mirror of
https://github.com/seaweedfs/seaweedfs.git
synced 2026-09-26 09:54:47 +00:00
Volume replication groundwork (Phase 6 hardening plan): - UDS handler: Add Store opcode (0x02) with opcode-based dispatch, StoreRequest/StoreResponse wire types, and size validation - storage/store.go: AppendRawNeedle() for accepting pre-serialized needle bytes from RDMA replication path - storage/needle: Return raw bytes from needle Append() to enable forwarding exact .dat bytes without re-serialization - storage/sra_transport.go: UDS client for outbound RDMA replication (connects to /tmp/sra-transport.sock, nil-safe fallback to HTTP) - command/volume.go: Wire -uds.transport flag - Integration test for Store opcode round-trip - Golden wire tests for Store request/response encoding Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
152 lines
4.5 KiB
Go
152 lines
4.5 KiB
Go
package needle
|
|
|
|
import (
|
|
"bytes"
|
|
"fmt"
|
|
|
|
"github.com/seaweedfs/seaweedfs/weed/glog"
|
|
"github.com/seaweedfs/seaweedfs/weed/storage/backend"
|
|
. "github.com/seaweedfs/seaweedfs/weed/storage/types"
|
|
"github.com/seaweedfs/seaweedfs/weed/util"
|
|
"github.com/seaweedfs/seaweedfs/weed/util/buffer_pool"
|
|
)
|
|
|
|
func (n *Needle) Append(w backend.BackendStorageFile, version Version) (offset uint64, size Size, actualSize int64, err error) {
|
|
end, _, e := w.GetStat()
|
|
if e != nil {
|
|
err = fmt.Errorf("Cannot Read Current Volume Position: %w", e)
|
|
return
|
|
}
|
|
offset = uint64(end)
|
|
if offset >= MaxPossibleVolumeSize && len(n.Data) != 0 {
|
|
err = fmt.Errorf("Volume Size %d Exceeded %d", offset, MaxPossibleVolumeSize)
|
|
return
|
|
}
|
|
bytesBuffer := buffer_pool.SyncPoolGetBuffer()
|
|
defer func() {
|
|
if err != nil {
|
|
if te := w.Truncate(end); te != nil {
|
|
// handle error or log
|
|
}
|
|
}
|
|
buffer_pool.SyncPoolPutBuffer(bytesBuffer)
|
|
}()
|
|
|
|
size, actualSize, err = writeNeedleByVersion(version, n, offset, bytesBuffer)
|
|
if err != nil {
|
|
return
|
|
}
|
|
|
|
_, err = w.WriteAt(bytesBuffer.Bytes(), int64(offset))
|
|
if err != nil {
|
|
err = fmt.Errorf("failed to write %d bytes to %s at offset %d: %w", actualSize, w.Name(), offset, err)
|
|
}
|
|
|
|
return offset, size, actualSize, err
|
|
}
|
|
|
|
// AppendGetBytes is like Append but also returns a copy of the raw serialized bytes
|
|
// that were written to disk. This enables RDMA replication to send the exact .dat bytes
|
|
// to a remote volume without re-serialization.
|
|
func (n *Needle) AppendGetBytes(w backend.BackendStorageFile, version Version) (offset uint64, size Size, actualSize int64, rawBytes []byte, err error) {
|
|
end, _, e := w.GetStat()
|
|
if e != nil {
|
|
err = fmt.Errorf("Cannot Read Current Volume Position: %w", e)
|
|
return
|
|
}
|
|
offset = uint64(end)
|
|
if offset >= MaxPossibleVolumeSize && len(n.Data) != 0 {
|
|
err = fmt.Errorf("Volume Size %d Exceeded %d", offset, MaxPossibleVolumeSize)
|
|
return
|
|
}
|
|
bytesBuffer := buffer_pool.SyncPoolGetBuffer()
|
|
defer func() {
|
|
if err != nil {
|
|
if te := w.Truncate(end); te != nil {
|
|
// handle error or log
|
|
}
|
|
}
|
|
buffer_pool.SyncPoolPutBuffer(bytesBuffer)
|
|
}()
|
|
|
|
size, actualSize, err = writeNeedleByVersion(version, n, offset, bytesBuffer)
|
|
if err != nil {
|
|
return
|
|
}
|
|
|
|
// Copy serialized bytes before they are written to disk and the buffer is returned to pool
|
|
src := bytesBuffer.Bytes()
|
|
rawBytes = make([]byte, len(src))
|
|
copy(rawBytes, src)
|
|
|
|
_, err = w.WriteAt(src, int64(offset))
|
|
if err != nil {
|
|
err = fmt.Errorf("failed to write %d bytes to %s at offset %d: %w", actualSize, w.Name(), offset, err)
|
|
}
|
|
|
|
return offset, size, actualSize, rawBytes, err
|
|
}
|
|
|
|
func WriteNeedleBlob(w backend.BackendStorageFile, dataSlice []byte, size Size, appendAtNs uint64, version Version) (offset uint64, err error) {
|
|
|
|
if end, _, e := w.GetStat(); e == nil {
|
|
defer func(w backend.BackendStorageFile, off int64) {
|
|
if err != nil {
|
|
if te := w.Truncate(end); te != nil {
|
|
glog.V(0).Infof("Failed to truncate %s back to %d with error: %v", w.Name(), end, te)
|
|
}
|
|
}
|
|
}(w, end)
|
|
offset = uint64(end)
|
|
} else {
|
|
err = fmt.Errorf("Cannot Read Current Volume Position: %v", e)
|
|
return
|
|
}
|
|
|
|
if version == Version3 {
|
|
// compute byte offset as int to compare and slice correctly
|
|
tsOffset := int(NeedleHeaderSize) + int(size) + NeedleChecksumSize
|
|
// Ensure dataSlice has enough capacity for the timestamp
|
|
if tsOffset < 0 {
|
|
err = fmt.Errorf("invalid needle size %d results in negative timestamp offset %d", size, tsOffset)
|
|
return
|
|
}
|
|
if tsOffset+TimestampSize > len(dataSlice) {
|
|
err = fmt.Errorf("needle blob buffer too small: need %d bytes, have %d", tsOffset+TimestampSize, len(dataSlice))
|
|
return
|
|
}
|
|
util.Uint64toBytes(dataSlice[tsOffset:tsOffset+TimestampSize], appendAtNs)
|
|
}
|
|
|
|
if err == nil {
|
|
_, err = w.WriteAt(dataSlice, int64(offset))
|
|
}
|
|
|
|
return
|
|
|
|
}
|
|
|
|
// prepareNeedleWrite encapsulates the common beginning logic for all versioned writeNeedle functions.
|
|
func prepareNeedleWrite(w backend.BackendStorageFile, n *Needle) (offset uint64, bytesBuffer *bytes.Buffer, cleanup func(err error), err error) {
|
|
end, _, e := w.GetStat()
|
|
if e != nil {
|
|
err = fmt.Errorf("Cannot Read Current Volume Position: %w", e)
|
|
return
|
|
}
|
|
offset = uint64(end)
|
|
if offset >= MaxPossibleVolumeSize && len(n.Data) != 0 {
|
|
err = fmt.Errorf("Volume Size %d Exceeded %d", offset, MaxPossibleVolumeSize)
|
|
return
|
|
}
|
|
bytesBuffer = buffer_pool.SyncPoolGetBuffer()
|
|
cleanup = func(err error) {
|
|
if err != nil {
|
|
if te := w.Truncate(end); te != nil {
|
|
// handle error or log
|
|
}
|
|
}
|
|
buffer_pool.SyncPoolPutBuffer(bytesBuffer)
|
|
}
|
|
return
|
|
}
|