seaweedfs/weed/topology/replication_health_checker.go
2019-04-10 19:41:55 +08:00

298 lines
9 KiB
Go

package topology
import (
"context"
"fmt"
"github.com/chrislusf/seaweedfs/weed/glog"
"github.com/chrislusf/seaweedfs/weed/operation"
"github.com/chrislusf/seaweedfs/weed/pb/volume_server_pb"
"github.com/chrislusf/seaweedfs/weed/storage"
"google.golang.org/grpc"
"sort"
"strings"
"sync"
)
/**
check the replication health
*/
func (t *Topology) RepairUnhealthyReplicationInLayout(grpcDialOption grpc.DialOption, layout *VolumeLayout, eVid storage.VolumeId) error {
ctx := context.Background()
locations, exist := layout.vid2location[eVid]
if !exist {
retErr := fmt.Errorf("the volume:%v has no locations", eVid)
glog.V(0).Infof(retErr.Error())
return retErr
}
//glog.V(5).Infof("volume:%v, locations:%v", eVid, locations.list)
fileStat, err := getReplicationInfo(grpcDialOption, ctx, eVid, locations)
if err != nil {
glog.Errorf("get replication status failed, %v", err)
return err
}
if isSameVolumeReplications(fileStat, layout.volumeSizeLimit) {
glog.V(0).Infof("the volume:%v has %d same replication, need not repair", eVid, len(fileStat))
return nil
}
// compact all the replications of volume
{
glog.V(4).Infof("begin compact all the replications of volume:%v", eVid)
allUrls := make([]string, 0, len(fileStat))
for _, fs := range fileStat {
allUrls = append(allUrls, fs.url)
}
if tryBatchCompactVolume(ctx, grpcDialOption, eVid, allUrls) == false {
err := fmt.Errorf("compact all the replications of volume:%v", eVid)
glog.Error(err.Error())
return err
}
glog.V(4).Infof("success compact all the replications of volume:%v", eVid)
}
// get replication status again
fileStat, err = getReplicationInfo(grpcDialOption, ctx, eVid, locations)
if err != nil {
return err
}
okUrls, errUrls := filterErrorReplication(fileStat)
if len(errUrls) == 0 {
return nil // they are the same
}
if len(okUrls) == 0 {
return fmt.Errorf("no correct volume replications, that's impossible")
}
glog.V(4).Infof("need repair replication : %v", errUrls)
if len(locations.list) <= 0 {
return fmt.Errorf("that's impossible, the locatins of volume:%v is empty", eVid)
}
for _, url := range errUrls {
vInfo := locations.list[0].volumes[eVid]
err = syncReplication(grpcDialOption, okUrls[0], url, vInfo)
if nil != err {
glog.Error(err)
return err
}
}
return nil
}
type FileStatus struct {
url string
fileStat *volume_server_pb.ReadVolumeFileStatusResponse
}
func getReplicationInfo(grpcDialOption grpc.DialOption, ctx context.Context, vid storage.VolumeId, locs *VolumeLocationList) (fs []FileStatus, err error) {
type ResponsePair struct {
url string
status *volume_server_pb.ReadVolumeFileStatusResponse
err error
}
var wg sync.WaitGroup
resultChan := make(chan ResponsePair, len(locs.list))
wg.Add(len(locs.list))
getFileStatFunc := func(url string, volumeId storage.VolumeId) {
defer wg.Done()
glog.V(4).Infof("volumeId:%v, location:%v", volumeId, url)
err := operation.WithVolumeServerClient(url, grpcDialOption, func(client volume_server_pb.VolumeServerClient) error {
req := &volume_server_pb.ReadVolumeFileStatusRequest{
VolumeId: uint32(volumeId),
}
respTmp, err := client.ReadVolumeFileStatus(ctx, req)
resultChan <- ResponsePair{
url: url,
status: respTmp,
err: err,
}
return nil
})
if nil != err {
glog.Error(err)
}
}
for _, node := range locs.list {
go getFileStatFunc(node.Url(), vid)
}
go func() { // close channel
wg.Wait()
close(resultChan)
}()
var errs []string
for result := range resultChan {
if result.err == nil {
fs = append(fs, FileStatus{
url: result.url,
fileStat: result.status,
})
continue
}
tmp := fmt.Sprintf("url : %s, error : %v", result.url, result.err)
errs = append(errs, tmp)
}
if len(fs) == len(locs.list) {
return fs, nil
}
err = fmt.Errorf("get volume[%v] replication status failed, err : %s", vid, strings.Join(errs, "; "))
return nil, err
}
/**
<see the class mapMetric and needleMap> :
the file count is the total count of the volume received from user clients
todo: this policy is not perfected or not rigorous, need fix
*/
func filterErrorReplication(fileStat []FileStatus) (okUrls, errUrls []string) {
sort.Slice(fileStat, func(i, j int) bool {
return fileStat[i].fileStat.FileCount > fileStat[j].fileStat.FileCount
})
if fileStat[0].fileStat.FileCount != fileStat[len(fileStat)-1].fileStat.FileCount {
okFileCounter := fileStat[0].fileStat.FileCount
for _, v := range fileStat {
if okFileCounter == v.fileStat.FileCount {
okUrls = append(okUrls, v.url)
} else {
errUrls = append(errUrls, v.url)
}
}
return
}
return
}
// execute the compact transaction
func compactVolume(ctx context.Context, grpcDialOption grpc.DialOption, volumeUrl string, vid storage.VolumeId) bool {
glog.V(0).Infoln("Start vacuuming", vid, "on", volumeUrl)
err := operation.WithVolumeServerClient(volumeUrl, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
_, err := volumeServerClient.VacuumVolumeCompact(ctx, &volume_server_pb.VacuumVolumeCompactRequest{
VolumeId: uint32(vid),
})
return err
})
if err != nil {
glog.Errorf("Error when vacuuming %d on %s: %v", vid, volumeUrl, err)
return false
}
glog.V(0).Infof("Complete vacuuming volume:%v on %s", vid, volumeUrl)
return true
}
// commit the compact transaction when compactVolume() return true
func commitCompactedVolume(ctx context.Context, grpcDialOption grpc.DialOption, volumeUrl string, vid storage.VolumeId) bool {
err := operation.WithVolumeServerClient(volumeUrl, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
_, err := volumeServerClient.VacuumVolumeCommit(ctx, &volume_server_pb.VacuumVolumeCommitRequest{
VolumeId: uint32(vid),
})
return err
})
if err != nil {
glog.Errorf("Error when committing vacuum %d on %s: %v", vid, volumeUrl, err)
return false
}
glog.V(0).Infof("Complete Committing vacuum %d on %s", vid, volumeUrl)
return true
}
// rollback the compact transaction when compactVolume return false
func cleanupCompactedVolume(ctx context.Context, grpcDialOption grpc.DialOption, volumeUrl string, vid storage.VolumeId) bool {
glog.V(0).Infoln("Start cleaning up", vid, "on", volumeUrl)
err := operation.WithVolumeServerClient(volumeUrl, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
_, err := volumeServerClient.VacuumVolumeCleanup(ctx, &volume_server_pb.VacuumVolumeCleanupRequest{
VolumeId: uint32(vid),
})
return err
})
if err != nil {
glog.Errorf("Error when cleaning up vacuum %d on %s: %v", vid, volumeUrl, err)
return false
}
glog.V(0).Infof("Complete cleaning up vacuum %d on %s", vid, volumeUrl)
return false
}
func tryCompactVolume(ctx context.Context, grpcDialOption grpc.DialOption, vid storage.VolumeId, volumeUrl string) bool {
if compactVolume(ctx, grpcDialOption, volumeUrl, vid) == false {
return cleanupCompactedVolume(ctx, grpcDialOption, volumeUrl, vid)
}
return commitCompactedVolume(ctx, grpcDialOption, volumeUrl, vid)
}
func tryBatchCompactVolume(ctx context.Context, grpcDialOption grpc.DialOption,
vid storage.VolumeId, urls []string) bool {
resultChan := make(chan error)
var wg sync.WaitGroup
wg.Add(len(urls))
for _, url := range urls {
go func(volumeUrl string) {
defer wg.Done()
if tryCompactVolume(ctx, grpcDialOption, vid, volumeUrl) == false {
resultChan <- fmt.Errorf("url:%s", volumeUrl)
}
}(url)
}
go func() {
wg.Wait()
close(resultChan)
}()
var errs []string
for result := range resultChan {
if result != nil {
errs = append(errs, result.Error())
}
}
if len(errs) > 0 {
glog.Errorf("consist volume:%v compact reversion failed, %s", vid, strings.Join(errs, "; "))
return false
}
return true
}
func isSameVolumeReplications(fileStat []FileStatus, volumeSizeLimit uint64) bool {
fileSizeSet := make(map[uint64]bool)
fileCountSet := make(map[uint64]bool)
lastModifiedSet := make(map[uint64]bool)
var oneFileSize uint64 = 0
for _, v := range fileStat {
fileCountSet[v.fileStat.FileCount] = true
lastModifiedSet[v.fileStat.DatFileTimestamp] = true
fileSizeSet[v.fileStat.DatFileSize] = true
oneFileSize = v.fileStat.DatFileSize
}
if (len(lastModifiedSet) == 1) && (len(fileCountSet) == 1) &&
(len(fileSizeSet) == 1) && (oneFileSize >= volumeSizeLimit) {
return true
}
return false
}
func syncReplication(grpcDialOption grpc.DialOption, srcUrl, destUrl string, vinfo storage.VolumeInfo) error {
ctx := context.Background()
err := operation.WithVolumeServerClient(destUrl, grpcDialOption,
func(client volume_server_pb.VolumeServerClient) error {
if _, err := client.ReplicateVolume(ctx, &volume_server_pb.ReplicateVolumeRequest{
VolumeId: uint32(vinfo.Id),
Collection: vinfo.Collection,
Replication: vinfo.ReplicaPlacement.String(),
Ttl: vinfo.Ttl.String(),
SourceDataNode: srcUrl,
}); err != nil {
glog.Errorf("sync replication failed, %v", err)
return err
}
return nil
})
return err
}