Limio is meant to be a dead simple rate-limiting library, primarily aimed at
providing an intuitive API surface for composable operational constraints. By
centering around chan int
, limio provides token bucket implementations with
decent backpressure and helps eliminate silly window syndrome by spreading out
token distribution over smaller units of time to provide a more even flow.
By using a channel, we also eliminate the need of Limiter implementations to follow a complex contract involving rates. They simply need to listen on a channel for some quantity, and then when a quantity arrives, perform that many operations.
Some examples usage of the limio.Reader
func slowCopy(w io.Writer, r io.Reader) error {
lr := limio.NewReader(r)
// Limit at 1MB/s
lr.SimpleLimit(1*MB, time.Second)
_, err := io.Copy(w, lr)
return err
}
func slowGroupCopy(ws []io.Writer, rs []io.Reader) error {
// For a simpler example, imagine len(ws) == len(rs) always
lmr := limio.NewSimpleManager()
// Limit all operations to an aggregate 1MB/s
lmr.SimpleLimit(1*MB, time.Second)
wg := &sync.WaitGroup{}
wg.Add(len(ws))
for i := range ws {
go func(i int) {
lr := limio.NewReader(rs[i])
lmr.Manage(lr)
// Obviously handle the errors in a real implementation
io.Copy(ws[i], lr)
wg.Done()
}(i)
}
wg.Wait()
return nil
}
-- import "astuart.co/limio"
Package limio provides an interface for rate limiting as well as a rate-limited Reader implementation.
In limio, there are two important interfaces for rate limiting. The first, Limiter, is more primitive. Most times, implementers of a Limiter should be creating a way to apply time constraints to a discretely quantifiable transaction.
The other interface is a Manager, which will likely be implemented in many more cases, as it allows consumers to take any number of Limiters and apply a strategy to the group. Most importantly, a Manager will also need to implement the Limiter interface, allowing consumers to treat its encapsulated group of Limiters as a single Limiter, knowing the strategy will be applied within the given limits.
const (
B int = 1 << (10 * (iota))
KB
MB
GB
TB
PB
EB
)
Some useful byte-sized (heh) constants
var DefaultWindow = 10 * time.Millisecond
DefaultWindow is the window used to smooth SimpleLimit rates. That is, SimpleLimit distributes the given quantity evenly into buckets of size t. This is useful for avoiding tcp silly window syndrome and providing predictable resource usage.
var ErrTimeoutExceeded error = errors.New("Timeout Exceeded")
ErrTimeoutExceeded will be returned upon a timeout lapsing without a read occuring
func Distribute(n int, t, w time.Duration) (int, time.Duration)
Distribute takes a rate (n, t) and window (w), evenly distributes the n/t to n'/t' (n'<=n && t'>=w)
type Limiter interface {
Limit(chan int) <-chan bool //The channel is useful for knowing that the channel has been unlimited. The boolean represents finality.
Unlimit()
}
A Limiter is an interface that meters some underlying discretely quantifiable operation with respect to time.
The Limit() function, when implemented, should apply a limit to some underlying operation when called. Supporting concurrency is up to the implementer and as such, should be documented. The semantics of the channel are that of a token bucket. The actual integer sent through the channel represents a quantity of operations that can take place. The implementation should be sure to specify its interpretation of the quantity.
Limit() returns a new boolean channel, used to comunicate that the given chan int
is no longer being used and may be closed. A false value indicates that the
Limiter has not been shut down and may still be acted upon. True indicates that
the limiter has been shutdown and any further function calls will have no
effect.
Unlimit() removes any formerly imposed limits and allows the underlying operation.
type Manager interface {
Limiter
Manage(Limiter) error
Unmanage(Limiter)
}
A Manager enables consumers to treat a group of Limiters as a single Limiter, enabling hierarchies of limiters. For example, a network interface could have a global limit that is distributed across connections, each of which can manage their own distribution of the bandwidth they are allocated.
type Reader struct {
}
Reader implements an io-limited reader that conforms to the io.Reader and limio.Limiter interface. Reader can have its limits updated concurrently with any Read() calls.
func NewReader(r io.Reader) *Reader
NewReader takes any io.Reader and returns a limio.Reader.
func (r *Reader) Close() error
Close allows the goroutines that were managing limits and reads to shut down and free up memory. It should be called by any clients of the limio.Reader, much as http.Response.Body should be closed to free up system resources.
func (r *Reader) Limit(lch chan int) <-chan bool
Limit can be used to precisely control the limit at which bytes can be Read, whether burstily or not.
func (r *Reader) Read(p []byte) (written int, err error)
Read implements io.Reader in a blocking manner according to the limits of the limio.Reader.
func (r *Reader) SetTimeout(t time.Duration) error
SetTimeout takes some time.Duration t and configures the underlying Reader to return a limio.TimedOut error if the timeout is exceeded while waiting for a read operation.
func (r *Reader) SimpleLimit(n int, t time.Duration) <-chan bool
SimpleLimit takes an integer and a time.Duration and limits the underlying reader non-burstily (given rate is averaged over a small time).
func (r *Reader) Unlimit()
Unlimit removes any restrictions on the underlying io.Reader.
type SimpleManager struct {
}
A SimpleManager is an implementation of the limio.Manager interface. It allows simple rate-based and arbitrary channel-based limits to be set.
A SimpleManager is designed so that Limit and Manage may be called concurrently.
func NewSimpleManager() *SimpleManager
NewSimpleManager creates and initializes a SimpleManager.
func (lm *SimpleManager) Close() error
Close allows the SimpleManager to free any resources it is using if the consumer has no further need for the SimpleManager.
func (lm *SimpleManager) Limit(l chan int) <-chan bool
Limit implements the limio.Limiter interface.
func (lm *SimpleManager) Manage(l Limiter) error
Manage takes a Limiter that will be adopted under the management policy of the SimpleManager.
func (lm *SimpleManager) NewReader(r io.Reader) *Reader
NewReader takes an io.Reader and Limits it according to its limit policy/strategy
func (lm *SimpleManager) SimpleLimit(n int, t time.Duration) <-chan bool
SimpleLimit takes an int and time.Duration that will be distributed evenly across all managed Limiters.
func (lm *SimpleManager) Unlimit()
Unlimit implements the limio.Limiter interface.
func (lm *SimpleManager) Unmanage(l Limiter)
Unmanage allows consumers to remove a specific Limiter from its management strategy