Files
tailscale/util/ioqueue/volatile.go
Joe Tsai 1a9090f14e util/ioqueue: add helper functions
This adds the following helpers:

* WaitLength waits until the buffer is non-empty or a context is canceled.
  It supports batchDelays so that even if data is ready, it blocks
  until the delay is over. This is useful for how we upload logs for iOS,
  where we deliberately wait a few minutes to reduce wakeup costs.

* DiscardOversize asynchronously discards data in the buffer
  once it exceeds the specified maxSize. It takes in an optional frameLen
  to ensure discarding maintains consistent frames in the buffer.
  This avoids known problems with today's ring buffer where we can get
  torn frames that lead to silent data corruption.

  An alternative approach would be to synchronously delete data upon
  an oversize condition at Write time, but there are two reasons
  not to do that:

    1. Doing requires teaching each Buffer implementation about
    the concept of framing, which the interface deliberately avoids.

    2. We want the write path to be extremely fast as we never want to
    be blocking production logic. Going over maxSize momentarily is
    considered a better tradeoff than synchronously blocking writes.

* StreamReader converts a non-blocking Buffer reader into a blocking one.
  This exists primarily for debugging where you can simply stream
  the entirety of a buffer to stdout.

We also adjust the package to avoid wrapping io.EOF and ErrEmpty
as those are sentinel errors with very specific meanings.

Updates tailscale/corp#21363

Signed-off-by: Joe Tsai <joetsai@digital-static.net>
2026-08-14 12:35:33 -07:00

187 lines
5.8 KiB
Go

// Copyright (c) Tailscale Inc & contributors
// SPDX-License-Identifier: BSD-3-Clause
package ioqueue
import (
"io"
"math/bits"
"sync"
"tailscale.com/types/bools"
)
// Statically verify that VolatileBuffer implements Buffer.
var _ Buffer = (*VolatileBuffer)(nil)
// VolatileBuffer is an in-memory implementation of [Buffer].
// The zero value is an empty buffer ready for use.
//
// Unread data is not durable: it is lost when the process exits.
// Close the write side with [VolatileBuffer.CloseWrite] to signal
// that no further bytes will be produced; subsequent [Buffer.Read]
// and [Buffer.Peek] calls return [io.EOF] once the buffer is drained,
// and outstanding [Buffer.WaitUntil] waiters are unblocked.
type VolatileBuffer struct {
mu sync.Mutex
// buf may retain a consumed prefix after reads/discards.
// Unread data is always the trailing writeOffset-readOffset bytes.
buf []byte
readOffset int64
writeOffset int64
peakLen int64 // decayed by compactLocked
closedWrite bool
waiters offsetWaiters
}
// Len reports the size of the buffer,
// which is the number of written, but unread bytes.
func (b *VolatileBuffer) Len() int64 {
b.mu.Lock()
defer b.mu.Unlock()
return b.lenLocked()
}
func (b *VolatileBuffer) lenLocked() int64 {
return b.writeOffset - b.readOffset
}
// WriteOffset is the total number of bytes written.
func (b *VolatileBuffer) WriteOffset() int64 {
b.mu.Lock()
defer b.mu.Unlock()
return b.writeOffset
}
// ReadOffset is the total number of bytes read.
func (b *VolatileBuffer) ReadOffset() int64 {
b.mu.Lock()
defer b.mu.Unlock()
return b.readOffset
}
// dataLocked returns the unread portion of buf.
func (b *VolatileBuffer) dataLocked() []byte {
n := int(b.writeOffset - b.readOffset)
return b.buf[len(b.buf)-n:]
}
// compactLocked reclaims backing storage after a consume,
// by sliding data forward or allocating a smaller capacity buffer.
func (b *VolatileBuffer) compactLocked() {
data := b.dataLocked()
// If already read data is greater than unread data, then compact.
if len(b.buf) > 2*len(data) {
b.peakLen = max(b.peakLen, b.lenLocked())
// If total capacity is 4x greater than the peak Len,
// then allocate smaller capacity, otherwise slide data to front.
if int64(cap(b.buf)/4) > b.peakLen && cap(b.buf) > 4<<10 {
newCap := 2 << bits.Len(uint(b.peakLen)-1) // double the peak Len, but round up to power-of-2
newCap = max(newCap, 4<<10) // minimum buffer capacity to shrink to
b.buf = append(make([]byte, 0, newCap), data...) // allocate smaller buffer
} else {
b.buf = b.buf[:copy(b.buf, data)] // slide data to the front
}
b.peakLen = b.peakLen * 7 / 8 // reduce peakLen by 12.5%
}
}
// Write writes data to the end of the buffer,
// atomically incrementing Len and WriteOffset
// by the amount of bytes written.
// Write does not block.
// After [CloseWrite], Write returns an error.
func (b *VolatileBuffer) Write(p []byte) (int, error) {
b.mu.Lock()
defer b.mu.Unlock()
if b.closedWrite {
return 0, wrapError("write", errClosed)
}
b.buf = append(b.buf, p...)
b.writeOffset += int64(len(p))
b.peakLen = max(b.peakLen, b.lenLocked())
b.waiters.notify(b.writeOffset, b.closedWrite)
return len(p), nil
}
// Read reads data from the front of the buffer, atomically decrementing
// Len and incrementing ReadOffset by the amount of bytes read.
// Rather than blocking, it returns [ErrEmpty] when the buffer is empty.
// After [CloseWrite], it returns [io.EOF] once drained.
func (b *VolatileBuffer) Read(p []byte) (int, error) {
b.mu.Lock()
defer b.mu.Unlock()
data := b.dataLocked()
if len(data) == 0 {
return 0, bools.IfElse(b.closedWrite, io.EOF, ErrEmpty)
}
n := copy(p, data)
b.readOffset += int64(n)
b.compactLocked()
return n, nil
}
// Peek copies data from the front of the buffer into p
// without affecting Len or ReadOffset.
// It reports the current ReadOffset and the number of bytes copied.
// If no bytes are available, it reports [ErrEmpty],
// or [io.EOF] after [VolatileBuffer.CloseWrite] once drained.
func (b *VolatileBuffer) Peek(p []byte) (readOffset int64, n int64, err error) {
b.mu.Lock()
defer b.mu.Unlock()
readOffset = b.readOffset
data := b.dataLocked()
if len(data) == 0 {
return readOffset, 0, bools.IfElse(b.closedWrite, io.EOF, ErrEmpty)
}
return readOffset, int64(copy(p, data)), nil
}
// DiscardUntil discards bytes from the front of the buffer by
// incrementing ReadOffset to match the specified readOffset.
// See [Buffer.DiscardUntil] for full semantics.
func (b *VolatileBuffer) DiscardUntil(readOffset int64) (n int64, err error) {
b.mu.Lock()
defer b.mu.Unlock()
if readOffset <= b.readOffset {
return 0, nil
}
if readOffset > b.writeOffset {
err = bools.IfElse(b.closedWrite, io.EOF, ErrEmpty)
}
readOffset = min(readOffset, b.writeOffset)
n = readOffset - b.readOffset
b.readOffset = readOffset
b.compactLocked()
return n, err
}
// WaitUntil returns a channel that is closed when WriteOffset exceeds
// the specified writeOffset, or when the write side is closed via
// [CloseWrite] (so waiters do not block indefinitely after close).
//
// Multiple waiters for the same writeOffset share one channel.
func (b *VolatileBuffer) WaitUntil(writeOffset int64) <-chan struct{} {
b.mu.Lock()
defer b.mu.Unlock()
return b.waiters.wait(b.writeOffset, writeOffset, b.closedWrite)
}
// CloseWrite closes the write side of the buffer.
// After CloseWrite, [Write] fails, [Read]/[Peek] return [io.EOF]
// once all buffered data has been consumed, and all [WaitUntil]
// waiters are unblocked.
// CloseWrite is idempotent and returns an error if already closed.
func (b *VolatileBuffer) CloseWrite() error {
b.mu.Lock()
defer b.mu.Unlock()
if b.closedWrite {
return wrapError("close", errClosed)
}
b.closedWrite = true
b.waiters.notify(b.writeOffset, b.closedWrite)
return nil
}