Add disk-aware EC rebalancing (#7597)
* Add placement package for EC shard placement logic - Consolidate EC shard placement algorithm for reuse across shell and worker tasks - Support multi-pass selection: racks, then servers, then disks - Include proper spread verification and scoring functions - Comprehensive test coverage for various cluster topologies * Make ec.balance disk-aware for multi-disk servers - Add EcDisk struct to track individual disks on volume servers - Update EcNode to maintain per-disk shard distribution - Parse disk_id from EC shard information during topology collection - Implement pickBestDiskOnNode() for selecting best disk per shard - Add diskDistributionScore() for tie-breaking node selection - Update all move operations to specify target disk in RPC calls - Improves shard balance within multi-disk servers, not just across servers * Use placement package in EC detection for consistent disk-level placement - Replace custom EC disk selection logic with shared placement package - Convert topology DiskInfo to placement.DiskCandidate format - Use SelectDestinations() for multi-rack/server/disk spreading - Convert placement results back to topology DiskInfo for task creation - Ensures EC detection uses same placement logic as shell commands * Make volume server evacuation disk-aware - Use pickBestDiskOnNode() when selecting evacuation target disk - Specify target disk in evacuation RPC requests - Maintains balanced disk distribution during server evacuations * Rename PlacementConfig to PlacementRequest for clarity PlacementRequest better reflects that this is a request for placement rather than a configuration object. This improves API semantics. * Rename DefaultConfig to DefaultPlacementRequest Aligns with the PlacementRequest type naming for consistency * Address review comments from Gemini and CodeRabbit Fix HIGH issues: - Fix empty disk discovery: Now discovers all disks from VolumeInfos, not just from EC shards. This ensures disks without EC shards are still considered for placement. - Fix EC shard count calculation in detection.go: Now correctly filters by DiskId and sums actual shard counts using ShardBits.ShardIdCount() instead of just counting EcShardInfo entries. Fix MEDIUM issues: - Add disk ID to evacuation log messages for consistency with other logging - Remove unused serverToDisks variable in placement.go - Fix comment that incorrectly said 'ascending' when sorting is 'descending' * add ec tests * Update ec-integration-tests.yml * Update ec_integration_test.go * Fix EC integration tests CI: build weed binary and update actions - Add 'Build weed binary' step before running tests - Update actions/setup-go from v4 to v6 (Node20 compatibility) - Update actions/checkout from v2 to v4 (Node20 compatibility) - Move working-directory to test step only * Add disk-aware EC rebalancing integration tests - Add TestDiskAwareECRebalancing test with multi-disk cluster setup - Test EC encode with disk awareness (shows disk ID in output) - Test EC balance with disk-level shard distribution - Add helper functions for disk-level verification: - startMultiDiskCluster: 3 servers x 4 disks each - countShardsPerDisk: track shards per disk per server - calculateDiskShardVariance: measure distribution balance - Verify no single disk is overloaded with shards
This commit is contained in:
Chris Lu
420
weed/storage/erasure_coding/placement/placement.go
Normal file
420
weed/storage/erasure_coding/placement/placement.go
Normal file
@@ -0,0 +1,420 @@
|
||||
// Package placement provides consolidated EC shard placement logic used by
|
||||
// both shell commands and worker tasks.
|
||||
//
|
||||
// This package encapsulates the algorithms for:
|
||||
// - Selecting destination nodes/disks for EC shards
|
||||
// - Ensuring proper spread across racks, servers, and disks
|
||||
// - Balancing shards across the cluster
|
||||
package placement
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"sort"
|
||||
)
|
||||
|
||||
// DiskCandidate represents a disk that can receive EC shards
|
||||
type DiskCandidate struct {
|
||||
NodeID string
|
||||
DiskID uint32
|
||||
DataCenter string
|
||||
Rack string
|
||||
|
||||
// Capacity information
|
||||
VolumeCount int64
|
||||
MaxVolumeCount int64
|
||||
ShardCount int // Current number of EC shards on this disk
|
||||
FreeSlots int // Available slots for new shards
|
||||
|
||||
// Load information
|
||||
LoadCount int // Number of active tasks on this disk
|
||||
}
|
||||
|
||||
// NodeCandidate represents a server node that can receive EC shards
|
||||
type NodeCandidate struct {
|
||||
NodeID string
|
||||
DataCenter string
|
||||
Rack string
|
||||
FreeSlots int
|
||||
ShardCount int // Total shards across all disks
|
||||
Disks []*DiskCandidate // All disks on this node
|
||||
}
|
||||
|
||||
// PlacementRequest configures EC shard placement behavior
|
||||
type PlacementRequest struct {
|
||||
// ShardsNeeded is the total number of shards to place
|
||||
ShardsNeeded int
|
||||
|
||||
// MaxShardsPerServer limits how many shards can be placed on a single server
|
||||
// 0 means no limit (but prefer spreading when possible)
|
||||
MaxShardsPerServer int
|
||||
|
||||
// MaxShardsPerRack limits how many shards can be placed in a single rack
|
||||
// 0 means no limit
|
||||
MaxShardsPerRack int
|
||||
|
||||
// MaxTaskLoad is the maximum task load count for a disk to be considered
|
||||
MaxTaskLoad int
|
||||
|
||||
// PreferDifferentServers when true, spreads shards across different servers
|
||||
// before using multiple disks on the same server
|
||||
PreferDifferentServers bool
|
||||
|
||||
// PreferDifferentRacks when true, spreads shards across different racks
|
||||
// before using multiple servers in the same rack
|
||||
PreferDifferentRacks bool
|
||||
}
|
||||
|
||||
// DefaultPlacementRequest returns the default placement configuration
|
||||
func DefaultPlacementRequest() PlacementRequest {
|
||||
return PlacementRequest{
|
||||
ShardsNeeded: 14,
|
||||
MaxShardsPerServer: 0,
|
||||
MaxShardsPerRack: 0,
|
||||
MaxTaskLoad: 5,
|
||||
PreferDifferentServers: true,
|
||||
PreferDifferentRacks: true,
|
||||
}
|
||||
}
|
||||
|
||||
// PlacementResult contains the selected destinations for EC shards
|
||||
type PlacementResult struct {
|
||||
SelectedDisks []*DiskCandidate
|
||||
|
||||
// Statistics
|
||||
ServersUsed int
|
||||
RacksUsed int
|
||||
DCsUsed int
|
||||
|
||||
// Distribution maps
|
||||
ShardsPerServer map[string]int
|
||||
ShardsPerRack map[string]int
|
||||
ShardsPerDC map[string]int
|
||||
}
|
||||
|
||||
// SelectDestinations selects the best disks for EC shard placement.
|
||||
// This is the main entry point for EC placement logic.
|
||||
//
|
||||
// The algorithm works in multiple passes:
|
||||
// 1. First pass: Select one disk from each rack (maximize rack diversity)
|
||||
// 2. Second pass: Select one disk from each unused server in used racks (maximize server diversity)
|
||||
// 3. Third pass: Select additional disks from servers already used (maximize disk diversity)
|
||||
func SelectDestinations(disks []*DiskCandidate, config PlacementRequest) (*PlacementResult, error) {
|
||||
if len(disks) == 0 {
|
||||
return nil, fmt.Errorf("no disk candidates provided")
|
||||
}
|
||||
if config.ShardsNeeded <= 0 {
|
||||
return nil, fmt.Errorf("shardsNeeded must be positive, got %d", config.ShardsNeeded)
|
||||
}
|
||||
|
||||
// Filter suitable disks
|
||||
suitable := filterSuitableDisks(disks, config)
|
||||
if len(suitable) == 0 {
|
||||
return nil, fmt.Errorf("no suitable disks found after filtering")
|
||||
}
|
||||
|
||||
// Build indexes for efficient lookup
|
||||
rackToDisks := groupDisksByRack(suitable)
|
||||
|
||||
result := &PlacementResult{
|
||||
SelectedDisks: make([]*DiskCandidate, 0, config.ShardsNeeded),
|
||||
ShardsPerServer: make(map[string]int),
|
||||
ShardsPerRack: make(map[string]int),
|
||||
ShardsPerDC: make(map[string]int),
|
||||
}
|
||||
|
||||
usedDisks := make(map[string]bool) // "nodeID:diskID" -> bool
|
||||
usedServers := make(map[string]bool) // nodeID -> bool
|
||||
usedRacks := make(map[string]bool) // "dc:rack" -> bool
|
||||
|
||||
// Pass 1: Select one disk from each rack (maximize rack diversity)
|
||||
if config.PreferDifferentRacks {
|
||||
// Sort racks by number of available servers (descending) to prioritize racks with more options
|
||||
sortedRacks := sortRacksByServerCount(rackToDisks)
|
||||
for _, rackKey := range sortedRacks {
|
||||
if len(result.SelectedDisks) >= config.ShardsNeeded {
|
||||
break
|
||||
}
|
||||
rackDisks := rackToDisks[rackKey]
|
||||
// Select best disk from this rack, preferring a new server
|
||||
disk := selectBestDiskFromRack(rackDisks, usedServers, usedDisks, config)
|
||||
if disk != nil {
|
||||
addDiskToResult(result, disk, usedDisks, usedServers, usedRacks)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Pass 2: Select disks from unused servers in already-used racks
|
||||
if config.PreferDifferentServers && len(result.SelectedDisks) < config.ShardsNeeded {
|
||||
for _, rackKey := range getSortedRackKeys(rackToDisks) {
|
||||
if len(result.SelectedDisks) >= config.ShardsNeeded {
|
||||
break
|
||||
}
|
||||
rackDisks := rackToDisks[rackKey]
|
||||
for _, disk := range sortDisksByScore(rackDisks) {
|
||||
if len(result.SelectedDisks) >= config.ShardsNeeded {
|
||||
break
|
||||
}
|
||||
diskKey := getDiskKey(disk)
|
||||
if usedDisks[diskKey] {
|
||||
continue
|
||||
}
|
||||
// Skip if server already used (we want different servers in this pass)
|
||||
if usedServers[disk.NodeID] {
|
||||
continue
|
||||
}
|
||||
// Check server limit
|
||||
if config.MaxShardsPerServer > 0 && result.ShardsPerServer[disk.NodeID] >= config.MaxShardsPerServer {
|
||||
continue
|
||||
}
|
||||
// Check rack limit
|
||||
if config.MaxShardsPerRack > 0 && result.ShardsPerRack[getRackKey(disk)] >= config.MaxShardsPerRack {
|
||||
continue
|
||||
}
|
||||
addDiskToResult(result, disk, usedDisks, usedServers, usedRacks)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Pass 3: Fill remaining slots from already-used servers (different disks)
|
||||
// Use round-robin across servers to balance shards evenly
|
||||
if len(result.SelectedDisks) < config.ShardsNeeded {
|
||||
// Group remaining disks by server
|
||||
serverToRemainingDisks := make(map[string][]*DiskCandidate)
|
||||
for _, disk := range suitable {
|
||||
if !usedDisks[getDiskKey(disk)] {
|
||||
serverToRemainingDisks[disk.NodeID] = append(serverToRemainingDisks[disk.NodeID], disk)
|
||||
}
|
||||
}
|
||||
|
||||
// Sort each server's disks by score
|
||||
for serverID := range serverToRemainingDisks {
|
||||
serverToRemainingDisks[serverID] = sortDisksByScore(serverToRemainingDisks[serverID])
|
||||
}
|
||||
|
||||
// Round-robin: repeatedly select from the server with the fewest shards
|
||||
for len(result.SelectedDisks) < config.ShardsNeeded {
|
||||
// Find server with fewest shards that still has available disks
|
||||
var bestServer string
|
||||
minShards := -1
|
||||
for serverID, disks := range serverToRemainingDisks {
|
||||
if len(disks) == 0 {
|
||||
continue
|
||||
}
|
||||
// Check server limit
|
||||
if config.MaxShardsPerServer > 0 && result.ShardsPerServer[serverID] >= config.MaxShardsPerServer {
|
||||
continue
|
||||
}
|
||||
shardCount := result.ShardsPerServer[serverID]
|
||||
if minShards == -1 || shardCount < minShards {
|
||||
minShards = shardCount
|
||||
bestServer = serverID
|
||||
} else if shardCount == minShards && serverID < bestServer {
|
||||
// Tie-break by server name for determinism
|
||||
bestServer = serverID
|
||||
}
|
||||
}
|
||||
|
||||
if bestServer == "" {
|
||||
// No more servers with available disks
|
||||
break
|
||||
}
|
||||
|
||||
// Pop the best disk from this server
|
||||
disks := serverToRemainingDisks[bestServer]
|
||||
disk := disks[0]
|
||||
serverToRemainingDisks[bestServer] = disks[1:]
|
||||
|
||||
// Check rack limit
|
||||
if config.MaxShardsPerRack > 0 && result.ShardsPerRack[getRackKey(disk)] >= config.MaxShardsPerRack {
|
||||
continue
|
||||
}
|
||||
|
||||
addDiskToResult(result, disk, usedDisks, usedServers, usedRacks)
|
||||
}
|
||||
}
|
||||
|
||||
// Calculate final statistics
|
||||
result.ServersUsed = len(usedServers)
|
||||
result.RacksUsed = len(usedRacks)
|
||||
dcSet := make(map[string]bool)
|
||||
for _, disk := range result.SelectedDisks {
|
||||
dcSet[disk.DataCenter] = true
|
||||
}
|
||||
result.DCsUsed = len(dcSet)
|
||||
|
||||
return result, nil
|
||||
}
|
||||
|
||||
// filterSuitableDisks filters disks that are suitable for EC placement
|
||||
func filterSuitableDisks(disks []*DiskCandidate, config PlacementRequest) []*DiskCandidate {
|
||||
var suitable []*DiskCandidate
|
||||
for _, disk := range disks {
|
||||
if disk.FreeSlots <= 0 {
|
||||
continue
|
||||
}
|
||||
if config.MaxTaskLoad > 0 && disk.LoadCount > config.MaxTaskLoad {
|
||||
continue
|
||||
}
|
||||
suitable = append(suitable, disk)
|
||||
}
|
||||
return suitable
|
||||
}
|
||||
|
||||
// groupDisksByRack groups disks by their rack (dc:rack key)
|
||||
func groupDisksByRack(disks []*DiskCandidate) map[string][]*DiskCandidate {
|
||||
result := make(map[string][]*DiskCandidate)
|
||||
for _, disk := range disks {
|
||||
key := getRackKey(disk)
|
||||
result[key] = append(result[key], disk)
|
||||
}
|
||||
return result
|
||||
}
|
||||
|
||||
// groupDisksByServer groups disks by their server
|
||||
func groupDisksByServer(disks []*DiskCandidate) map[string][]*DiskCandidate {
|
||||
result := make(map[string][]*DiskCandidate)
|
||||
for _, disk := range disks {
|
||||
result[disk.NodeID] = append(result[disk.NodeID], disk)
|
||||
}
|
||||
return result
|
||||
}
|
||||
|
||||
// getRackKey returns the unique key for a rack (dc:rack)
|
||||
func getRackKey(disk *DiskCandidate) string {
|
||||
return fmt.Sprintf("%s:%s", disk.DataCenter, disk.Rack)
|
||||
}
|
||||
|
||||
// getDiskKey returns the unique key for a disk (nodeID:diskID)
|
||||
func getDiskKey(disk *DiskCandidate) string {
|
||||
return fmt.Sprintf("%s:%d", disk.NodeID, disk.DiskID)
|
||||
}
|
||||
|
||||
// sortRacksByServerCount returns rack keys sorted by number of servers (ascending)
|
||||
func sortRacksByServerCount(rackToDisks map[string][]*DiskCandidate) []string {
|
||||
// Count unique servers per rack
|
||||
rackServerCount := make(map[string]int)
|
||||
for rackKey, disks := range rackToDisks {
|
||||
servers := make(map[string]bool)
|
||||
for _, disk := range disks {
|
||||
servers[disk.NodeID] = true
|
||||
}
|
||||
rackServerCount[rackKey] = len(servers)
|
||||
}
|
||||
|
||||
keys := getSortedRackKeys(rackToDisks)
|
||||
sort.Slice(keys, func(i, j int) bool {
|
||||
// Sort by server count (descending) to pick from racks with more options first
|
||||
return rackServerCount[keys[i]] > rackServerCount[keys[j]]
|
||||
})
|
||||
return keys
|
||||
}
|
||||
|
||||
// getSortedRackKeys returns rack keys in a deterministic order
|
||||
func getSortedRackKeys(rackToDisks map[string][]*DiskCandidate) []string {
|
||||
keys := make([]string, 0, len(rackToDisks))
|
||||
for k := range rackToDisks {
|
||||
keys = append(keys, k)
|
||||
}
|
||||
sort.Strings(keys)
|
||||
return keys
|
||||
}
|
||||
|
||||
// selectBestDiskFromRack selects the best disk from a rack for EC placement
|
||||
// It prefers servers that haven't been used yet
|
||||
func selectBestDiskFromRack(disks []*DiskCandidate, usedServers, usedDisks map[string]bool, config PlacementRequest) *DiskCandidate {
|
||||
var bestDisk *DiskCandidate
|
||||
bestScore := -1.0
|
||||
bestIsFromUnusedServer := false
|
||||
|
||||
for _, disk := range disks {
|
||||
if usedDisks[getDiskKey(disk)] {
|
||||
continue
|
||||
}
|
||||
isFromUnusedServer := !usedServers[disk.NodeID]
|
||||
score := calculateDiskScore(disk)
|
||||
|
||||
// Prefer unused servers
|
||||
if isFromUnusedServer && !bestIsFromUnusedServer {
|
||||
bestDisk = disk
|
||||
bestScore = score
|
||||
bestIsFromUnusedServer = true
|
||||
} else if isFromUnusedServer == bestIsFromUnusedServer && score > bestScore {
|
||||
bestDisk = disk
|
||||
bestScore = score
|
||||
}
|
||||
}
|
||||
|
||||
return bestDisk
|
||||
}
|
||||
|
||||
// sortDisksByScore returns disks sorted by score (best first)
|
||||
func sortDisksByScore(disks []*DiskCandidate) []*DiskCandidate {
|
||||
sorted := make([]*DiskCandidate, len(disks))
|
||||
copy(sorted, disks)
|
||||
sort.Slice(sorted, func(i, j int) bool {
|
||||
return calculateDiskScore(sorted[i]) > calculateDiskScore(sorted[j])
|
||||
})
|
||||
return sorted
|
||||
}
|
||||
|
||||
// calculateDiskScore calculates a score for a disk candidate
|
||||
// Higher score is better
|
||||
func calculateDiskScore(disk *DiskCandidate) float64 {
|
||||
score := 0.0
|
||||
|
||||
// Primary factor: available capacity (lower utilization is better)
|
||||
if disk.MaxVolumeCount > 0 {
|
||||
utilization := float64(disk.VolumeCount) / float64(disk.MaxVolumeCount)
|
||||
score += (1.0 - utilization) * 60.0 // Up to 60 points
|
||||
} else {
|
||||
score += 30.0 // Default if no max count
|
||||
}
|
||||
|
||||
// Secondary factor: fewer shards already on this disk is better
|
||||
score += float64(10-disk.ShardCount) * 2.0 // Up to 20 points
|
||||
|
||||
// Tertiary factor: lower load is better
|
||||
score += float64(10 - disk.LoadCount) // Up to 10 points
|
||||
|
||||
return score
|
||||
}
|
||||
|
||||
// addDiskToResult adds a disk to the result and updates tracking maps
|
||||
func addDiskToResult(result *PlacementResult, disk *DiskCandidate,
|
||||
usedDisks, usedServers, usedRacks map[string]bool) {
|
||||
diskKey := getDiskKey(disk)
|
||||
rackKey := getRackKey(disk)
|
||||
|
||||
result.SelectedDisks = append(result.SelectedDisks, disk)
|
||||
usedDisks[diskKey] = true
|
||||
usedServers[disk.NodeID] = true
|
||||
usedRacks[rackKey] = true
|
||||
result.ShardsPerServer[disk.NodeID]++
|
||||
result.ShardsPerRack[rackKey]++
|
||||
result.ShardsPerDC[disk.DataCenter]++
|
||||
}
|
||||
|
||||
// VerifySpread checks if the placement result meets diversity requirements
|
||||
func VerifySpread(result *PlacementResult, minServers, minRacks int) error {
|
||||
if result.ServersUsed < minServers {
|
||||
return fmt.Errorf("only %d servers used, need at least %d", result.ServersUsed, minServers)
|
||||
}
|
||||
if result.RacksUsed < minRacks {
|
||||
return fmt.Errorf("only %d racks used, need at least %d", result.RacksUsed, minRacks)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// CalculateIdealDistribution returns the ideal number of shards per server
|
||||
// when we have a certain number of shards and servers
|
||||
func CalculateIdealDistribution(totalShards, numServers int) (min, max int) {
|
||||
if numServers <= 0 {
|
||||
return 0, totalShards
|
||||
}
|
||||
min = totalShards / numServers
|
||||
max = min
|
||||
if totalShards%numServers != 0 {
|
||||
max = min + 1
|
||||
}
|
||||
return
|
||||
}
|
||||
Reference in New Issue
Block a user