diff --git a/weed/command/mount.go b/weed/command/mount.go index 98f139c6f..ba0639bc1 100644 --- a/weed/command/mount.go +++ b/weed/command/mount.go @@ -43,6 +43,13 @@ type MountOptions struct { rdmaReadOnly *bool rdmaMaxConcurrent *int rdmaTimeoutMs *int + + // ML optimization options + mlOptimizationEnabled *bool + mlPrefetchWorkers *int + mlConfidenceThreshold *float64 + mlMaxPrefetchAhead *int + mlBatchSize *int } var ( @@ -90,6 +97,13 @@ func init() { mountOptions.rdmaReadOnly = cmdMount.Flag.Bool("rdma.readOnly", false, "use RDMA for reads only (writes use HTTP)") mountOptions.rdmaMaxConcurrent = cmdMount.Flag.Int("rdma.maxConcurrent", 64, "max concurrent RDMA operations") mountOptions.rdmaTimeoutMs = cmdMount.Flag.Int("rdma.timeoutMs", 5000, "RDMA operation timeout in milliseconds") + + // ML optimization flags + mountOptions.mlOptimizationEnabled = cmdMount.Flag.Bool("ml.enabled", false, "enable ML-aware optimizations for machine learning workloads") + mountOptions.mlPrefetchWorkers = cmdMount.Flag.Int("ml.prefetchWorkers", 8, "number of prefetch worker threads for ML workloads") + mountOptions.mlConfidenceThreshold = cmdMount.Flag.Float64("ml.confidenceThreshold", 0.6, "minimum confidence threshold to trigger ML prefetch") + mountOptions.mlMaxPrefetchAhead = cmdMount.Flag.Int("ml.maxPrefetchAhead", 8, "maximum number of chunks to prefetch ahead") + mountOptions.mlBatchSize = cmdMount.Flag.Int("ml.batchSize", 3, "batch size for ML prefetch operations") mountCpuProfile = cmdMount.Flag.String("cpuprofile", "", "cpu profile output file") mountMemProfile = cmdMount.Flag.String("memprofile", "", "memory profile output file") @@ -124,5 +138,17 @@ var cmdMount = &Command{ -rdma.maxConcurrent=64 Max concurrent RDMA operations -rdma.timeoutMs=5000 RDMA operation timeout in milliseconds + ML Optimization: + For machine learning workloads, enable intelligent prefetching and caching: + weed mount -filer=localhost:8888 -dir=/mnt/seaweedfs \ + -ml.enabled=true + + ML Options: + -ml.enabled=false Enable ML-aware optimizations + -ml.prefetchWorkers=8 Number of concurrent prefetch workers + -ml.confidenceThreshold=0.6 Minimum confidence to trigger ML prefetch + -ml.maxPrefetchAhead=8 Maximum chunks to prefetch ahead + -ml.batchSize=3 Batch size for prefetch operations + `, } diff --git a/weed/command/mount_std.go b/weed/command/mount_std.go index 53b09589d..afc00e781 100644 --- a/weed/command/mount_std.go +++ b/weed/command/mount_std.go @@ -260,6 +260,12 @@ func RunMount(option *MountOptions, umask os.FileMode) bool { RdmaReadOnly: *option.rdmaReadOnly, RdmaMaxConcurrent: *option.rdmaMaxConcurrent, RdmaTimeoutMs: *option.rdmaTimeoutMs, + // ML optimization options + MLOptimizationEnabled: *option.mlOptimizationEnabled, + MLPrefetchWorkers: *option.mlPrefetchWorkers, + MLConfidenceThreshold: *option.mlConfidenceThreshold, + MLMaxPrefetchAhead: *option.mlMaxPrefetchAhead, + MLBatchSize: *option.mlBatchSize, }) // create mount root diff --git a/weed/mount/ml_integration.go b/weed/mount/ml_integration.go index fefd23cd6..7dc18b00d 100644 --- a/weed/mount/ml_integration.go +++ b/weed/mount/ml_integration.go @@ -1,6 +1,8 @@ package mount import ( + "time" + "github.com/hanwen/go-fuse/v2/fuse" "github.com/seaweedfs/seaweedfs/weed/glog" "github.com/seaweedfs/seaweedfs/weed/mount/ml" @@ -35,6 +37,42 @@ func NewMLIntegrationManager(chunkCache chunk_cache.ChunkCache, lookupFn wdclien return manager } +// NewMLIntegrationManagerWithConfig creates a new ML integration manager with custom configuration +func NewMLIntegrationManagerWithConfig( + chunkCache chunk_cache.ChunkCache, + lookupFn wdclient.LookupFileIdFunctionType, + prefetchWorkers int, + confidenceThreshold float64, + maxPrefetchAhead int, + batchSize int, +) *MLIntegrationManager { + config := &ml.MLConfig{ + PrefetchWorkers: prefetchWorkers, + PrefetchQueueSize: prefetchWorkers * 4, // 4x workers for queue depth + PrefetchTimeout: 30 * time.Second, + EnableMLHeuristics: true, + SequentialThreshold: 5, + ConfidenceThreshold: confidenceThreshold, + MaxPrefetchAhead: maxPrefetchAhead, + PrefetchBatchSize: batchSize, + } + + mlOpt := ml.NewMLOptimization(config, chunkCache, lookupFn) + + // Create FUSE integration + fuseInt := ml.NewFUSEMLIntegration(mlOpt) + + manager := &MLIntegrationManager{ + mlOptimization: mlOpt, + fuseIntegration: fuseInt, + enabled: true, + } + + glog.V(1).Infof("ML integration manager initialized with custom config: workers=%d, confidence=%.2f, prefetchAhead=%d, batchSize=%d", + prefetchWorkers, confidenceThreshold, maxPrefetchAhead, batchSize) + return manager +} + // EnableMLOptimization enables or disables ML optimization func (mgr *MLIntegrationManager) EnableMLOptimization(enabled bool) { mgr.enabled = enabled diff --git a/weed/mount/weedfs.go b/weed/mount/weedfs.go index 95864ef00..445136f96 100644 --- a/weed/mount/weedfs.go +++ b/weed/mount/weedfs.go @@ -70,6 +70,13 @@ type Option struct { RdmaReadOnly bool RdmaMaxConcurrent int RdmaTimeoutMs int + + // ML optimization options + MLOptimizationEnabled bool + MLPrefetchWorkers int + MLConfidenceThreshold float64 + MLMaxPrefetchAhead int + MLBatchSize int uniqueCacheDirForRead string uniqueCacheDirForWrite string @@ -96,6 +103,7 @@ type WFS struct { IsOverQuota bool fhLockTable *util.LockTable[FileHandleId] rdmaClient *RDMAMountClient + mlIntegration *MLIntegrationManager FilerConf *filer.FilerConf } @@ -151,6 +159,9 @@ func NewSeaweedFileSystem(option *Option) *WFS { if wfs.rdmaClient != nil { wfs.rdmaClient.Close() } + if wfs.mlIntegration != nil { + wfs.mlIntegration.Shutdown() + } }) // Initialize RDMA client if enabled @@ -169,6 +180,20 @@ func NewSeaweedFileSystem(option *Option) *WFS { option.RdmaSidecarAddr, option.RdmaMaxConcurrent, option.RdmaTimeoutMs) } } + + // Initialize ML optimization if enabled + if option.MLOptimizationEnabled { + wfs.mlIntegration = NewMLIntegrationManagerWithConfig( + wfs.chunkCache, + wfs.LookupFn(), + option.MLPrefetchWorkers, + option.MLConfidenceThreshold, + option.MLMaxPrefetchAhead, + option.MLBatchSize, + ) + glog.Infof("ML optimization enabled: prefetchWorkers=%d, confidenceThreshold=%.2f, maxPrefetchAhead=%d", + option.MLPrefetchWorkers, option.MLConfidenceThreshold, option.MLMaxPrefetchAhead) + } if wfs.option.ConcurrentWriters > 0 { wfs.concurrentWriters = util.NewLimitedConcurrentExecutor(wfs.option.ConcurrentWriters) diff --git a/weed/mount/weedfs_attr.go b/weed/mount/weedfs_attr.go index 0bd5771cd..ccdeb3c9f 100644 --- a/weed/mount/weedfs_attr.go +++ b/weed/mount/weedfs_attr.go @@ -22,6 +22,12 @@ func (wfs *WFS) GetAttr(cancel <-chan struct{}, input *fuse.GetAttrIn, out *fuse _, _, entry, status := wfs.maybeReadEntry(inode) if status == fuse.OK { out.AttrValid = 1 + + // Apply ML-specific attribute cache optimizations if enabled + if wfs.mlIntegration != nil { + wfs.mlIntegration.OptimizeAttributes(inode, out) + } + wfs.setAttrByPbEntry(&out.Attr, inode, entry, true) return status } else { diff --git a/weed/mount/weedfs_file_io.go b/weed/mount/weedfs_file_io.go index 04fe7f21c..add20caea 100644 --- a/weed/mount/weedfs_file_io.go +++ b/weed/mount/weedfs_file_io.go @@ -67,6 +67,14 @@ func (wfs *WFS) Open(cancel <-chan struct{}, in *fuse.OpenIn, out *fuse.OpenOut) if status == fuse.OK { out.Fh = uint64(fileHandle.fh) out.OpenFlags = in.Flags + + // Apply ML optimizations if enabled + if wfs.mlIntegration != nil { + if path, _, entry, pathStatus := wfs.maybeReadEntry(in.NodeId); pathStatus == fuse.OK { + wfs.mlIntegration.OnFileOpen(in.NodeId, entry, string(path), in.Flags, out) + } + } + if wfs.option.IsMacOs { // remove the direct_io flag, as it is not well-supported on macOS // https://code.google.com/archive/p/macfuse/wikis/OPTIONS.wiki recommended to avoid the direct_io flag @@ -106,5 +114,10 @@ func (wfs *WFS) Open(cancel <-chan struct{}, in *fuse.OpenIn, out *fuse.OpenOut) * @param fi file information */ func (wfs *WFS) Release(cancel <-chan struct{}, in *fuse.ReleaseIn) { + // Notify ML integration of file close + if wfs.mlIntegration != nil { + wfs.mlIntegration.OnFileClose(in.NodeId) + } + wfs.ReleaseHandle(FileHandleId(in.Fh)) } diff --git a/weed/mount/weedfs_file_read.go b/weed/mount/weedfs_file_read.go index dc79d3dc7..b1cf2d948 100644 --- a/weed/mount/weedfs_file_read.go +++ b/weed/mount/weedfs_file_read.go @@ -62,6 +62,11 @@ func (wfs *WFS) Read(cancel <-chan struct{}, in *fuse.ReadIn, buff []byte) (fuse glog.Warningf("file handle read %s %d: %v", fh.FullPath(), totalRead, err) return nil, fuse.EIO } + + // Notify ML integration of file read for pattern detection + if wfs.mlIntegration != nil && totalRead > 0 { + wfs.mlIntegration.OnFileRead(in.NodeId, offset, int(totalRead)) + } if IsDebugFileReadWrite { // print(".")