mirror of
https://github.com/minio/minio.git
synced 2024-12-26 15:15:55 -05:00
d4b391de1b
Replace the `io.Pipe` from streamingBitrotWriter -> CreateFile with a fixed size ring buffer. This will add an output buffer for encoded shards to be written to disk - potentially via RPC. This will remove blocking when `(*streamingBitrotWriter).Write` is called, and it writes hashes and data. With current settings, the write looks like this: ``` Outbound ┌───────────────────┐ ┌────────────────┐ ┌───────────────┐ ┌────────────────┐ │ │ Parr. │ │ (http body) │ │ │ │ │ Bitrot Hash │ Write │ Pipe │ Read │ HTTP buffer │ Write (syscall) │ TCP Buffer │ │ Erasure Shard │ ──────────► │ (unbuffered) │ ────────────► │ (64K Max) │ ───────────────────► │ (4MB) │ │ │ │ │ │ (io.Copy) │ │ │ └───────────────────┘ └────────────────┘ └───────────────┘ └────────────────┘ ``` We write a Hash (32 bytes). Since the pipe is unbuffered, it will block until the 32 bytes have been delivered to the TCP buffer, and the next Read hits the Pipe. Then we write the shard data. This will typically be bigger than 64KB, so it will block until two blocks have been read from the pipe. When we insert a ring buffer: ``` Outbound ┌───────────────────┐ ┌────────────────┐ ┌───────────────┐ ┌────────────────┐ │ │ │ │ (http body) │ │ │ │ │ Bitrot Hash │ Write │ Ring Buffer │ Read │ HTTP buffer │ Write (syscall) │ TCP Buffer │ │ Erasure Shard │ ──────────► │ (2MB) │ ────────────► │ (64K Max) │ ───────────────────► │ (4MB) │ │ │ │ │ │ (io.Copy) │ │ │ └───────────────────┘ └────────────────┘ └───────────────┘ └────────────────┘ ``` The hash+shard will fit within the ring buffer, so writes will not block - but will complete after a memcopy. Reads can fill the 64KB buffer if there is data for it. If the network is congested, the ring buffer will become filled, and all syscalls will be on full buffers. Only when the ring buffer is filled will erasure coding start blocking. Since there is always "space" to write output data, we remove the parallel writing since we are always writing to memory now, and the goroutine synchronization overhead probably not worth taking. If the output were blocked in the existing, we would still wait for it to unblock in parallel write, so it would make no difference there - except now the ring buffer smoothes out the load. There are some micro-optimizations we could look at later. The biggest is that, in most cases, we could encode directly to the ring buffer - if we are not at a boundary. Also, "force filling" the Read requests (i.e., blocking until a full read can be completed) could be investigated and maybe allow concurrent memory on read and write.
619 lines
13 KiB
Go
619 lines
13 KiB
Go
// Copyright 2019 smallnest. All rights reserved.
|
|
// Use of this source code is governed by a MIT-style
|
|
// license that can be found in the LICENSE file.
|
|
|
|
package ringbuffer
|
|
|
|
import (
|
|
"context"
|
|
"errors"
|
|
"io"
|
|
"sync"
|
|
"unsafe"
|
|
)
|
|
|
|
var (
|
|
// ErrTooMuchDataToWrite is returned when the data to write is more than the buffer size.
|
|
ErrTooMuchDataToWrite = errors.New("too much data to write")
|
|
|
|
// ErrIsFull is returned when the buffer is full and not blocking.
|
|
ErrIsFull = errors.New("ringbuffer is full")
|
|
|
|
// ErrIsEmpty is returned when the buffer is empty and not blocking.
|
|
ErrIsEmpty = errors.New("ringbuffer is empty")
|
|
|
|
// ErrIsNotEmpty is returned when the buffer is not empty and not blocking.
|
|
ErrIsNotEmpty = errors.New("ringbuffer is not empty")
|
|
|
|
// ErrAcquireLock is returned when the lock is not acquired on Try operations.
|
|
ErrAcquireLock = errors.New("unable to acquire lock")
|
|
|
|
// ErrWriteOnClosed is returned when write on a closed ringbuffer.
|
|
ErrWriteOnClosed = errors.New("write on closed ringbuffer")
|
|
)
|
|
|
|
// RingBuffer is a circular buffer that implement io.ReaderWriter interface.
|
|
// It operates like a buffered pipe, where data written to a RingBuffer
|
|
// and can be read back from another goroutine.
|
|
// It is safe to concurrently read and write RingBuffer.
|
|
type RingBuffer struct {
|
|
buf []byte
|
|
size int
|
|
r int // next position to read
|
|
w int // next position to write
|
|
isFull bool
|
|
err error
|
|
block bool
|
|
mu sync.Mutex
|
|
wg sync.WaitGroup
|
|
readCond *sync.Cond // Signaled when data has been read.
|
|
writeCond *sync.Cond // Signaled when data has been written.
|
|
}
|
|
|
|
// New returns a new RingBuffer whose buffer has the given size.
|
|
func New(size int) *RingBuffer {
|
|
return &RingBuffer{
|
|
buf: make([]byte, size),
|
|
size: size,
|
|
}
|
|
}
|
|
|
|
// NewBuffer returns a new RingBuffer whose buffer is provided.
|
|
func NewBuffer(b []byte) *RingBuffer {
|
|
return &RingBuffer{
|
|
buf: b,
|
|
size: len(b),
|
|
}
|
|
}
|
|
|
|
// SetBlocking sets the blocking mode of the ring buffer.
|
|
// If block is true, Read and Write will block when there is no data to read or no space to write.
|
|
// If block is false, Read and Write will return ErrIsEmpty or ErrIsFull immediately.
|
|
// By default, the ring buffer is not blocking.
|
|
// This setting should be called before any Read or Write operation or after a Reset.
|
|
func (r *RingBuffer) SetBlocking(block bool) *RingBuffer {
|
|
r.block = block
|
|
if block {
|
|
r.readCond = sync.NewCond(&r.mu)
|
|
r.writeCond = sync.NewCond(&r.mu)
|
|
}
|
|
return r
|
|
}
|
|
|
|
// WithCancel sets a context to cancel the ring buffer.
|
|
// When the context is canceled, the ring buffer will be closed with the context error.
|
|
// A goroutine will be started and run until the provided context is canceled.
|
|
func (r *RingBuffer) WithCancel(ctx context.Context) *RingBuffer {
|
|
go func() {
|
|
select {
|
|
case <-ctx.Done():
|
|
r.CloseWithError(ctx.Err())
|
|
}
|
|
}()
|
|
return r
|
|
}
|
|
|
|
func (r *RingBuffer) setErr(err error, locked bool) error {
|
|
if !locked {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
}
|
|
if r.err != nil && r.err != io.EOF {
|
|
return r.err
|
|
}
|
|
|
|
switch err {
|
|
// Internal errors are transient
|
|
case nil, ErrIsEmpty, ErrIsFull, ErrAcquireLock, ErrTooMuchDataToWrite, ErrIsNotEmpty:
|
|
return err
|
|
default:
|
|
r.err = err
|
|
if r.block {
|
|
r.readCond.Broadcast()
|
|
r.writeCond.Broadcast()
|
|
}
|
|
}
|
|
return err
|
|
}
|
|
|
|
func (r *RingBuffer) readErr(locked bool) error {
|
|
if !locked {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
}
|
|
if r.err != nil {
|
|
if r.err == io.EOF {
|
|
if r.w == r.r && !r.isFull {
|
|
return io.EOF
|
|
}
|
|
return nil
|
|
}
|
|
return r.err
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// Read reads up to len(p) bytes into p. It returns the number of bytes read (0 <= n <= len(p)) and any error encountered.
|
|
// Even if Read returns n < len(p), it may use all of p as scratch space during the call.
|
|
// If some data is available but not len(p) bytes, Read conventionally returns what is available instead of waiting for more.
|
|
// When Read encounters an error or end-of-file condition after successfully reading n > 0 bytes, it returns the number of bytes read.
|
|
// It may return the (non-nil) error from the same call or return the error (and n == 0) from a subsequent call.
|
|
// Callers should always process the n > 0 bytes returned before considering the error err.
|
|
// Doing so correctly handles I/O errors that happen after reading some bytes and also both of the allowed EOF behaviors.
|
|
func (r *RingBuffer) Read(p []byte) (n int, err error) {
|
|
if len(p) == 0 {
|
|
return 0, r.readErr(false)
|
|
}
|
|
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
if err := r.readErr(true); err != nil {
|
|
return 0, err
|
|
}
|
|
|
|
r.wg.Add(1)
|
|
defer r.wg.Done()
|
|
n, err = r.read(p)
|
|
for err == ErrIsEmpty && r.block {
|
|
r.writeCond.Wait()
|
|
if err = r.readErr(true); err != nil {
|
|
break
|
|
}
|
|
n, err = r.read(p)
|
|
}
|
|
if r.block && n > 0 {
|
|
r.readCond.Broadcast()
|
|
}
|
|
return n, err
|
|
}
|
|
|
|
// TryRead read up to len(p) bytes into p like Read but it is not blocking.
|
|
// If it has not succeeded to acquire the lock, it return 0 as n and ErrAcquireLock.
|
|
func (r *RingBuffer) TryRead(p []byte) (n int, err error) {
|
|
ok := r.mu.TryLock()
|
|
if !ok {
|
|
return 0, ErrAcquireLock
|
|
}
|
|
defer r.mu.Unlock()
|
|
if err := r.readErr(true); err != nil {
|
|
return 0, err
|
|
}
|
|
if len(p) == 0 {
|
|
return 0, r.readErr(true)
|
|
}
|
|
|
|
n, err = r.read(p)
|
|
if r.block && n > 0 {
|
|
r.readCond.Broadcast()
|
|
}
|
|
return n, err
|
|
}
|
|
|
|
func (r *RingBuffer) read(p []byte) (n int, err error) {
|
|
if r.w == r.r && !r.isFull {
|
|
return 0, ErrIsEmpty
|
|
}
|
|
|
|
if r.w > r.r {
|
|
n = r.w - r.r
|
|
if n > len(p) {
|
|
n = len(p)
|
|
}
|
|
copy(p, r.buf[r.r:r.r+n])
|
|
r.r = (r.r + n) % r.size
|
|
return
|
|
}
|
|
|
|
n = r.size - r.r + r.w
|
|
if n > len(p) {
|
|
n = len(p)
|
|
}
|
|
|
|
if r.r+n <= r.size {
|
|
copy(p, r.buf[r.r:r.r+n])
|
|
} else {
|
|
c1 := r.size - r.r
|
|
copy(p, r.buf[r.r:r.size])
|
|
c2 := n - c1
|
|
copy(p[c1:], r.buf[0:c2])
|
|
}
|
|
r.r = (r.r + n) % r.size
|
|
|
|
r.isFull = false
|
|
|
|
return n, r.readErr(true)
|
|
}
|
|
|
|
// ReadByte reads and returns the next byte from the input or ErrIsEmpty.
|
|
func (r *RingBuffer) ReadByte() (b byte, err error) {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
if err = r.readErr(true); err != nil {
|
|
return 0, err
|
|
}
|
|
for r.w == r.r && !r.isFull {
|
|
if r.block {
|
|
r.writeCond.Wait()
|
|
err = r.readErr(true)
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
continue
|
|
}
|
|
return 0, ErrIsEmpty
|
|
}
|
|
b = r.buf[r.r]
|
|
r.r++
|
|
if r.r == r.size {
|
|
r.r = 0
|
|
}
|
|
|
|
r.isFull = false
|
|
return b, r.readErr(true)
|
|
}
|
|
|
|
// Write writes len(p) bytes from p to the underlying buf.
|
|
// It returns the number of bytes written from p (0 <= n <= len(p))
|
|
// and any error encountered that caused the write to stop early.
|
|
// If blocking n < len(p) will be returned only if an error occurred.
|
|
// Write returns a non-nil error if it returns n < len(p).
|
|
// Write will not modify the slice data, even temporarily.
|
|
func (r *RingBuffer) Write(p []byte) (n int, err error) {
|
|
if len(p) == 0 {
|
|
return 0, r.setErr(nil, false)
|
|
}
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
if err := r.err; err != nil {
|
|
if err == io.EOF {
|
|
err = ErrWriteOnClosed
|
|
}
|
|
return 0, err
|
|
}
|
|
wrote := 0
|
|
for len(p) > 0 {
|
|
n, err = r.write(p)
|
|
wrote += n
|
|
if !r.block || err == nil {
|
|
break
|
|
}
|
|
err = r.setErr(err, true)
|
|
if r.block && (err == ErrIsFull || err == ErrTooMuchDataToWrite) {
|
|
r.writeCond.Broadcast()
|
|
r.readCond.Wait()
|
|
p = p[n:]
|
|
err = nil
|
|
continue
|
|
}
|
|
break
|
|
}
|
|
if r.block && wrote > 0 {
|
|
r.writeCond.Broadcast()
|
|
}
|
|
|
|
return wrote, r.setErr(err, true)
|
|
}
|
|
|
|
// TryWrite writes len(p) bytes from p to the underlying buf like Write, but it is not blocking.
|
|
// If it has not succeeded to acquire the lock, it return 0 as n and ErrAcquireLock.
|
|
func (r *RingBuffer) TryWrite(p []byte) (n int, err error) {
|
|
if len(p) == 0 {
|
|
return 0, r.setErr(nil, false)
|
|
}
|
|
ok := r.mu.TryLock()
|
|
if !ok {
|
|
return 0, ErrAcquireLock
|
|
}
|
|
defer r.mu.Unlock()
|
|
if err := r.err; err != nil {
|
|
if err == io.EOF {
|
|
err = ErrWriteOnClosed
|
|
}
|
|
return 0, err
|
|
}
|
|
|
|
n, err = r.write(p)
|
|
if r.block && n > 0 {
|
|
r.writeCond.Broadcast()
|
|
}
|
|
return n, r.setErr(err, true)
|
|
}
|
|
|
|
func (r *RingBuffer) write(p []byte) (n int, err error) {
|
|
if r.isFull {
|
|
return 0, ErrIsFull
|
|
}
|
|
|
|
var avail int
|
|
if r.w >= r.r {
|
|
avail = r.size - r.w + r.r
|
|
} else {
|
|
avail = r.r - r.w
|
|
}
|
|
|
|
if len(p) > avail {
|
|
err = ErrTooMuchDataToWrite
|
|
p = p[:avail]
|
|
}
|
|
n = len(p)
|
|
|
|
if r.w >= r.r {
|
|
c1 := r.size - r.w
|
|
if c1 >= n {
|
|
copy(r.buf[r.w:], p)
|
|
r.w += n
|
|
} else {
|
|
copy(r.buf[r.w:], p[:c1])
|
|
c2 := n - c1
|
|
copy(r.buf[0:], p[c1:])
|
|
r.w = c2
|
|
}
|
|
} else {
|
|
copy(r.buf[r.w:], p)
|
|
r.w += n
|
|
}
|
|
|
|
if r.w == r.size {
|
|
r.w = 0
|
|
}
|
|
if r.w == r.r {
|
|
r.isFull = true
|
|
}
|
|
|
|
return n, err
|
|
}
|
|
|
|
// WriteByte writes one byte into buffer, and returns ErrIsFull if buffer is full.
|
|
func (r *RingBuffer) WriteByte(c byte) error {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
if err := r.err; err != nil {
|
|
if err == io.EOF {
|
|
err = ErrWriteOnClosed
|
|
}
|
|
return err
|
|
}
|
|
err := r.writeByte(c)
|
|
for err == ErrIsFull && r.block {
|
|
r.readCond.Wait()
|
|
err = r.setErr(r.writeByte(c), true)
|
|
}
|
|
if r.block && err == nil {
|
|
r.writeCond.Broadcast()
|
|
}
|
|
return err
|
|
}
|
|
|
|
// TryWriteByte writes one byte into buffer without blocking.
|
|
// If it has not succeeded to acquire the lock, it return ErrAcquireLock.
|
|
func (r *RingBuffer) TryWriteByte(c byte) error {
|
|
ok := r.mu.TryLock()
|
|
if !ok {
|
|
return ErrAcquireLock
|
|
}
|
|
defer r.mu.Unlock()
|
|
if err := r.err; err != nil {
|
|
if err == io.EOF {
|
|
err = ErrWriteOnClosed
|
|
}
|
|
return err
|
|
}
|
|
|
|
err := r.writeByte(c)
|
|
if err == nil && r.block {
|
|
r.writeCond.Broadcast()
|
|
}
|
|
return err
|
|
}
|
|
|
|
func (r *RingBuffer) writeByte(c byte) error {
|
|
if r.w == r.r && r.isFull {
|
|
return ErrIsFull
|
|
}
|
|
r.buf[r.w] = c
|
|
r.w++
|
|
|
|
if r.w == r.size {
|
|
r.w = 0
|
|
}
|
|
if r.w == r.r {
|
|
r.isFull = true
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// Length return the length of available read bytes.
|
|
func (r *RingBuffer) Length() int {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
|
|
if r.w == r.r {
|
|
if r.isFull {
|
|
return r.size
|
|
}
|
|
return 0
|
|
}
|
|
|
|
if r.w > r.r {
|
|
return r.w - r.r
|
|
}
|
|
|
|
return r.size - r.r + r.w
|
|
}
|
|
|
|
// Capacity returns the size of the underlying buffer.
|
|
func (r *RingBuffer) Capacity() int {
|
|
return r.size
|
|
}
|
|
|
|
// Free returns the length of available bytes to write.
|
|
func (r *RingBuffer) Free() int {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
|
|
if r.w == r.r {
|
|
if r.isFull {
|
|
return 0
|
|
}
|
|
return r.size
|
|
}
|
|
|
|
if r.w < r.r {
|
|
return r.r - r.w
|
|
}
|
|
|
|
return r.size - r.w + r.r
|
|
}
|
|
|
|
// WriteString writes the contents of the string s to buffer, which accepts a slice of bytes.
|
|
func (r *RingBuffer) WriteString(s string) (n int, err error) {
|
|
x := (*[2]uintptr)(unsafe.Pointer(&s))
|
|
h := [3]uintptr{x[0], x[1], x[1]}
|
|
buf := *(*[]byte)(unsafe.Pointer(&h))
|
|
return r.Write(buf)
|
|
}
|
|
|
|
// Bytes returns all available read bytes.
|
|
// It does not move the read pointer and only copy the available data.
|
|
// If the dst is big enough it will be used as destination,
|
|
// otherwise a new buffer will be allocated.
|
|
func (r *RingBuffer) Bytes(dst []byte) []byte {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
getDst := func(n int) []byte {
|
|
if cap(dst) < n {
|
|
return make([]byte, n)
|
|
}
|
|
return dst[:n]
|
|
}
|
|
|
|
if r.w == r.r {
|
|
if r.isFull {
|
|
buf := getDst(r.size)
|
|
copy(buf, r.buf[r.r:])
|
|
copy(buf[r.size-r.r:], r.buf[:r.w])
|
|
return buf
|
|
}
|
|
return nil
|
|
}
|
|
|
|
if r.w > r.r {
|
|
buf := getDst(r.w - r.r)
|
|
copy(buf, r.buf[r.r:r.w])
|
|
return buf
|
|
}
|
|
|
|
n := r.size - r.r + r.w
|
|
buf := getDst(n)
|
|
|
|
if r.r+n < r.size {
|
|
copy(buf, r.buf[r.r:r.r+n])
|
|
} else {
|
|
c1 := r.size - r.r
|
|
copy(buf, r.buf[r.r:r.size])
|
|
c2 := n - c1
|
|
copy(buf[c1:], r.buf[0:c2])
|
|
}
|
|
|
|
return buf
|
|
}
|
|
|
|
// IsFull returns this ringbuffer is full.
|
|
func (r *RingBuffer) IsFull() bool {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
|
|
return r.isFull
|
|
}
|
|
|
|
// IsEmpty returns this ringbuffer is empty.
|
|
func (r *RingBuffer) IsEmpty() bool {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
|
|
return !r.isFull && r.w == r.r
|
|
}
|
|
|
|
// CloseWithError closes the writer; reads will return
|
|
// no bytes and the error err, or EOF if err is nil.
|
|
//
|
|
// CloseWithError never overwrites the previous error if it exists
|
|
// and always returns nil.
|
|
func (r *RingBuffer) CloseWithError(err error) {
|
|
if err == nil {
|
|
err = io.EOF
|
|
}
|
|
r.setErr(err, false)
|
|
}
|
|
|
|
// CloseWriter closes the writer.
|
|
// Reads will return any remaining bytes and io.EOF.
|
|
func (r *RingBuffer) CloseWriter() {
|
|
r.setErr(io.EOF, false)
|
|
}
|
|
|
|
// Flush waits for the buffer to be empty and fully read.
|
|
// If not blocking ErrIsNotEmpty will be returned if the buffer still contains data.
|
|
func (r *RingBuffer) Flush() error {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
for r.w != r.r || r.isFull {
|
|
err := r.readErr(true)
|
|
if err != nil {
|
|
if err == io.EOF {
|
|
err = nil
|
|
}
|
|
return err
|
|
}
|
|
if !r.block {
|
|
return ErrIsNotEmpty
|
|
}
|
|
r.readCond.Wait()
|
|
}
|
|
|
|
err := r.readErr(true)
|
|
if err == io.EOF {
|
|
return nil
|
|
}
|
|
return err
|
|
}
|
|
|
|
// Reset the read pointer and writer pointer to zero.
|
|
func (r *RingBuffer) Reset() {
|
|
r.mu.Lock()
|
|
defer r.mu.Unlock()
|
|
|
|
// Set error so any readers/writers will return immediately.
|
|
r.setErr(errors.New("reset called"), true)
|
|
if r.block {
|
|
r.readCond.Broadcast()
|
|
r.writeCond.Broadcast()
|
|
}
|
|
|
|
// Unlock the mutex so readers/writers can finish.
|
|
r.mu.Unlock()
|
|
r.wg.Wait()
|
|
r.mu.Lock()
|
|
r.r = 0
|
|
r.w = 0
|
|
r.err = nil
|
|
r.isFull = false
|
|
}
|
|
|
|
// WriteCloser returns a WriteCloser that writes to the ring buffer.
|
|
// When the returned WriteCloser is closed, it will wait for all data to be read before returning.
|
|
func (r *RingBuffer) WriteCloser() io.WriteCloser {
|
|
return &writeCloser{RingBuffer: r}
|
|
}
|
|
|
|
type writeCloser struct {
|
|
*RingBuffer
|
|
}
|
|
|
|
// Close provides a close method for the WriteCloser.
|
|
func (wc *writeCloser) Close() error {
|
|
wc.CloseWriter()
|
|
return wc.Flush()
|
|
}
|