feat(balance): replica placement validation for volume moves (#8622)
* feat(balance): add replica placement validation for volume moves When the volume balance detection proposes moving a volume, validate that the move does not violate the volume's replication policy (e.g., ReplicaPlacement=010 requires replicas on different racks). If the preferred destination violates the policy, fall back to score-based planning; if that also violates, skip the volume entirely. - Add ReplicaLocation type and VolumeReplicaMap to ClusterInfo - Build replica map from all volumes before collection filtering - Port placement validation logic from command_volume_fix_replication.go - Thread replica map through collectVolumeMetrics call chain - Add IsGoodMove check in createBalanceTask before destination use * address PR review: extract validation closure, add defensive checks - Extract validateMove closure to eliminate duplicated ReplicaLocation construction and IsGoodMove calls - Add defensive check for empty replica map entries (len(replicas) == 0) - Add bounds check for int-to-byte cast on ExpectedReplicas (0-255) * address nitpick: rp test helper accepts *testing.T and fails on error Prevents silent failures from typos in replica placement codes. * address review: add composite replica placement tests (011, 110) Test multi-constraint placement policies where both rack and DC rules must be satisfied simultaneously. * address review: use struct keys instead of string concatenation Replace string-concatenated map keys with typed rackKey/nodeKey structs to eliminate allocations and avoid ambiguity if IDs contain spaces. * address review: simplify bounds check, log fallback error, guard source - Remove unreachable ExpectedReplicas < 0 branch (outer condition already guarantees > 0), fold bounds check into single condition - Log error from planBalanceDestination in replica validation fallback - Return false from IsGoodMove when sourceNodeID not found in existing replicas (inconsistent cluster state) * address review: use slices.Contains instead of hand-rolled helpers Replace isAmongDC and isAmongRack with slices.Contains from the standard library, reducing boilerplate.
This commit is contained in:
Chris Lu
@@ -10,6 +10,7 @@ import (
|
||||
"github.com/seaweedfs/seaweedfs/weed/admin/topology"
|
||||
"github.com/seaweedfs/seaweedfs/weed/glog"
|
||||
"github.com/seaweedfs/seaweedfs/weed/pb/worker_pb"
|
||||
"github.com/seaweedfs/seaweedfs/weed/storage/super_block"
|
||||
"github.com/seaweedfs/seaweedfs/weed/worker/tasks/base"
|
||||
"github.com/seaweedfs/seaweedfs/weed/worker/tasks/util"
|
||||
"github.com/seaweedfs/seaweedfs/weed/worker/types"
|
||||
@@ -389,6 +390,46 @@ func createBalanceTask(diskType string, selectedVolume *types.VolumeHealthMetric
|
||||
}
|
||||
}
|
||||
|
||||
// Validate move against replica placement policy
|
||||
if selectedVolume.ExpectedReplicas > 0 && selectedVolume.ExpectedReplicas <= 255 && clusterInfo.VolumeReplicaMap != nil {
|
||||
rpBytes, rpErr := super_block.NewReplicaPlacementFromByte(byte(selectedVolume.ExpectedReplicas))
|
||||
if rpErr == nil && rpBytes.HasReplication() {
|
||||
replicas := clusterInfo.VolumeReplicaMap[selectedVolume.VolumeID]
|
||||
if len(replicas) == 0 {
|
||||
glog.V(1).Infof("BALANCE [%s]: No replica locations found for volume %d, skipping placement validation",
|
||||
diskType, selectedVolume.VolumeID)
|
||||
} else {
|
||||
validateMove := func(plan *topology.DestinationPlan) bool {
|
||||
if plan == nil {
|
||||
return false
|
||||
}
|
||||
target := types.ReplicaLocation{
|
||||
DataCenter: plan.TargetDC,
|
||||
Rack: plan.TargetRack,
|
||||
NodeID: plan.TargetNode,
|
||||
}
|
||||
return IsGoodMove(rpBytes, replicas, selectedVolume.Server, target)
|
||||
}
|
||||
|
||||
if !validateMove(destinationPlan) {
|
||||
glog.V(1).Infof("BALANCE [%s]: Destination %s violates replica placement for volume %d (rp=%03d), falling back",
|
||||
diskType, destinationPlan.TargetNode, selectedVolume.VolumeID, selectedVolume.ExpectedReplicas)
|
||||
// Fall back to score-based planning
|
||||
destinationPlan, err = planBalanceDestination(clusterInfo.ActiveTopology, selectedVolume)
|
||||
if err != nil {
|
||||
glog.Warningf("BALANCE [%s]: Failed to plan fallback destination for volume %d: %v", diskType, selectedVolume.VolumeID, err)
|
||||
return nil, ""
|
||||
}
|
||||
if !validateMove(destinationPlan) {
|
||||
glog.V(1).Infof("BALANCE [%s]: Fallback destination %s also violates replica placement for volume %d",
|
||||
diskType, destinationPlan.TargetNode, selectedVolume.VolumeID)
|
||||
return nil, ""
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Find the actual disk containing the volume on the source server
|
||||
sourceDisk, found := base.FindVolumeDisk(clusterInfo.ActiveTopology, selectedVolume.VolumeID, selectedVolume.Collection, selectedVolume.Server)
|
||||
if !found {
|
||||
|
||||
146
weed/worker/tasks/balance/replica_placement.go
Normal file
146
weed/worker/tasks/balance/replica_placement.go
Normal file
@@ -0,0 +1,146 @@
|
||||
package balance
|
||||
|
||||
import (
|
||||
"slices"
|
||||
|
||||
"github.com/seaweedfs/seaweedfs/weed/storage/super_block"
|
||||
"github.com/seaweedfs/seaweedfs/weed/worker/types"
|
||||
)
|
||||
|
||||
// rackKey uniquely identifies a rack within a data center.
|
||||
type rackKey struct {
|
||||
DataCenter string
|
||||
Rack string
|
||||
}
|
||||
|
||||
// nodeKey uniquely identifies a node within a rack.
|
||||
type nodeKey struct {
|
||||
DataCenter string
|
||||
Rack string
|
||||
NodeID string
|
||||
}
|
||||
|
||||
// IsGoodMove checks whether moving a volume from sourceNodeID to target
|
||||
// would satisfy the volume's replica placement policy, given the current
|
||||
// set of replica locations.
|
||||
func IsGoodMove(rp *super_block.ReplicaPlacement, existingReplicas []types.ReplicaLocation, sourceNodeID string, target types.ReplicaLocation) bool {
|
||||
if rp == nil || !rp.HasReplication() {
|
||||
return true // no replication constraint
|
||||
}
|
||||
|
||||
// Build the replica set after the move: remove source, add target
|
||||
afterMove := make([]types.ReplicaLocation, 0, len(existingReplicas))
|
||||
sourceFound := false
|
||||
for _, r := range existingReplicas {
|
||||
if r.NodeID == sourceNodeID {
|
||||
sourceFound = true
|
||||
} else {
|
||||
afterMove = append(afterMove, r)
|
||||
}
|
||||
}
|
||||
if !sourceFound {
|
||||
// Source not in replica list — cluster state may be inconsistent.
|
||||
// Treat as unsafe to avoid incorrect placement decisions.
|
||||
return false
|
||||
}
|
||||
|
||||
return satisfyReplicaPlacement(rp, afterMove, target)
|
||||
}
|
||||
|
||||
// satisfyReplicaPlacement checks whether placing a replica at target
|
||||
// is consistent with the replication policy, given the existing replicas.
|
||||
// Ported from weed/shell/command_volume_fix_replication.go
|
||||
func satisfyReplicaPlacement(rp *super_block.ReplicaPlacement, replicas []types.ReplicaLocation, target types.ReplicaLocation) bool {
|
||||
existingDCs, _, existingNodes := countReplicas(replicas)
|
||||
|
||||
targetNK := nodeKey{DataCenter: target.DataCenter, Rack: target.Rack, NodeID: target.NodeID}
|
||||
if _, found := existingNodes[targetNK]; found {
|
||||
// avoid duplicated volume on the same data node
|
||||
return false
|
||||
}
|
||||
|
||||
primaryDCs, _ := findTopDCKeys(existingDCs)
|
||||
|
||||
// ensure data center count is within limit
|
||||
if _, found := existingDCs[target.DataCenter]; !found {
|
||||
// different from existing dcs
|
||||
if len(existingDCs) < rp.DiffDataCenterCount+1 {
|
||||
return true
|
||||
}
|
||||
return false
|
||||
}
|
||||
// now same as one of existing data centers
|
||||
if !slices.Contains(primaryDCs, target.DataCenter) {
|
||||
return false
|
||||
}
|
||||
|
||||
// now on a primary dc - check racks within this DC
|
||||
primaryDcRacks := make(map[rackKey]int)
|
||||
for _, r := range replicas {
|
||||
if r.DataCenter != target.DataCenter {
|
||||
continue
|
||||
}
|
||||
primaryDcRacks[rackKey{DataCenter: r.DataCenter, Rack: r.Rack}]++
|
||||
}
|
||||
|
||||
targetRK := rackKey{DataCenter: target.DataCenter, Rack: target.Rack}
|
||||
primaryRacks, _ := findTopRackKeys(primaryDcRacks)
|
||||
sameRackCount := primaryDcRacks[targetRK]
|
||||
|
||||
if _, found := primaryDcRacks[targetRK]; !found {
|
||||
// different from existing racks
|
||||
if len(primaryDcRacks) < rp.DiffRackCount+1 {
|
||||
return true
|
||||
}
|
||||
return false
|
||||
}
|
||||
// same as one of existing racks
|
||||
if !slices.Contains(primaryRacks, targetRK) {
|
||||
return false
|
||||
}
|
||||
|
||||
// on primary rack - check same-rack count
|
||||
if sameRackCount < rp.SameRackCount+1 {
|
||||
return true
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
func countReplicas(replicas []types.ReplicaLocation) (dcCounts map[string]int, rackCounts map[rackKey]int, nodeCounts map[nodeKey]int) {
|
||||
dcCounts = make(map[string]int)
|
||||
rackCounts = make(map[rackKey]int)
|
||||
nodeCounts = make(map[nodeKey]int)
|
||||
for _, r := range replicas {
|
||||
dcCounts[r.DataCenter]++
|
||||
rackCounts[rackKey{DataCenter: r.DataCenter, Rack: r.Rack}]++
|
||||
nodeCounts[nodeKey{DataCenter: r.DataCenter, Rack: r.Rack, NodeID: r.NodeID}]++
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
func findTopDCKeys(m map[string]int) (topKeys []string, max int) {
|
||||
for k, c := range m {
|
||||
if max < c {
|
||||
topKeys = topKeys[:0]
|
||||
topKeys = append(topKeys, k)
|
||||
max = c
|
||||
} else if max == c {
|
||||
topKeys = append(topKeys, k)
|
||||
}
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
func findTopRackKeys(m map[rackKey]int) (topKeys []rackKey, max int) {
|
||||
for k, c := range m {
|
||||
if max < c {
|
||||
topKeys = topKeys[:0]
|
||||
topKeys = append(topKeys, k)
|
||||
max = c
|
||||
} else if max == c {
|
||||
topKeys = append(topKeys, k)
|
||||
}
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
127
weed/worker/tasks/balance/replica_placement_test.go
Normal file
127
weed/worker/tasks/balance/replica_placement_test.go
Normal file
@@ -0,0 +1,127 @@
|
||||
package balance
|
||||
|
||||
import (
|
||||
"testing"
|
||||
|
||||
"github.com/seaweedfs/seaweedfs/weed/storage/super_block"
|
||||
"github.com/seaweedfs/seaweedfs/weed/worker/types"
|
||||
)
|
||||
|
||||
func rp(t *testing.T, code string) *super_block.ReplicaPlacement {
|
||||
t.Helper()
|
||||
r, err := super_block.NewReplicaPlacementFromString(code)
|
||||
if err != nil {
|
||||
t.Fatalf("invalid replica placement code %q: %v", code, err)
|
||||
}
|
||||
return r
|
||||
}
|
||||
|
||||
func loc(dc, rack, node string) types.ReplicaLocation {
|
||||
return types.ReplicaLocation{DataCenter: dc, Rack: rack, NodeID: node}
|
||||
}
|
||||
|
||||
func TestIsGoodMove_NoReplication(t *testing.T) {
|
||||
// 000 = no replication. Any move is fine.
|
||||
if !IsGoodMove(rp(t, "000"), []types.ReplicaLocation{loc("dc1", "r1", "n1")}, "n1", loc("dc1", "r1", "n2")) {
|
||||
t.Error("000: any move should be allowed")
|
||||
}
|
||||
}
|
||||
|
||||
func TestIsGoodMove_001_SameRack(t *testing.T) {
|
||||
// 001 = 1 replica on same rack (2 total on same rack)
|
||||
existing := []types.ReplicaLocation{
|
||||
loc("dc1", "r1", "n1"),
|
||||
loc("dc1", "r1", "n2"),
|
||||
}
|
||||
// Move n1 -> n3 on same rack: good
|
||||
if !IsGoodMove(rp(t, "001"), existing, "n1", loc("dc1", "r1", "n3")) {
|
||||
t.Error("001: move to same rack should be allowed")
|
||||
}
|
||||
// Move n1 -> n3 on different rack: bad (would leave only 1 on r1, need 2)
|
||||
if IsGoodMove(rp(t, "001"), existing, "n1", loc("dc1", "r2", "n3")) {
|
||||
t.Error("001: move to different rack should not be allowed when it breaks same-rack count")
|
||||
}
|
||||
}
|
||||
|
||||
func TestIsGoodMove_010_DiffRack(t *testing.T) {
|
||||
// 010 = 1 replica on different rack (2 racks total)
|
||||
existing := []types.ReplicaLocation{
|
||||
loc("dc1", "r1", "n1"),
|
||||
loc("dc1", "r2", "n2"),
|
||||
}
|
||||
// Move n1 -> n3 on r2: bad (both replicas on same rack)
|
||||
if IsGoodMove(rp(t, "010"), existing, "n1", loc("dc1", "r2", "n3")) {
|
||||
t.Error("010: move to same rack as other replica should not be allowed")
|
||||
}
|
||||
// Move n1 -> n3 on r3: good (still 2 different racks)
|
||||
if !IsGoodMove(rp(t, "010"), existing, "n1", loc("dc1", "r3", "n3")) {
|
||||
t.Error("010: move to different rack should be allowed")
|
||||
}
|
||||
}
|
||||
|
||||
func TestIsGoodMove_100_DiffDC(t *testing.T) {
|
||||
// 100 = 1 replica in different DC
|
||||
existing := []types.ReplicaLocation{
|
||||
loc("dc1", "r1", "n1"),
|
||||
loc("dc2", "r1", "n2"),
|
||||
}
|
||||
// Move n1 -> n3 in dc2: bad (both in same DC)
|
||||
if IsGoodMove(rp(t, "100"), existing, "n1", loc("dc2", "r1", "n3")) {
|
||||
t.Error("100: move to same DC as other replica should not be allowed")
|
||||
}
|
||||
// Move n1 -> n3 in dc3: good (different DCs)
|
||||
if !IsGoodMove(rp(t, "100"), existing, "n1", loc("dc3", "r1", "n3")) {
|
||||
t.Error("100: move to different DC should be allowed")
|
||||
}
|
||||
}
|
||||
|
||||
func TestIsGoodMove_SameNode(t *testing.T) {
|
||||
// Moving to the same node as an existing replica should always be rejected
|
||||
existing := []types.ReplicaLocation{
|
||||
loc("dc1", "r1", "n1"),
|
||||
loc("dc1", "r2", "n2"),
|
||||
}
|
||||
if IsGoodMove(rp(t, "010"), existing, "n1", loc("dc1", "r2", "n2")) {
|
||||
t.Error("should reject move to same node as existing replica")
|
||||
}
|
||||
}
|
||||
|
||||
func TestIsGoodMove_011_Composite(t *testing.T) {
|
||||
// 011 = 1 same-rack + 1 different-rack (3 replicas: 2 on same rack, 1 on different)
|
||||
existing := []types.ReplicaLocation{
|
||||
loc("dc1", "r1", "n1"),
|
||||
loc("dc1", "r1", "n2"),
|
||||
loc("dc1", "r2", "n3"),
|
||||
}
|
||||
// Move n1 -> n4 on r1: good (maintains 2 on r1, 1 on r2)
|
||||
if !IsGoodMove(rp(t, "011"), existing, "n1", loc("dc1", "r1", "n4")) {
|
||||
t.Error("011: move within same rack should be allowed")
|
||||
}
|
||||
// Move n3 -> n4 on r1: bad (would have 3 on r1, 0 on different rack)
|
||||
if IsGoodMove(rp(t, "011"), existing, "n3", loc("dc1", "r1", "n4")) {
|
||||
t.Error("011: move that eliminates different-rack replica should not be allowed")
|
||||
}
|
||||
}
|
||||
|
||||
func TestIsGoodMove_110_Composite(t *testing.T) {
|
||||
// 110 = 1 different-rack + 1 different-DC (3 replicas across 2 DCs and 2 racks)
|
||||
existing := []types.ReplicaLocation{
|
||||
loc("dc1", "r1", "n1"),
|
||||
loc("dc1", "r2", "n2"),
|
||||
loc("dc2", "r1", "n3"),
|
||||
}
|
||||
// Move n1 -> n4 in dc1/r3: good (dc1 still has r2+r3, dc2 has r1)
|
||||
if !IsGoodMove(rp(t, "110"), existing, "n1", loc("dc1", "r3", "n4")) {
|
||||
t.Error("110: move to new rack in same DC should be allowed")
|
||||
}
|
||||
// Move n3 -> n4 in dc1/r1: bad (would lose the different-DC replica)
|
||||
if IsGoodMove(rp(t, "110"), existing, "n3", loc("dc1", "r1", "n4")) {
|
||||
t.Error("110: move that eliminates different-DC replica should not be allowed")
|
||||
}
|
||||
}
|
||||
|
||||
func TestIsGoodMove_NilReplicaPlacement(t *testing.T) {
|
||||
if !IsGoodMove(nil, []types.ReplicaLocation{loc("dc1", "r1", "n1")}, "n1", loc("dc1", "r1", "n2")) {
|
||||
t.Error("nil replica placement should allow any move")
|
||||
}
|
||||
}
|
||||
@@ -6,13 +6,21 @@ import (
|
||||
"github.com/seaweedfs/seaweedfs/weed/admin/topology"
|
||||
)
|
||||
|
||||
// ReplicaLocation identifies where a volume replica lives.
|
||||
type ReplicaLocation struct {
|
||||
DataCenter string
|
||||
Rack string
|
||||
NodeID string
|
||||
}
|
||||
|
||||
// ClusterInfo contains cluster information for task detection
|
||||
type ClusterInfo struct {
|
||||
Servers []*VolumeServerInfo
|
||||
TotalVolumes int
|
||||
TotalServers int
|
||||
LastUpdated time.Time
|
||||
ActiveTopology *topology.ActiveTopology // Added for destination planning in detection
|
||||
Servers []*VolumeServerInfo
|
||||
TotalVolumes int
|
||||
TotalServers int
|
||||
LastUpdated time.Time
|
||||
ActiveTopology *topology.ActiveTopology // Added for destination planning in detection
|
||||
VolumeReplicaMap map[uint32][]ReplicaLocation
|
||||
}
|
||||
|
||||
// VolumeHealthMetrics contains health information about a volume (simplified)
|
||||
|
||||
Reference in New Issue
Block a user