gitlab-com-seaweedfs/seaweedfs
1
0
Fork 0
mirror of https://github.com/seaweedfs/seaweedfs.git synced 2024-01-19 02:48:24 +00:00
Code Issues Projects Releases Wiki Activity

fsck replicas

Browse source
This commit is contained in:
Konstantin Lebedev 2022-03-31 19:10:06 +05:00
parent 67d870dfc2
commit 1d9e30d8c0
1 changed files with 103 additions and 81 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

78
weed/shell/command_volume_fsck.go
View file

@ -11,6 +11,7 @@ import (
"net/http" "net/http"
"net/url" "net/url"
"os" "os"
"path"
"path/filepath" "path/filepath"
"strings" "strings"
"sync" "sync"
@ -65,8 +66,11 @@ func (c *commandVolumeFsck) Do(args []string, commandEnv *CommandEnv, writer io.
verbose := fsckCommand.Bool("v", false, "verbose mode") verbose := fsckCommand.Bool("v", false, "verbose mode")
findMissingChunksInFiler := fsckCommand.Bool("findMissingChunksInFiler", false, "see \"help volume.fsck\"") findMissingChunksInFiler := fsckCommand.Bool("findMissingChunksInFiler", false, "see \"help volume.fsck\"")
findMissingChunksInFilerPath := fsckCommand.String("findMissingChunksInFilerPath", "/", "used together with findMissingChunksInFiler") findMissingChunksInFilerPath := fsckCommand.String("findMissingChunksInFilerPath", "/", "used together with findMissingChunksInFiler")
findMissingChunksInVolumeId := fsckCommand.Int("findMissingChunksInVolumeId", 0, "used together with findMissingChunksInFiler")
applyPurging := fsckCommand.Bool("reallyDeleteFromVolume", false, "<expert only!> after detection, delete missing data from volumes / delete missing file entries from filer") applyPurging := fsckCommand.Bool("reallyDeleteFromVolume", false, "<expert only!> after detection, delete missing data from volumes / delete missing file entries from filer")
purgeAbsent := fsckCommand.Bool("reallyDeleteFilerEntries", false, "<expert only!> delete missing file entries from filer if the corresponding volume is missing for any reason, please ensure all still existing/expected volumes are connected! used together with findMissingChunksInFiler") purgeAbsent := fsckCommand.Bool("reallyDeleteFilerEntries", false, "<expert only!> delete missing file entries from filer if the corresponding volume is missing for any reason, please ensure all still existing/expected volumes are connected! used together with findMissingChunksInFiler")
tempPath := fsckCommand.String("tempPath", path.Join(os.TempDir()), "path for temporary idx files")
if err = fsckCommand.Parse(args); err != nil { if err = fsckCommand.Parse(args); err != nil {
return nil return nil
} }
@ -78,7 +82,7 @@ func (c *commandVolumeFsck) Do(args []string, commandEnv *CommandEnv, writer io.
c.env = commandEnv c.env = commandEnv
// create a temp folder // create a temp folder
tempFolder, err := os.MkdirTemp("", "sw_fsck") tempFolder, err := os.MkdirTemp(*tempPath, "sw_fsck")
if err != nil { if err != nil {
return fmt.Errorf("failed to create temp folder: %v", err) return fmt.Errorf("failed to create temp folder: %v", err)
} }
@ -88,14 +92,14 @@ func (c *commandVolumeFsck) Do(args []string, commandEnv *CommandEnv, writer io.
defer os.RemoveAll(tempFolder) defer os.RemoveAll(tempFolder)
// collect all volume id locations // collect all volume id locations
volumeIdToVInfo, err := c.collectVolumeIds(commandEnv, *verbose, writer) dataNodeVolumeIdToVInfo, err := c.collectVolumeIds(commandEnv, *verbose, writer)
if err != nil { if err != nil {
return fmt.Errorf("failed to collect all volume locations: %v", err) return fmt.Errorf("failed to collect all volume locations: %v", err)
} }
isBucketsPath := false isBucketsPath := false
var fillerBucketsPath string var fillerBucketsPath string
if *findMissingChunksInFiler && *findMissingChunksInFilerPath != "" { if *findMissingChunksInFiler && *findMissingChunksInFilerPath != "/" {
fillerBucketsPath, err = readFilerBucketsPath(commandEnv) fillerBucketsPath, err = readFilerBucketsPath(commandEnv)
if err != nil { if err != nil {
return fmt.Errorf("read filer buckets path: %v", err) return fmt.Errorf("read filer buckets path: %v", err)
@ -109,33 +113,41 @@ func (c *commandVolumeFsck) Do(args []string, commandEnv *CommandEnv, writer io.
} }
// collect each volume file ids // collect each volume file ids
for dataNodeId, volumeIdToVInfo := range dataNodeVolumeIdToVInfo {
for volumeId, vinfo := range volumeIdToVInfo { for volumeId, vinfo := range volumeIdToVInfo {
if *findMissingChunksInVolumeId > 0 && uint32(*findMissingChunksInVolumeId) != volumeId {
delete(volumeIdToVInfo, volumeId)
continue
}
if isBucketsPath && !strings.HasPrefix(*findMissingChunksInFilerPath, fillerBucketsPath+"/"+vinfo.collection) { if isBucketsPath && !strings.HasPrefix(*findMissingChunksInFilerPath, fillerBucketsPath+"/"+vinfo.collection) {
delete(volumeIdToVInfo, volumeId) delete(volumeIdToVInfo, volumeId)
continue continue
} }
err = c.collectOneVolumeFileIds(tempFolder, volumeId, vinfo, *verbose, writer) err = c.collectOneVolumeFileIds(tempFolder, dataNodeId, volumeId, vinfo, *verbose, writer)
if err != nil { if err != nil {
return fmt.Errorf("failed to collect file ids from volume %d on %s: %v", volumeId, vinfo.server, err) return fmt.Errorf("failed to collect file ids from volume %d on %s: %v", volumeId, vinfo.server, err)
} }
} }
}
if *findMissingChunksInFiler { if *findMissingChunksInFiler {
// collect all filer file ids and paths // collect all filer file ids and paths
if err = c.collectFilerFileIdAndPaths(volumeIdToVInfo, tempFolder, writer, *findMissingChunksInFilerPath, *verbose, *purgeAbsent); err != nil { if err = c.collectFilerFileIdAndPaths(dataNodeVolumeIdToVInfo, tempFolder, writer, *findMissingChunksInFilerPath, *verbose, *purgeAbsent); err != nil {
return fmt.Errorf("collectFilerFileIdAndPaths: %v", err) return fmt.Errorf("collectFilerFileIdAndPaths: %v", err)
} }
for dataNodeId, volumeIdToVInfo := range dataNodeVolumeIdToVInfo {
// for each volume, check filer file ids // for each volume, check filer file ids
if err = c.findFilerChunksMissingInVolumeServers(volumeIdToVInfo, tempFolder, writer, *verbose, *applyPurging); err != nil { if err = c.findFilerChunksMissingInVolumeServers(volumeIdToVInfo, tempFolder, dataNodeId, writer, *verbose, *applyPurging); err != nil {
return fmt.Errorf("findFilerChunksMissingInVolumeServers: %v", err) return fmt.Errorf("findFilerChunksMissingInVolumeServers: %v", err)
} }
}
} else { } else {
// collect all filer file ids // collect all filer file ids
if err = c.collectFilerFileIds(volumeIdToVInfo, tempFolder, writer, *verbose); err != nil { if err = c.collectFilerFileIds(dataNodeVolumeIdToVInfo, tempFolder, writer, *verbose); err != nil {
return fmt.Errorf("failed to collect file ids from filer: %v", err) return fmt.Errorf("failed to collect file ids from filer: %v", err)
} }
// volume file ids subtract filer file ids // volume file ids subtract filer file ids
if err = c.findExtraChunksInVolumeServers(volumeIdToVInfo, tempFolder, writer, *verbose, *applyPurging); err != nil { if err = c.findExtraChunksInVolumeServers(dataNodeVolumeIdToVInfo, tempFolder, writer, *verbose, *applyPurging); err != nil {
return fmt.Errorf("findExtraChunksInVolumeServers: %v", err) return fmt.Errorf("findExtraChunksInVolumeServers: %v", err)
} }
} }
@ -143,20 +155,24 @@ func (c *commandVolumeFsck) Do(args []string, commandEnv *CommandEnv, writer io.
return nil return nil
} }
func (c *commandVolumeFsck) collectFilerFileIdAndPaths(volumeIdToServer map[uint32]VInfo, tempFolder string, writer io.Writer, filerPath string, verbose bool, purgeAbsent bool) error { func (c *commandVolumeFsck) collectFilerFileIdAndPaths(dataNodeVolumeIdToVInfo map[string]map[uint32]VInfo, tempFolder string, writer io.Writer, filerPath string, verbose bool, purgeAbsent bool) error {
if verbose { if verbose {
fmt.Fprintf(writer, "checking each file from filer ...\n") fmt.Fprintf(writer, "checking each file from filer ...\n")
} }
files := make(map[uint32]*os.File) files := make(map[uint32]*os.File)
for _, volumeIdToServer := range dataNodeVolumeIdToVInfo {
for vid := range volumeIdToServer { for vid := range volumeIdToServer {
dst, openErr := os.OpenFile(getFilerFileIdFile(tempFolder, vid), os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644) dst, openErr := os.OpenFile(getFilerFileIdFile(tempFolder, vid), os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644)
if openErr != nil { if openErr != nil {
return fmt.Errorf("failed to create file %s: %v", getFilerFileIdFile(tempFolder, vid), openErr) return fmt.Errorf("failed to create file %s: %v", getFilerFileIdFile(tempFolder, vid), openErr)
} }
if _, ok := volumeIdToServer[vid]; !ok {
files[vid] = dst files[vid] = dst
} }
}
}
defer func() { defer func() {
for _, f := range files { for _, f := range files {
f.Close() f.Close()
@ -210,10 +226,10 @@ func (c *commandVolumeFsck) collectFilerFileIdAndPaths(volumeIdToServer map[uint
} }
func (c *commandVolumeFsck) findFilerChunksMissingInVolumeServers(volumeIdToVInfo map[uint32]VInfo, tempFolder string, writer io.Writer, verbose bool, applyPurging bool) error { func (c *commandVolumeFsck) findFilerChunksMissingInVolumeServers(volumeIdToVInfo map[uint32]VInfo, tempFolder string, dataNodeId string, writer io.Writer, verbose bool, applyPurging bool) error {
for volumeId, vinfo := range volumeIdToVInfo { for volumeId, vinfo := range volumeIdToVInfo {
checkErr := c.oneVolumeFileIdsCheckOneVolume(tempFolder, volumeId, writer, verbose, applyPurging) checkErr := c.oneVolumeFileIdsCheckOneVolume(tempFolder, dataNodeId, volumeId, writer, verbose, applyPurging)
if checkErr != nil { if checkErr != nil {
return fmt.Errorf("failed to collect file ids from volume %d on %s: %v", volumeId, vinfo.server, checkErr) return fmt.Errorf("failed to collect file ids from volume %d on %s: %v", volumeId, vinfo.server, checkErr)
} }
@ -221,12 +237,13 @@ func (c *commandVolumeFsck) findFilerChunksMissingInVolumeServers(volumeIdToVInf
return nil return nil
} }
func (c *commandVolumeFsck) findExtraChunksInVolumeServers(volumeIdToVInfo map[uint32]VInfo, tempFolder string, writer io.Writer, verbose bool, applyPurging bool) error { func (c *commandVolumeFsck) findExtraChunksInVolumeServers(dataNodeVolumeIdToVInfo map[string]map[uint32]VInfo, tempFolder string, writer io.Writer, verbose bool, applyPurging bool) error {
var totalInUseCount, totalOrphanChunkCount, totalOrphanDataSize uint64 var totalInUseCount, totalOrphanChunkCount, totalOrphanDataSize uint64
for dataNodeId, volumeIdToVInfo := range dataNodeVolumeIdToVInfo {
for volumeId, vinfo := range volumeIdToVInfo { for volumeId, vinfo := range volumeIdToVInfo {
inUseCount, orphanFileIds, orphanDataSize, checkErr := c.oneVolumeFileIdsSubtractFilerFileIds(tempFolder, volumeId, writer, verbose) inUseCount, orphanFileIds, orphanDataSize, checkErr := c.oneVolumeFileIdsSubtractFilerFileIds(tempFolder, dataNodeId, volumeId, writer, verbose)
if checkErr != nil { if checkErr != nil {
return fmt.Errorf("failed to collect file ids from volume %d on %s: %v", volumeId, vinfo.server, checkErr) return fmt.Errorf("failed to collect file ids from volume %d on %s: %v", volumeId, vinfo.server, checkErr)
} }
@ -273,6 +290,7 @@ func (c *commandVolumeFsck) findExtraChunksInVolumeServers(volumeIdToVInfo map[u
} }
} }
} }
}
if !applyPurging { if !applyPurging {
pct := float64(totalOrphanChunkCount*100) / (float64(totalOrphanChunkCount + totalInUseCount)) pct := float64(totalOrphanChunkCount*100) / (float64(totalOrphanChunkCount + totalInUseCount))
@ -290,7 +308,7 @@ func (c *commandVolumeFsck) findExtraChunksInVolumeServers(volumeIdToVInfo map[u
return nil return nil
} }
func (c *commandVolumeFsck) collectOneVolumeFileIds(tempFolder string, volumeId uint32, vinfo VInfo, verbose bool, writer io.Writer) error { func (c *commandVolumeFsck) collectOneVolumeFileIds(tempFolder string, dataNodeId string, volumeId uint32, vinfo VInfo, verbose bool, writer io.Writer) error {
if verbose { if verbose {
fmt.Fprintf(writer, "collecting volume %d file ids from %s ...\n", volumeId, vinfo.server) fmt.Fprintf(writer, "collecting volume %d file ids from %s ...\n", volumeId, vinfo.server)
@ -316,7 +334,7 @@ func (c *commandVolumeFsck) collectOneVolumeFileIds(tempFolder string, volumeId
return fmt.Errorf("failed to start copying volume %d%s: %v", volumeId, ext, err) return fmt.Errorf("failed to start copying volume %d%s: %v", volumeId, ext, err)
} }
err = writeToFile(copyFileClient, getVolumeFileIdFile(tempFolder, volumeId)) err = writeToFile(copyFileClient, getVolumeFileIdFile(tempFolder, dataNodeId, volumeId))
if err != nil { if err != nil {
return fmt.Errorf("failed to copy %d%s from %s: %v", volumeId, ext, vinfo.server, err) return fmt.Errorf("failed to copy %d%s from %s: %v", volumeId, ext, vinfo.server, err)
} }
@ -327,13 +345,14 @@ func (c *commandVolumeFsck) collectOneVolumeFileIds(tempFolder string, volumeId
} }
func (c *commandVolumeFsck) collectFilerFileIds(volumeIdToServer map[uint32]VInfo, tempFolder string, writer io.Writer, verbose bool) error { func (c *commandVolumeFsck) collectFilerFileIds(dataNodeVolumeIdToVInfo map[string]map[uint32]VInfo, tempFolder string, writer io.Writer, verbose bool) error {
if verbose { if verbose {
fmt.Fprintf(writer, "collecting file ids from filer ...\n") fmt.Fprintf(writer, "collecting file ids from filer ...\n")
} }
files := make(map[uint32]*os.File) files := make(map[uint32]*os.File)
for _, volumeIdToServer := range dataNodeVolumeIdToVInfo {
for vid := range volumeIdToServer { for vid := range volumeIdToServer {
dst, openErr := os.OpenFile(getFilerFileIdFile(tempFolder, vid), os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644) dst, openErr := os.OpenFile(getFilerFileIdFile(tempFolder, vid), os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644)
if openErr != nil { if openErr != nil {
@ -341,6 +360,7 @@ func (c *commandVolumeFsck) collectFilerFileIds(volumeIdToServer map[uint32]VInf
} }
files[vid] = dst files[vid] = dst
} }
}
defer func() { defer func() {
for _, f := range files { for _, f := range files {
f.Close() f.Close()
@ -377,7 +397,7 @@ func (c *commandVolumeFsck) collectFilerFileIds(volumeIdToServer map[uint32]VInf
}) })
} }
func (c *commandVolumeFsck) oneVolumeFileIdsCheckOneVolume(tempFolder string, volumeId uint32, writer io.Writer, verbose bool, applyPurging bool) (err error) { func (c *commandVolumeFsck) oneVolumeFileIdsCheckOneVolume(tempFolder string, dataNodeId string, volumeId uint32, writer io.Writer, verbose bool, applyPurging bool) (err error) {
if verbose { if verbose {
fmt.Fprintf(writer, "find missing file chunks in volume %d ...\n", volumeId) fmt.Fprintf(writer, "find missing file chunks in volume %d ...\n", volumeId)
@ -386,7 +406,7 @@ func (c *commandVolumeFsck) oneVolumeFileIdsCheckOneVolume(tempFolder string, vo
db := needle_map.NewMemDb() db := needle_map.NewMemDb()
defer db.Close() defer db.Close()
if err = db.LoadFromIdx(getVolumeFileIdFile(tempFolder, volumeId)); err != nil { if err = db.LoadFromIdx(getVolumeFileIdFile(tempFolder, dataNodeId, volumeId)); err != nil {
return return
} }
@ -473,12 +493,12 @@ func (c *commandVolumeFsck) httpDelete(path util.FullPath, verbose bool) {
} }
} }
func (c *commandVolumeFsck) oneVolumeFileIdsSubtractFilerFileIds(tempFolder string, volumeId uint32, writer io.Writer, verbose bool) (inUseCount uint64, orphanFileIds []string, orphanDataSize uint64, err error) { func (c *commandVolumeFsck) oneVolumeFileIdsSubtractFilerFileIds(tempFolder string, dataNodeId string, volumeId uint32, writer io.Writer, verbose bool) (inUseCount uint64, orphanFileIds []string, orphanDataSize uint64, err error) {
db := needle_map.NewMemDb() db := needle_map.NewMemDb()
defer db.Close() defer db.Close()
if err = db.LoadFromIdx(getVolumeFileIdFile(tempFolder, volumeId)); err != nil { if err = db.LoadFromIdx(getVolumeFileIdFile(tempFolder, dataNodeId, volumeId)); err != nil {
return return
} }
@ -509,8 +529,8 @@ func (c *commandVolumeFsck) oneVolumeFileIdsSubtractFilerFileIds(tempFolder stri
if orphanFileCount > 0 { if orphanFileCount > 0 {
pct := float64(orphanFileCount*100) / (float64(orphanFileCount + inUseCount)) pct := float64(orphanFileCount*100) / (float64(orphanFileCount + inUseCount))
fmt.Fprintf(writer, "volume:%d\tentries:%d\torphan:%d\t%.2f%%\t%dB\n", fmt.Fprintf(writer, "dataNode:%s\tvolume:%d\tentries:%d\torphan:%d\t%.2f%%\t%dB\n",
volumeId, orphanFileCount+inUseCount, orphanFileCount, pct, orphanDataSize) dataNodeId, volumeId, orphanFileCount+inUseCount, orphanFileCount, pct, orphanDataSize)
} }
return return
@ -524,13 +544,13 @@ type VInfo struct {
isReadOnly bool isReadOnly bool
} }
func (c *commandVolumeFsck) collectVolumeIds(commandEnv *CommandEnv, verbose bool, writer io.Writer) (volumeIdToServer map[uint32]VInfo, err error) { func (c *commandVolumeFsck) collectVolumeIds(commandEnv *CommandEnv, verbose bool, writer io.Writer) (volumeIdToServer map[string]map[uint32]VInfo, err error) {
if verbose { if verbose {
fmt.Fprintf(writer, "collecting volume id and locations from master ...\n") fmt.Fprintf(writer, "collecting volume id and locations from master ...\n")
} }
volumeIdToServer = make(map[uint32]VInfo) volumeIdToServer = make(map[string]map[uint32]VInfo)
// collect topology information // collect topology information
topologyInfo, _, err := collectTopologyInfo(commandEnv, 0) topologyInfo, _, err := collectTopologyInfo(commandEnv, 0)
if err != nil { if err != nil {
@ -539,8 +559,10 @@ func (c *commandVolumeFsck) collectVolumeIds(commandEnv *CommandEnv, verbose boo
eachDataNode(topologyInfo, func(dc string, rack RackId, t *master_pb.DataNodeInfo) { eachDataNode(topologyInfo, func(dc string, rack RackId, t *master_pb.DataNodeInfo) {
for _, diskInfo := range t.DiskInfos { for _, diskInfo := range t.DiskInfos {
dataNodeId := t.GetId()
volumeIdToServer[dataNodeId] = make(map[uint32]VInfo)
for _, vi := range diskInfo.VolumeInfos { for _, vi := range diskInfo.VolumeInfos {
volumeIdToServer[vi.Id] = VInfo{ volumeIdToServer[dataNodeId][vi.Id] = VInfo{
server: pb.NewServerAddressFromDataNode(t), server: pb.NewServerAddressFromDataNode(t),
collection: vi.Collection, collection: vi.Collection,
isEcVolume: false, isEcVolume: false,
@ -548,7 +570,7 @@ func (c *commandVolumeFsck) collectVolumeIds(commandEnv *CommandEnv, verbose boo
} }
} }
for _, ecShardInfo := range diskInfo.EcShardInfos { for _, ecShardInfo := range diskInfo.EcShardInfos {
volumeIdToServer[ecShardInfo.Id] = VInfo{ volumeIdToServer[dataNodeId][ecShardInfo.Id] = VInfo{
server: pb.NewServerAddressFromDataNode(t), server: pb.NewServerAddressFromDataNode(t),
collection: ecShardInfo.Collection, collection: ecShardInfo.Collection,
isEcVolume: true, isEcVolume: true,
@ -600,8 +622,8 @@ func (c *commandVolumeFsck) purgeFileIdsForOneVolume(volumeId uint32, fileIds []
return return
} }
func getVolumeFileIdFile(tempFolder string, vid uint32) string { func getVolumeFileIdFile(tempFolder string, dataNodeid string, vid uint32) string {
return filepath.Join(tempFolder, fmt.Sprintf("%d.idx", vid)) return filepath.Join(tempFolder, fmt.Sprintf("%s_%d.idx", dataNodeid, vid))
} }
func getFilerFileIdFile(tempFolder string, vid uint32) string { func getFilerFileIdFile(tempFolder string, vid uint32) string {