIp = cmdServer . Flag . String ( "ip.bind" , "0.0.0.0" , "ip address to bind to" )
2014-03-03 14:16:54 +08:00
serverMaxCpu = cmdServer . Flag . Int ( "maxCpu" , 0 , "maximum number of CPUs. 0 means all available CPUs" )
2017-01-10 17:01:12 +08:00
serverTimeout = cmdServer . Flag . Int ( "idleTimeout" , 30 , "connection idle seconds" )
2014-03-03 14:16:54 +08:00
serverDataCenter = cmdServer . Flag . String ( "dataCenter" , "" , "current volume server's data center name" )
serverRack = cmdServer . Flag . String ( "rack" , "" , "current volume server's rack name" )
2016-08-06 05:45:48 +08:00
serverWhiteListOption = cmdServer . Flag . String ( "whiteList" , "" , "comma separated Ip addresses having write permission. No limit if empty." )
2018-08-13 05:25:31 +08:00
serverPeers = cmdServer . Flag . String ( "master.peers" , "" , "all master nodes in comma separated ip:masterPort list" )
2015-02-08 07:35:28 +08:00
serverSecureKey = cmdServer . Flag . String ( "secure.secret" , "" , "secret to encrypt Json Web Token(JWT)" )
2014-05-27 08:34:54 +08:00
serverGarbageThreshold = cmdServer . Flag . String ( "garbageThreshold" , "0.3" , "threshold to vacuum and reclaim spaces" )
2014-04-26 13:09:42 +08:00
masterPort = cmdServer . Flag . Int ( "master.port" , 9333 , "master server http listen port" )
masterMetaFolder = cmdServer . Flag . String ( "master.dir" , "" , "data directory to store meta data, default to same as -dir specified" )
masterVolumeSizeLimitMB = cmdServer . Flag . Uint ( "master.volumeSizeLimitMB" , 30 * 1000 , "Master stops directing writes to oversized volumes." )
2017-01-09 03:01:46 +08:00
masterVolumePreallocate = cmdServer . Flag . Bool ( "master.volumePreallocate" , false , "Preallocate disk space for volumes." )
2014-04-26 13:09:42 +08:00
masterDefaultReplicaPlacement = cmdServer . Flag . String ( "master.defaultReplicaPlacement" , "000" , "Default replication type if not specified." )
volumePort = cmdServer . Flag . Int ( "volume.port" , 8080 , "volume server http listen port" )
2015-03-09 16:10:01 +08:00
volumePublicPort = cmdServer . Flag . Int ( "volume.port.public" , 0 , "volume server public port" )
2014-03-03 14:16:54 +08:00
volumeDataFolders = cmdServer . Flag . String ( "dir" , os . TempDir ( ) , "directories to store data files. dir[,dir]..." )
2014-04-26 13:09:42 +08:00
volumeMaxDataVolumeCounts = cmdServer . Flag . String ( "volume.max" , "7" , "maximum numbers of volumes, count[,count]..." )
2014-03-16 14:03:49 +08:00
volumePulse = cmdServer . Flag . Int ( "pulseSeconds" , 5 , "number of seconds between heartbeats" )
2017-05-27 13:51:25 +08:00
volumeIndexType = cmdServer . Flag . String ( "volume.index" , "memory" , "Choose [memory|leveldb|boltdb|btree] mode for memory~performance balance." )
2014-05-15 16:16:56 +08:00
volumeFixJpgOrientation = cmdServer . Flag . Bool ( "volume.images.fix.orientation" , true , "Adjust jpg orientation when uploading." )
2015-08-04 05:43:15 +08:00
volumeReadRedirect = cmdServer . Flag . Bool ( "volume.read.redirect" , true , "Redirect moved or non-local volumes." )
2015-04-17 00:16:43 +08:00
volumeServerPublicUrl = cmdServer . Flag . String ( "volume.publicUrl" , "" , "publicly accessible address" )
2014-03-31 11:57:25 +08:00
isStartingFiler = cmdServer . Flag . Bool ( "filer" , false , "whether to start filer" )
2013-12-04 15:22:26 +08:00
2016-08-06 05:45:48 +08:00
serverWhiteList [ ] string
2013-12-04 15:22:26 +08:00
)
2014-03-31 11:57:25 +08:00
func init ( ) {
2014-11-29 08:34:03 +08:00
serverOptions . cpuprofile = cmdServer . Flag . String ( "cpuprofile" , "" , "cpu profile output file" )
2014-05-13 15:03:10 +08:00
filerOptions . collection = cmdServer . Flag . String ( "filer.collection" , "" , "all data will be stored in this collection" )
filerOptions . port = cmdServer . Flag . Int ( "filer.port" , 8888 , "filer server http listen port" )
2018-08-13 16:20:49 +08:00
filerOptions . grpcPort = cmdServer . Flag . Int ( "filer.port.grpc" , 0 , "filer grpc server listen port, default to http port + 10000" )
2017-05-28 11:14:22 +08:00
filerOptions . publicPort = cmdServer . Flag . Int ( "filer.port.public" , 0 , "filer server public http listen port" )
2014-05-13 15:03:10 +08:00
filerOptions . defaultReplicaPlacement = cmdServer . Flag . String ( "filer.defaultReplicaPlacement" , "" , "Default replication type if not specified during runtime." )
2014-12-09 12:27:26 +08:00
filerOptions . redirectOnRead = cmdServer . Flag . Bool ( "filer.redirectOnRead" , false , "whether proxy or redirect to volume server during file GET request" )
2015-04-14 14:38:46 +08:00
filerOptions . disableDirListing = cmdServer . Flag . Bool ( "filer.disableDirListing" , false , "turn off directory listing" )
2017-09-15 23:24:30 +08:00
filerOptions . maxMB = cmdServer . Flag . Int ( "filer.maxMB" , 32 , "split files larger than the limit" )
2018-07-08 17:11:36 +08:00
filerOptions . dirListingLimit = cmdServer . Flag . Int ( "filer.dirListLimit" , 1000 , "limit sub dir listing size" )
2014-03-31 11:57:25 +08:00
}
2013-12-04 15:22:26 +08:00
func runServer ( cmd * Command , args [ ] string ) bool {
2015-02-08 07:35:28 +08:00
filerOptions . secretKey = serverSecureKey
2014-05-08 01:17:06 +08:00
if * serverOptions . cpuprofile != "" {
f , err := os . Create ( * serverOptions . cpuprofile )
if err != nil {
glog . Fatal ( err )
}
pprof . StartCPUProfile ( f )
defer pprof . StopCPUProfile ( )
}
2014-03-31 11:57:25 +08:00
2014-12-14 16:33:16 +08:00
if * filerOptions . redirectOnRead {
* isStartingFiler = true
}
2018-06-01 15:39:39 +08:00
master := * serverIp + ":" + strconv . Itoa ( * masterPort )
2017-05-28 11:14:22 +08:00
filerOptions . ip = serverIp
2014-03-31 11:57:25 +08:00
2018-07-09 17:22:48 +08:00
filerOptions . dataCenter = serverDataCenter
2014-05-13 15:03:10 +08:00
if * filerOptions . defaultReplicaPlacement == "" {
* filerOptions . defaultReplicaPlacement = * masterDefaultReplicaPlacement
2014-03-31 11:57:25 +08:00
}
2015-03-09 16:10:01 +08:00
if * volumePublicPort == 0 {
* volumePublicPort = * volumePort
2015-02-26 13:28:55 +08:00
}
2013-12-04 15:22:26 +08:00
if * serverMaxCpu < 1 {
* serverMaxCpu = runtime . NumCPU ( )
}
runtime . GOMAXPROCS ( * serverMaxCpu )
folders := strings . Split ( * volumeDataFolders , "," )
maxCountStrings := strings . Split ( * volumeMaxDataVolumeCounts , "," )
2015-03-10 15:20:31 +08:00
var maxCounts [ ] int
2013-12-04 15:22:26 +08:00
for _ , maxString := range maxCountStrings {
if max , e := strconv . Atoi ( maxString ) ; e == nil {
maxCounts = append ( maxCounts , max )
} else {
2014-04-17 15:16:44 +08:00
glog . Fatalf ( "The max specified in -max not a valid number %s" , maxString )
2013-12-04 15:22:26 +08:00
}
}
if len ( folders ) != len ( maxCounts ) {
glog . Fatalf ( "%d directories by -dir, but only %d max is set by -max" , len ( folders ) , len ( maxCounts ) )
}
for _ , folder := range folders {
2013-12-10 05:27:09 +08:00
if err := util . TestFolderWritable ( folder ) ; err != nil {
glog . Fatalf ( "Check Data Folder(-dir) Writable %s : %s" , folder , err )
2013-12-04 15:22:26 +08:00
}
}
2017-07-17 12:40:47 +08:00
if * masterVolumeSizeLimitMB > 30 * 1000 {
glog . Fatalf ( "masterVolumeSizeLimitMB should be less than 30000" )
}
2014-03-31 11:57:25 +08:00
if * masterMetaFolder == "" {
* masterMetaFolder = folders [ 0 ]
}
if err := util . TestFolderWritable ( * masterMetaFolder ) ; err != nil {
glog . Fatalf ( "Check Meta Folder (-mdir=\"%s\") Writable: %s" , * masterMetaFolder , err )
}
2016-08-06 05:45:48 +08:00
if * serverWhiteListOption != "" {
serverWhiteList = strings . Split ( * serverWhiteListOption , "," )
2013-12-04 15:22:26 +08:00
}
2014-03-31 11:57:25 +08:00
if * isStartingFiler {
go func ( ) {
2016-07-21 14:45:55 +08:00
time . Sleep ( 1 * time . Second )
2017-05-28 11:14:22 +08:00
filerOptions . start ( )
2014-03-31 11:57:25 +08:00
} ( )
}
2014-03-31 02:28:04 +08:00
2014-03-16 14:03:49 +08:00
var raftWaitForMaster sync . WaitGroup
var volumeWait sync . WaitGroup
raftWaitForMaster . Add ( 1 )
volumeWait . Add ( 1 )
2013-12-04 15:22:26 +08:00
go func ( ) {
r := mux . NewRouter ( )
2014-03-26 04:46:59 +08:00
ms := weed_server . NewMasterServer ( r , * masterPort , * masterMetaFolder ,
2017-01-09 03:01:46 +08:00
* masterVolumeSizeLimitMB , * masterVolumePreallocate ,
2017-02-13 13:58:44 +08:00
* volumePulse , * masterDefaultReplicaPlacement , * serverGarbageThreshold ,
2016-08-06 05:45:48 +08:00
serverWhiteList , * serverSecureKey ,
2013-12-04 15:22:26 +08:00
)
2014-09-21 14:30:35 +08:00
glog . V ( 0 ) . Infoln ( "Start Seaweed Master" , util . VERSION , "at" , * serverIp + ":" + strconv . Itoa ( * masterPort ) )
2017-01-10 17:30:00 +08:00
masterListener , e := util . NewListener ( * serverBindIp + ":" + strconv . Itoa ( * masterPort ) , 0 )
2014-03-21 02:07:15 +08:00
if e != nil {
2015-01-14 09:04:41 +08:00
glog . Fatalf ( "Master startup error: %v" , e )
2013-12-04 15:22:26 +08:00
}
2013-12-10 05:27:09 +08:00
go func ( ) {
2014-03-16 14:03:49 +08:00
raftWaitForMaster . Wait ( )
2013-12-10 05:27:09 +08:00
time . Sleep ( 100 * time . Millisecond )
2018-08-13 05:25:31 +08:00
myAddress , peers := checkPeers ( * serverIp , * masterPort , * serverPeers )
2014-04-17 14:43:27 +08:00
raftServer := weed_server . NewRaftServer ( r , peers , myAddress , * masterMetaFolder , ms . Topo , * volumePulse )
2014-02-05 17:54:52 +08:00
ms . SetRaftServer ( raftServer )
2014-03-16 14:03:49 +08:00
volumeWait . Done ( )
2013-12-10 05:27:09 +08:00
} ( )
2014-03-16 14:03:49 +08:00
raftWaitForMaster . Done ( )
2017-01-10 17:01:12 +08:00
// start grpc and http server
m := cmux . New ( masterListener )
grpcL := m . Match ( cmux . HTTP2HeaderField ( "content-type" , "application/grpc" ) )
httpL := m . Match ( cmux . Any ( ) )
// Create your protocol servers.
2018-07-04 10:07:55 +08:00
grpcS := util . NewGrpcServer ( )
2018-05-10 14:11:54 +08:00
master_pb . RegisterSeaweedServer ( grpcS , ms )
2017-01-10 17:01:12 +08:00
reflection . Register ( grpcS )
httpS := & http . Server { Handler : r }
go grpcS . Serve ( grpcL )
go httpS . Serve ( httpL )
if err := m . Serve ( ) ; err != nil {
glog . Fatalf ( "master server failed to serve: %v" , err )
2013-12-04 15:22:26 +08:00
}
2017-01-10 17:01:12 +08:00
2013-12-04 15:22:26 +08:00
} ( )
2014-03-16 14:03:49 +08:00
volumeWait . Wait ( )
2013-12-10 05:27:09 +08:00
time . Sleep ( 100 * time . Millisecond )
2015-03-09 16:10:01 +08:00
if * volumePublicPort == 0 {
* volumePublicPort = * volumePort
}
2015-04-17 00:16:43 +08:00
if * volumeServerPublicUrl == "" {
* volumeServerPublicUrl = * serverIp + ":" + strconv . Itoa ( * volumePublicPort )
2015-04-09 02:08:08 +08:00
}
2015-03-09 16:10:01 +08:00
isSeperatedPublicPort := * volumePublicPort != * volumePort
volumeMux := http . NewServeMux ( )
publicVolumeMux := volumeMux
if isSeperatedPublicPort {
publicVolumeMux = http . NewServeMux ( )
}
Add boltdb for volume needle map
boltdb is fairly slow to write, about 6 minutes for recreating index
for 1553934 files. Boltdb loads 1,553,934 x 16 = 24,862,944bytes from
disk, and generate the boltdb as large as 134,217,728 bytes in 6
minutes.
To compare, for leveldb, it recreates index in leveldb as large as
27,188,148 bytes in 8 seconds.
For in memory version, it loads the index in
To test the memory consumption, the leveldb or boltdb index are
created. And the server is restarted. Using the benchmark tool to read
lots of files. There are 7 volumes in benchmark collection, each with
about 1553K files.
For leveldb, the memory starts at 142,884KB, and stays at 179,340KB.
For boltdb, the memory starts at 73,756KB, and stays at 144,564KB.
For in-memory, the memory starts at 368,152KB, and stays at 448,032KB.
2015-03-30 02:04:32 +08:00
volumeNeedleMapKind := storage . NeedleMapInMemory
switch * volumeIndexType {
case "leveldb" :
volumeNeedleMapKind = storage . NeedleMapLevelDb
case "boltdb" :
volumeNeedleMapKind = storage . NeedleMapBoltDb
2017-05-27 13:51:25 +08:00
case "btree" :
volumeNeedleMapKind = storage . NeedleMapBtree
Add boltdb for volume needle map
boltdb is fairly slow to write, about 6 minutes for recreating index
for 1553934 files. Boltdb loads 1,553,934 x 16 = 24,862,944bytes from
disk, and generate the boltdb as large as 134,217,728 bytes in 6
minutes.
To compare, for leveldb, it recreates index in leveldb as large as
27,188,148 bytes in 8 seconds.
For in memory version, it loads the index in
To test the memory consumption, the leveldb or boltdb index are
created. And the server is restarted. Using the benchmark tool to read
lots of files. There are 7 volumes in benchmark collection, each with
about 1553K files.
For leveldb, the memory starts at 142,884KB, and stays at 179,340KB.
For boltdb, the memory starts at 73,756KB, and stays at 144,564KB.
For in-memory, the memory starts at 368,152KB, and stays at 448,032KB.
2015-03-30 02:04:32 +08:00
}
2015-03-09 16:10:01 +08:00
volumeServer := weed_server . NewVolumeServer ( volumeMux , publicVolumeMux ,
2015-04-17 00:16:43 +08:00
* serverIp , * volumePort , * volumeServerPublicUrl ,
2015-01-19 09:03:38 +08:00
folders , maxCounts ,
Add boltdb for volume needle map
boltdb is fairly slow to write, about 6 minutes for recreating index
for 1553934 files. Boltdb loads 1,553,934 x 16 = 24,862,944bytes from
disk, and generate the boltdb as large as 134,217,728 bytes in 6
minutes.
To compare, for leveldb, it recreates index in leveldb as large as
27,188,148 bytes in 8 seconds.
For in memory version, it loads the index in
To test the memory consumption, the leveldb or boltdb index are
created. And the server is restarted. Using the benchmark tool to read
lots of files. There are 7 volumes in benchmark collection, each with
about 1553K files.
For leveldb, the memory starts at 142,884KB, and stays at 179,340KB.
For boltdb, the memory starts at 73,756KB, and stays at 144,564KB.
For in-memory, the memory starts at 368,152KB, and stays at 448,032KB.
2015-03-30 02:04:32 +08:00
volumeNeedleMapKind ,
2018-06-01 15:39:39 +08:00
[ ] string { master } , * volumePulse , * serverDataCenter , * serverRack ,
2016-08-06 05:45:48 +08:00
serverWhiteList , * volumeFixJpgOrientation , * volumeReadRedirect ,
2013-12-04 15:22:26 +08:00
)
2014-09-21 14:30:35 +08:00
glog . V ( 0 ) . Infoln ( "Start Seaweed volume server" , util . VERSION , "at" , * serverIp + ":" + strconv . Itoa ( * volumePort ) )
2015-02-26 13:28:55 +08:00
volumeListener , eListen := util . NewListener (
2014-09-21 14:30:35 +08:00
* serverBindIp + ":" + strconv . Itoa ( * volumePort ) ,
2014-03-21 02:07:15 +08:00
time . Duration ( * serverTimeout ) * time . Second ,
)
2015-02-26 13:28:55 +08:00
if eListen != nil {
glog . Fatalf ( "Volume server listener error: %v" , eListen )
2014-03-21 02:07:15 +08:00
}
2015-03-09 16:10:01 +08:00
if isSeperatedPublicPort {
publicListeningAddress := * serverIp + ":" + strconv . Itoa ( * volumePublicPort )
glog . V ( 0 ) . Infoln ( "Start Seaweed volume server" , util . VERSION , "public at" , publicListeningAddress )
publicListener , e := util . NewListener ( publicListeningAddress , time . Duration ( * serverTimeout ) * time . Second )
if e != nil {
glog . Fatalf ( "Volume server listener error:%v" , e )
}
go func ( ) {
if e := http . Serve ( publicListener , publicVolumeMux ) ; e != nil {
glog . Fatalf ( "Volume server fail to serve public: %v" , e )
}
} ( )
}
2014-05-13 15:03:10 +08:00
2017-06-22 16:33:58 +08:00
util . OnInterrupt ( func ( ) {
2014-05-14 06:04:04 +08:00
volumeServer . Shutdown ( )
pprof . StopCPUProfile ( )
} )
2014-05-13 15:03:10 +08:00
2015-03-09 16:10:01 +08:00
if e := http . Serve ( volumeListener , volumeMux ) ; e != nil {
2015-01-14 09:04:41 +08:00
glog . Fatalf ( "Volume server fail to serve:%v" , e )
2013-12-04 15:22:26 +08:00
}
return true
}