lsm/seaweedfs
1
0
Fork 0
mirror of https://github.com/seaweedfs/seaweedfs.git synced 2025-11-24 16:53:14 +08:00
Code Issues Actions 17 Packages Projects Releases Wiki Activity

sum from all shards

Browse Source
This commit is contained in:
chrislu
2025-08-13 18:48:05 -07:00
parent 722aebdf11
commit 1de251d575
1 changed files with 28 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
weed/admin/dash/ec_shard_management.go
View File

@@ -769,7 +769,7 @@ func (s *AdminServer) GetEcVolumeDetails(volumeID uint32, sortBy string, sortOrd
// Get EC volume health metrics (deletion information) // Get EC volume health metrics (deletion information)
volumeHealth, err := s.getEcVolumeHealthMetrics(volumeID) volumeHealth, err := s.getEcVolumeHealthMetrics(volumeID)
if err != nil { if err != nil {
glog.V(1).Infof("Failed to get EC volume health metrics for volume %d: %v", volumeID, err) glog.V(0).Infof("ERROR: Failed to get EC volume health metrics for volume %d: %v", volumeID, err)
// Don't fail the request, just use default values // Don't fail the request, just use default values
volumeHealth = &EcVolumeHealthInfo{ volumeHealth = &EcVolumeHealthInfo{
TotalSize: 0, TotalSize: 0,
@@ -853,6 +853,7 @@ func (s *AdminServer) GetEcVolumeDetails(volumeID uint32, sortBy string, sortOrd
// getEcVolumeHealthMetrics retrieves health metrics for an EC volume // getEcVolumeHealthMetrics retrieves health metrics for an EC volume
func (s *AdminServer) getEcVolumeHealthMetrics(volumeID uint32) (*EcVolumeHealthInfo, error) { func (s *AdminServer) getEcVolumeHealthMetrics(volumeID uint32) (*EcVolumeHealthInfo, error) {
glog.V(0).Infof("DEBUG: getEcVolumeHealthMetrics called for volume %d", volumeID)
// Get list of servers that have shards for this EC volume // Get list of servers that have shards for this EC volume
var servers []string var servers []string
@@ -889,16 +890,18 @@ func (s *AdminServer) getEcVolumeHealthMetrics(volumeID uint32) (*EcVolumeHealth
return nil, fmt.Errorf("failed to get topology info: %v", err) return nil, fmt.Errorf("failed to get topology info: %v", err)
} }
glog.V(0).Infof("DEBUG: Found %d servers with EC shards for volume %d: %v", len(servers), volumeID, servers)
if len(servers) == 0 { if len(servers) == 0 {
return nil, fmt.Errorf("no servers found with EC shards for volume %d", volumeID) return nil, fmt.Errorf("no servers found with EC shards for volume %d", volumeID)
} }
// Aggregate health metrics from ALL servers that have EC shards // Aggregate health metrics from ALL servers that have EC shards
var aggregatedHealth *EcVolumeHealthInfo var aggregatedHealth *EcVolumeHealthInfo
var maxTotalSize uint64 var totalSize uint64
var maxFileCount uint64 var totalFileCount uint64
var maxDeletedBytes uint64 var totalDeletedBytes uint64
var maxDeletedCount uint64 var totalDeletedCount uint64
validServers := 0
for _, server := range servers { for _, server := range servers {
healthInfo, err := s.getVolumeHealthFromServer(server, volumeID) healthInfo, err := s.getVolumeHealthFromServer(server, volumeID)
@@ -906,20 +909,16 @@ func (s *AdminServer) getEcVolumeHealthMetrics(volumeID uint32) (*EcVolumeHealth
glog.V(2).Infof("Failed to get volume health from server %s for volume %d: %v", server, volumeID, err) glog.V(2).Infof("Failed to get volume health from server %s for volume %d: %v", server, volumeID, err)
continue // Try next server continue // Try next server
} }
glog.V(0).Infof("DEBUG: getVolumeHealthFromServer returned for %s: healthInfo=%v", server, healthInfo != nil)
if healthInfo != nil { if healthInfo != nil {
// Use the maximum values across servers // Sum the values across all servers (each server contributes its shard data)
if healthInfo.TotalSize > maxTotalSize { totalSize += healthInfo.TotalSize
maxTotalSize = healthInfo.TotalSize totalFileCount += healthInfo.FileCount
} totalDeletedBytes += healthInfo.DeletedByteCount
if healthInfo.FileCount > maxFileCount { totalDeletedCount += healthInfo.DeleteCount
maxFileCount = healthInfo.FileCount validServers++
}
if healthInfo.DeletedByteCount > maxDeletedBytes { glog.V(0).Infof("DEBUG: Added server %s data: size=%d, files=%d, deleted_bytes=%d", server, healthInfo.TotalSize, healthInfo.FileCount, healthInfo.DeletedByteCount)
maxDeletedBytes = healthInfo.DeletedByteCount
}
if healthInfo.DeleteCount > maxDeletedCount {
maxDeletedCount = healthInfo.DeleteCount
}
// Store first non-nil health info as template for aggregated result // Store first non-nil health info as template for aggregated result
if aggregatedHealth == nil { if aggregatedHealth == nil {
@@ -929,17 +928,22 @@ func (s *AdminServer) getEcVolumeHealthMetrics(volumeID uint32) (*EcVolumeHealth
} }
// If we got aggregated data, finalize it // If we got aggregated data, finalize it
if aggregatedHealth != nil { glog.V(0).Infof("DEBUG: Aggregation check - aggregatedHealth=%v, validServers=%d", aggregatedHealth != nil, validServers)
aggregatedHealth.TotalSize = maxTotalSize if aggregatedHealth != nil && validServers > 0 {
aggregatedHealth.FileCount = maxFileCount // Use summed totals from all servers
aggregatedHealth.DeletedByteCount = maxDeletedBytes aggregatedHealth.TotalSize = totalSize
aggregatedHealth.DeleteCount = maxDeletedCount aggregatedHealth.FileCount = totalFileCount
aggregatedHealth.DeletedByteCount = totalDeletedBytes
aggregatedHealth.DeleteCount = totalDeletedCount
// Calculate garbage ratio from aggregated data // Calculate garbage ratio from aggregated data
if aggregatedHealth.TotalSize > 0 { if aggregatedHealth.TotalSize > 0 {
aggregatedHealth.GarbageRatio = float64(aggregatedHealth.DeletedByteCount) / float64(aggregatedHealth.TotalSize) aggregatedHealth.GarbageRatio = float64(aggregatedHealth.DeletedByteCount) / float64(aggregatedHealth.TotalSize)
} }
glog.V(0).Infof("SUCCESS: Aggregated EC volume %d from %d servers: %d total bytes -> %d MB",
volumeID, validServers, totalSize, totalSize/1024/1024)
return aggregatedHealth, nil return aggregatedHealth, nil
} }
@@ -1009,7 +1013,7 @@ func (s *AdminServer) getVolumeHealthFromServer(server string, volumeID uint32)
volumeID, server, healthInfo.DeletedByteCount, healthInfo.DeleteCount, healthInfo.TotalSize) volumeID, server, healthInfo.DeletedByteCount, healthInfo.DeleteCount, healthInfo.TotalSize)
} }
return nil return nil // Return from WithVolumeServerClient callback - healthInfo is captured by closure
}) })
return healthInfo, err return healthInfo, err