JakeHillion/storj
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Keep some unhealthy pieces in the event of a partial repair (#2380)

Browse Source 
* if repair is partial, keep saving "unhealthy" pieces that are not duplicates
This commit is contained in:
Maximillian von Briesen 2019-06-28 15:48:51 -04:00 committed by GitHub
parent 7e0c5d51d0
commit b750bc2d0f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 22 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
pkg/storage/segments/repairer.go
View File

@ -67,7 +67,8 @@ func (repairer *Repairer) Repair(ctx context.Context, path storj.Path) (err erro
expiration := pointer.GetExpirationDate()
var excludeNodeIDs storj.NodeIDList
var healthyPieces []*pb.RemotePiece
var healthyPieces, unhealthyPieces []*pb.RemotePiece
healthyMap := make(map[int32]bool)
pieces := pointer.GetRemote().GetRemotePieces()
missingPieces, err := repairer.cache.GetMissingPieces(ctx, pieces)
if err != nil {
@ -100,6 +101,9 @@ func (repairer *Repairer) Repair(ctx context.Context, path storj.Path) (err erro
excludeNodeIDs = append(excludeNodeIDs, piece.NodeId)
if _, ok := lostPiecesSet[piece.GetPieceNum()]; !ok {
healthyPieces = append(healthyPieces, piece)
healthyMap[piece.GetPieceNum()] = true
} else {
unhealthyPieces = append(unhealthyPieces, piece)
}
}
@ -160,16 +164,14 @@ func (repairer *Repairer) Repair(ctx context.Context, path storj.Path) (err erro
NodeId: node.Id,
Hash: hashes[i],
})
healthyMap[int32(i)] = true
}
// Update the remote pieces in the pointer
pointer.GetRemote().RemotePieces = healthyPieces
length := int32(len(healthyPieces))
healthyLength := int32(len(healthyPieces))
switch {
case length <= pointer.Remote.Redundancy.RepairThreshold:
case healthyLength <= pointer.Remote.Redundancy.RepairThreshold:
mon.Meter("repair_failed").Mark(1)
case length < pointer.Remote.Redundancy.SuccessThreshold:
case healthyLength < pointer.Remote.Redundancy.SuccessThreshold:
mon.Meter("repair_partial").Mark(1)
default:
mon.Meter("repair_success").Mark(1)
@ -177,10 +179,22 @@ func (repairer *Repairer) Repair(ctx context.Context, path storj.Path) (err erro
healthyRatioAfterRepair := 0.0
if pointer.Remote.Redundancy.Total != 0 {
healthyRatioAfterRepair = float64(len(healthyPieces)) / float64(pointer.Remote.Redundancy.Total)
healthyRatioAfterRepair = float64(healthyLength) / float64(pointer.Remote.Redundancy.Total)
}
mon.FloatVal("healthy_ratio_after_repair").Observe(healthyRatioAfterRepair)
// if partial repair, include "unhealthy" pieces that are not duplicates
if healthyLength < pointer.Remote.Redundancy.SuccessThreshold {
for _, p := range unhealthyPieces {
if _, ok := healthyMap[p.GetPieceNum()]; !ok {
healthyPieces = append(healthyPieces, p)
}
}
}
// Update the remote pieces in the pointer
pointer.GetRemote().RemotePieces = healthyPieces
// Update the segment pointer in the metainfo
return repairer.metainfo.Put(ctx, path, pointer)
}