diff --git a/metrics-server/vendor/gopkg.in/inf.v0/LICENSE b/metrics-server/vendor/gopkg.in/inf.v0/LICENSE
new file mode 100644
index 0000000..87a5ced
--- /dev/null
+++ b/metrics-server/vendor/gopkg.in/inf.v0/LICENSE
@@ -0,0 +1,28 @@
+Copyright (c) 2012 Péter Surányi. Portions Copyright (c) 2009 The Go
+Authors. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   * Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+   * Redistributions in binary form must reproduce the above
+copyright notice, this list of conditions and the following disclaimer
+in the documentation and/or other materials provided with the
+distribution.
+   * Neither the name of Google Inc. nor the names of its
+contributors may be used to endorse or promote products derived from
+this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/metrics-server/vendor/gopkg.in/inf.v0/dec.go b/metrics-server/vendor/gopkg.in/inf.v0/dec.go
new file mode 100644
index 0000000..26548b6
--- /dev/null
+++ b/metrics-server/vendor/gopkg.in/inf.v0/dec.go
@@ -0,0 +1,615 @@
+// Package inf (type inf.Dec) implements "infinite-precision" decimal
+// arithmetic.
+// "Infinite precision" describes two characteristics: practically unlimited
+// precision for decimal number representation and no support for calculating
+// with any specific fixed precision.
+// (Although there is no practical limit on precision, inf.Dec can only
+// represent finite decimals.)
+//
+// This package is currently in experimental stage and the API may change.
+//
+// This package does NOT support:
+//  - rounding to specific precisions (as opposed to specific decimal positions)
+//  - the notion of context (each rounding must be explicit)
+//  - NaN and Inf values, and distinguishing between positive and negative zero
+//  - conversions to and from float32/64 types
+//
+// Features considered for possible addition:
+//  + formatting options
+//  + Exp method
+//  + combined operations such as AddRound/MulAdd etc
+//  + exchanging data in decimal32/64/128 formats
+//
+package inf // import "gopkg.in/inf.v0"
+
+// TODO:
+//  - avoid excessive deep copying (quo and rounders)
+
+import (
+	"fmt"
+	"io"
+	"math/big"
+	"strings"
+)
+
+// A Dec represents a signed arbitrary-precision decimal.
+// It is a combination of a sign, an arbitrary-precision integer coefficient
+// value, and a signed fixed-precision exponent value.
+// The sign and the coefficient value are handled together as a signed value
+// and referred to as the unscaled value.
+// (Positive and negative zero values are not distinguished.)
+// Since the exponent is most commonly non-positive, it is handled in negated
+// form and referred to as scale.
+//
+// The mathematical value of a Dec equals:
+//
+//  unscaled * 10**(-scale)
+//
+// Note that different Dec representations may have equal mathematical values.
+//
+//  unscaled  scale  String()
+//  -------------------------
+//         0      0    "0"
+//         0      2    "0.00"
+//         0     -2    "0"
+//         1      0    "1"
+//       100      2    "1.00"
+//        10      0   "10"
+//         1     -1   "10"
+//
+// The zero value for a Dec represents the value 0 with scale 0.
+//
+// Operations are typically performed through the *Dec type.
+// The semantics of the assignment operation "=" for "bare" Dec values is
+// undefined and should not be relied on.
+//
+// Methods are typically of the form:
+//
+//	func (z *Dec) Op(x, y *Dec) *Dec
+//
+// and implement operations z = x Op y with the result as receiver; if it
+// is one of the operands it may be overwritten (and its memory reused).
+// To enable chaining of operations, the result is also returned. Methods
+// returning a result other than *Dec take one of the operands as the receiver.
+//
+// A "bare" Quo method (quotient / division operation) is not provided, as the
+// result is not always a finite decimal and thus in general cannot be
+// represented as a Dec.
+// Instead, in the common case when rounding is (potentially) necessary,
+// QuoRound should be used with a Scale and a Rounder.
+// QuoExact or QuoRound with RoundExact can be used in the special cases when it
+// is known that the result is always a finite decimal.
+//
+type Dec struct {
+	unscaled big.Int
+	scale    Scale
+}
+
+// Scale represents the type used for the scale of a Dec.
+type Scale int32
+
+const scaleSize = 4 // bytes in a Scale value
+
+// Scaler represents a method for obtaining the scale to use for the result of
+// an operation on x and y.
+type scaler interface {
+	Scale(x *Dec, y *Dec) Scale
+}
+
+var bigInt = [...]*big.Int{
+	big.NewInt(0), big.NewInt(1), big.NewInt(2), big.NewInt(3), big.NewInt(4),
+	big.NewInt(5), big.NewInt(6), big.NewInt(7), big.NewInt(8), big.NewInt(9),
+	big.NewInt(10),
+}
+
+var exp10cache [64]big.Int = func() [64]big.Int {
+	e10, e10i := [64]big.Int{}, bigInt[1]
+	for i := range e10 {
+		e10[i].Set(e10i)
+		e10i = new(big.Int).Mul(e10i, bigInt[10])
+	}
+	return e10
+}()
+
+// NewDec allocates and returns a new Dec set to the given int64 unscaled value
+// and scale.
+func NewDec(unscaled int64, scale Scale) *Dec {
+	return new(Dec).SetUnscaled(unscaled).SetScale(scale)
+}
+
+// NewDecBig allocates and returns a new Dec set to the given *big.Int unscaled
+// value and scale.
+func NewDecBig(unscaled *big.Int, scale Scale) *Dec {
+	return new(Dec).SetUnscaledBig(unscaled).SetScale(scale)
+}
+
+// Scale returns the scale of x.
+func (x *Dec) Scale() Scale {
+	return x.scale
+}
+
+// Unscaled returns the unscaled value of x for u and true for ok when the
+// unscaled value can be represented as int64; otherwise it returns an undefined
+// int64 value for u and false for ok. Use x.UnscaledBig().Int64() to avoid
+// checking the validity of the value when the check is known to be redundant.
+func (x *Dec) Unscaled() (u int64, ok bool) {
+	u = x.unscaled.Int64()
+	var i big.Int
+	ok = i.SetInt64(u).Cmp(&x.unscaled) == 0
+	return
+}
+
+// UnscaledBig returns the unscaled value of x as *big.Int.
+func (x *Dec) UnscaledBig() *big.Int {
+	return &x.unscaled
+}
+
+// SetScale sets the scale of z, with the unscaled value unchanged, and returns
+// z.
+// The mathematical value of the Dec changes as if it was multiplied by
+// 10**(oldscale-scale).
+func (z *Dec) SetScale(scale Scale) *Dec {
+	z.scale = scale
+	return z
+}
+
+// SetUnscaled sets the unscaled value of z, with the scale unchanged, and
+// returns z.
+func (z *Dec) SetUnscaled(unscaled int64) *Dec {
+	z.unscaled.SetInt64(unscaled)
+	return z
+}
+
+// SetUnscaledBig sets the unscaled value of z, with the scale unchanged, and
+// returns z.
+func (z *Dec) SetUnscaledBig(unscaled *big.Int) *Dec {
+	z.unscaled.Set(unscaled)
+	return z
+}
+
+// Set sets z to the value of x and returns z.
+// It does nothing if z == x.
+func (z *Dec) Set(x *Dec) *Dec {
+	if z != x {
+		z.SetUnscaledBig(x.UnscaledBig())
+		z.SetScale(x.Scale())
+	}
+	return z
+}
+
+// Sign returns:
+//
+//	-1 if x <  0
+//	 0 if x == 0
+//	+1 if x >  0
+//
+func (x *Dec) Sign() int {
+	return x.UnscaledBig().Sign()
+}
+
+// Neg sets z to -x and returns z.
+func (z *Dec) Neg(x *Dec) *Dec {
+	z.SetScale(x.Scale())
+	z.UnscaledBig().Neg(x.UnscaledBig())
+	return z
+}
+
+// Cmp compares x and y and returns:
+//
+//   -1 if x <  y
+//    0 if x == y
+//   +1 if x >  y
+//
+func (x *Dec) Cmp(y *Dec) int {
+	xx, yy := upscale(x, y)
+	return xx.UnscaledBig().Cmp(yy.UnscaledBig())
+}
+
+// Abs sets z to |x| (the absolute value of x) and returns z.
+func (z *Dec) Abs(x *Dec) *Dec {
+	z.SetScale(x.Scale())
+	z.UnscaledBig().Abs(x.UnscaledBig())
+	return z
+}
+
+// Add sets z to the sum x+y and returns z.
+// The scale of z is the greater of the scales of x and y.
+func (z *Dec) Add(x, y *Dec) *Dec {
+	xx, yy := upscale(x, y)
+	z.SetScale(xx.Scale())
+	z.UnscaledBig().Add(xx.UnscaledBig(), yy.UnscaledBig())
+	return z
+}
+
+// Sub sets z to the difference x-y and returns z.
+// The scale of z is the greater of the scales of x and y.
+func (z *Dec) Sub(x, y *Dec) *Dec {
+	xx, yy := upscale(x, y)
+	z.SetScale(xx.Scale())
+	z.UnscaledBig().Sub(xx.UnscaledBig(), yy.UnscaledBig())
+	return z
+}
+
+// Mul sets z to the product x*y and returns z.
+// The scale of z is the sum of the scales of x and y.
+func (z *Dec) Mul(x, y *Dec) *Dec {
+	z.SetScale(x.Scale() + y.Scale())
+	z.UnscaledBig().Mul(x.UnscaledBig(), y.UnscaledBig())
+	return z
+}
+
+// Round sets z to the value of x rounded to Scale s using Rounder r, and
+// returns z.
+func (z *Dec) Round(x *Dec, s Scale, r Rounder) *Dec {
+	return z.QuoRound(x, NewDec(1, 0), s, r)
+}
+
+// QuoRound sets z to the quotient x/y, rounded using the given Rounder to the
+// specified scale.
+//
+// If the rounder is RoundExact but the result can not be expressed exactly at
+// the specified scale, QuoRound returns nil, and the value of z is undefined.
+//
+// There is no corresponding Div method; the equivalent can be achieved through
+// the choice of Rounder used.
+//
+func (z *Dec) QuoRound(x, y *Dec, s Scale, r Rounder) *Dec {
+	return z.quo(x, y, sclr{s}, r)
+}
+
+func (z *Dec) quo(x, y *Dec, s scaler, r Rounder) *Dec {
+	scl := s.Scale(x, y)
+	var zzz *Dec
+	if r.UseRemainder() {
+		zz, rA, rB := new(Dec).quoRem(x, y, scl, true, new(big.Int), new(big.Int))
+		zzz = r.Round(new(Dec), zz, rA, rB)
+	} else {
+		zz, _, _ := new(Dec).quoRem(x, y, scl, false, nil, nil)
+		zzz = r.Round(new(Dec), zz, nil, nil)
+	}
+	if zzz == nil {
+		return nil
+	}
+	return z.Set(zzz)
+}
+
+// QuoExact sets z to the quotient x/y and returns z when x/y is a finite
+// decimal. Otherwise it returns nil and the value of z is undefined.
+//
+// The scale of a non-nil result is "x.Scale() - y.Scale()" or greater; it is
+// calculated so that the remainder will be zero whenever x/y is a finite
+// decimal.
+func (z *Dec) QuoExact(x, y *Dec) *Dec {
+	return z.quo(x, y, scaleQuoExact{}, RoundExact)
+}
+
+// quoRem sets z to the quotient x/y with the scale s, and if useRem is true,
+// it sets remNum and remDen to the numerator and denominator of the remainder.
+// It returns z, remNum and remDen.
+//
+// The remainder is normalized to the range -1 < r < 1 to simplify rounding;
+// that is, the results satisfy the following equation:
+//
+//  x / y = z + (remNum/remDen) * 10**(-z.Scale())
+//
+// See Rounder for more details about rounding.
+//
+func (z *Dec) quoRem(x, y *Dec, s Scale, useRem bool,
+	remNum, remDen *big.Int) (*Dec, *big.Int, *big.Int) {
+	// difference (required adjustment) compared to "canonical" result scale
+	shift := s - (x.Scale() - y.Scale())
+	// pointers to adjusted unscaled dividend and divisor
+	var ix, iy *big.Int
+	switch {
+	case shift > 0:
+		// increased scale: decimal-shift dividend left
+		ix = new(big.Int).Mul(x.UnscaledBig(), exp10(shift))
+		iy = y.UnscaledBig()
+	case shift < 0:
+		// decreased scale: decimal-shift divisor left
+		ix = x.UnscaledBig()
+		iy = new(big.Int).Mul(y.UnscaledBig(), exp10(-shift))
+	default:
+		ix = x.UnscaledBig()
+		iy = y.UnscaledBig()
+	}
+	// save a copy of iy in case it to be overwritten with the result
+	iy2 := iy
+	if iy == z.UnscaledBig() {
+		iy2 = new(big.Int).Set(iy)
+	}
+	// set scale
+	z.SetScale(s)
+	// set unscaled
+	if useRem {
+		// Int division
+		_, intr := z.UnscaledBig().QuoRem(ix, iy, new(big.Int))
+		// set remainder
+		remNum.Set(intr)
+		remDen.Set(iy2)
+	} else {
+		z.UnscaledBig().Quo(ix, iy)
+	}
+	return z, remNum, remDen
+}
+
+type sclr struct{ s Scale }
+
+func (s sclr) Scale(x, y *Dec) Scale {
+	return s.s
+}
+
+type scaleQuoExact struct{}
+
+func (sqe scaleQuoExact) Scale(x, y *Dec) Scale {
+	rem := new(big.Rat).SetFrac(x.UnscaledBig(), y.UnscaledBig())
+	f2, f5 := factor2(rem.Denom()), factor(rem.Denom(), bigInt[5])
+	var f10 Scale
+	if f2 > f5 {
+		f10 = Scale(f2)
+	} else {
+		f10 = Scale(f5)
+	}
+	return x.Scale() - y.Scale() + f10
+}
+
+func factor(n *big.Int, p *big.Int) int {
+	// could be improved for large factors
+	d, f := n, 0
+	for {
+		dd, dm := new(big.Int).DivMod(d, p, new(big.Int))
+		if dm.Sign() == 0 {
+			f++
+			d = dd
+		} else {
+			break
+		}
+	}
+	return f
+}
+
+func factor2(n *big.Int) int {
+	// could be improved for large factors
+	f := 0
+	for ; n.Bit(f) == 0; f++ {
+	}
+	return f
+}
+
+func upscale(a, b *Dec) (*Dec, *Dec) {
+	if a.Scale() == b.Scale() {
+		return a, b
+	}
+	if a.Scale() > b.Scale() {
+		bb := b.rescale(a.Scale())
+		return a, bb
+	}
+	aa := a.rescale(b.Scale())
+	return aa, b
+}
+
+func exp10(x Scale) *big.Int {
+	if int(x) < len(exp10cache) {
+		return &exp10cache[int(x)]
+	}
+	return new(big.Int).Exp(bigInt[10], big.NewInt(int64(x)), nil)
+}
+
+func (x *Dec) rescale(newScale Scale) *Dec {
+	shift := newScale - x.Scale()
+	switch {
+	case shift < 0:
+		e := exp10(-shift)
+		return NewDecBig(new(big.Int).Quo(x.UnscaledBig(), e), newScale)
+	case shift > 0:
+		e := exp10(shift)
+		return NewDecBig(new(big.Int).Mul(x.UnscaledBig(), e), newScale)
+	}
+	return x
+}
+
+var zeros = []byte("00000000000000000000000000000000" +
+	"00000000000000000000000000000000")
+var lzeros = Scale(len(zeros))
+
+func appendZeros(s []byte, n Scale) []byte {
+	for i := Scale(0); i < n; i += lzeros {
+		if n > i+lzeros {
+			s = append(s, zeros...)
+		} else {
+			s = append(s, zeros[0:n-i]...)
+		}
+	}
+	return s
+}
+
+func (x *Dec) String() string {
+	if x == nil {
+		return "<nil>"
+	}
+	scale := x.Scale()
+	s := []byte(x.UnscaledBig().String())
+	if scale <= 0 {
+		if scale != 0 && x.unscaled.Sign() != 0 {
+			s = appendZeros(s, -scale)
+		}
+		return string(s)
+	}
+	negbit := Scale(-((x.Sign() - 1) / 2))
+	// scale > 0
+	lens := Scale(len(s))
+	if lens-negbit <= scale {
+		ss := make([]byte, 0, scale+2)
+		if negbit == 1 {
+			ss = append(ss, '-')
+		}
+		ss = append(ss, '0', '.')
+		ss = appendZeros(ss, scale-lens+negbit)
+		ss = append(ss, s[negbit:]...)
+		return string(ss)
+	}
+	// lens > scale
+	ss := make([]byte, 0, lens+1)
+	ss = append(ss, s[:lens-scale]...)
+	ss = append(ss, '.')
+	ss = append(ss, s[lens-scale:]...)
+	return string(ss)
+}
+
+// Format is a support routine for fmt.Formatter. It accepts the decimal
+// formats 'd' and 'f', and handles both equivalently.
+// Width, precision, flags and bases 2, 8, 16 are not supported.
+func (x *Dec) Format(s fmt.State, ch rune) {
+	if ch != 'd' && ch != 'f' && ch != 'v' && ch != 's' {
+		fmt.Fprintf(s, "%%!%c(dec.Dec=%s)", ch, x.String())
+		return
+	}
+	fmt.Fprintf(s, x.String())
+}
+
+func (z *Dec) scan(r io.RuneScanner) (*Dec, error) {
+	unscaled := make([]byte, 0, 256) // collects chars of unscaled as bytes
+	dp, dg := -1, -1                 // indexes of decimal point, first digit
+loop:
+	for {
+		ch, _, err := r.ReadRune()
+		if err == io.EOF {
+			break loop
+		}
+		if err != nil {
+			return nil, err
+		}
+		switch {
+		case ch == '+' || ch == '-':
+			if len(unscaled) > 0 || dp >= 0 { // must be first character
+				r.UnreadRune()
+				break loop
+			}
+		case ch == '.':
+			if dp >= 0 {
+				r.UnreadRune()
+				break loop
+			}
+			dp = len(unscaled)
+			continue // don't add to unscaled
+		case ch >= '0' && ch <= '9':
+			if dg == -1 {
+				dg = len(unscaled)
+			}
+		default:
+			r.UnreadRune()
+			break loop
+		}
+		unscaled = append(unscaled, byte(ch))
+	}
+	if dg == -1 {
+		return nil, fmt.Errorf("no digits read")
+	}
+	if dp >= 0 {
+		z.SetScale(Scale(len(unscaled) - dp))
+	} else {
+		z.SetScale(0)
+	}
+	_, ok := z.UnscaledBig().SetString(string(unscaled), 10)
+	if !ok {
+		return nil, fmt.Errorf("invalid decimal: %s", string(unscaled))
+	}
+	return z, nil
+}
+
+// SetString sets z to the value of s, interpreted as a decimal (base 10),
+// and returns z and a boolean indicating success. The scale of z is the
+// number of digits after the decimal point (including any trailing 0s),
+// or 0 if there is no decimal point. If SetString fails, the value of z
+// is undefined but the returned value is nil.
+func (z *Dec) SetString(s string) (*Dec, bool) {
+	r := strings.NewReader(s)
+	_, err := z.scan(r)
+	if err != nil {
+		return nil, false
+	}
+	_, _, err = r.ReadRune()
+	if err != io.EOF {
+		return nil, false
+	}
+	// err == io.EOF => scan consumed all of s
+	return z, true
+}
+
+// Scan is a support routine for fmt.Scanner; it sets z to the value of
+// the scanned number. It accepts the decimal formats 'd' and 'f', and
+// handles both equivalently. Bases 2, 8, 16 are not supported.
+// The scale of z is the number of digits after the decimal point
+// (including any trailing 0s), or 0 if there is no decimal point.
+func (z *Dec) Scan(s fmt.ScanState, ch rune) error {
+	if ch != 'd' && ch != 'f' && ch != 's' && ch != 'v' {
+		return fmt.Errorf("Dec.Scan: invalid verb '%c'", ch)
+	}
+	s.SkipSpace()
+	_, err := z.scan(s)
+	return err
+}
+
+// Gob encoding version
+const decGobVersion byte = 1
+
+func scaleBytes(s Scale) []byte {
+	buf := make([]byte, scaleSize)
+	i := scaleSize
+	for j := 0; j < scaleSize; j++ {
+		i--
+		buf[i] = byte(s)
+		s >>= 8
+	}
+	return buf
+}
+
+func scale(b []byte) (s Scale) {
+	for j := 0; j < scaleSize; j++ {
+		s <<= 8
+		s |= Scale(b[j])
+	}
+	return
+}
+
+// GobEncode implements the gob.GobEncoder interface.
+func (x *Dec) GobEncode() ([]byte, error) {
+	buf, err := x.UnscaledBig().GobEncode()
+	if err != nil {
+		return nil, err
+	}
+	buf = append(append(buf, scaleBytes(x.Scale())...), decGobVersion)
+	return buf, nil
+}
+
+// GobDecode implements the gob.GobDecoder interface.
+func (z *Dec) GobDecode(buf []byte) error {
+	if len(buf) == 0 {
+		return fmt.Errorf("Dec.GobDecode: no data")
+	}
+	b := buf[len(buf)-1]
+	if b != decGobVersion {
+		return fmt.Errorf("Dec.GobDecode: encoding version %d not supported", b)
+	}
+	l := len(buf) - scaleSize - 1
+	err := z.UnscaledBig().GobDecode(buf[:l])
+	if err != nil {
+		return err
+	}
+	z.SetScale(scale(buf[l : l+scaleSize]))
+	return nil
+}
+
+// MarshalText implements the encoding.TextMarshaler interface.
+func (x *Dec) MarshalText() ([]byte, error) {
+	return []byte(x.String()), nil
+}
+
+// UnmarshalText implements the encoding.TextUnmarshaler interface.
+func (z *Dec) UnmarshalText(data []byte) error {
+	_, ok := z.SetString(string(data))
+	if !ok {
+		return fmt.Errorf("invalid inf.Dec")
+	}
+	return nil
+}
diff --git a/metrics-server/vendor/gopkg.in/inf.v0/rounder.go b/metrics-server/vendor/gopkg.in/inf.v0/rounder.go
new file mode 100644
index 0000000..3a97ef5
--- /dev/null
+++ b/metrics-server/vendor/gopkg.in/inf.v0/rounder.go
@@ -0,0 +1,145 @@
+package inf
+
+import (
+	"math/big"
+)
+
+// Rounder represents a method for rounding the (possibly infinite decimal)
+// result of a division to a finite Dec. It is used by Dec.Round() and
+// Dec.Quo().
+//
+// See the Example for results of using each Rounder with some sample values.
+//
+type Rounder rounder
+
+// See http://speleotrove.com/decimal/damodel.html#refround for more detailed
+// definitions of these rounding modes.
+var (
+	RoundDown     Rounder // towards 0
+	RoundUp       Rounder // away from 0
+	RoundFloor    Rounder // towards -infinity
+	RoundCeil     Rounder // towards +infinity
+	RoundHalfDown Rounder // to nearest; towards 0 if same distance
+	RoundHalfUp   Rounder // to nearest; away from 0 if same distance
+	RoundHalfEven Rounder // to nearest; even last digit if same distance
+)
+
+// RoundExact is to be used in the case when rounding is not necessary.
+// When used with Quo or Round, it returns the result verbatim when it can be
+// expressed exactly with the given precision, and it returns nil otherwise.
+// QuoExact is a shorthand for using Quo with RoundExact.
+var RoundExact Rounder
+
+type rounder interface {
+
+	// When UseRemainder() returns true, the Round() method is passed the
+	// remainder of the division, expressed as the numerator and denominator of
+	// a rational.
+	UseRemainder() bool
+
+	// Round sets the rounded value of a quotient to z, and returns z.
+	// quo is rounded down (truncated towards zero) to the scale obtained from
+	// the Scaler in Quo().
+	//
+	// When the remainder is not used, remNum and remDen are nil.
+	// When used, the remainder is normalized between -1 and 1; that is:
+	//
+	//  -|remDen| < remNum < |remDen|
+	//
+	// remDen has the same sign as y, and remNum is zero or has the same sign
+	// as x.
+	Round(z, quo *Dec, remNum, remDen *big.Int) *Dec
+}
+
+type rndr struct {
+	useRem bool
+	round  func(z, quo *Dec, remNum, remDen *big.Int) *Dec
+}
+
+func (r rndr) UseRemainder() bool {
+	return r.useRem
+}
+
+func (r rndr) Round(z, quo *Dec, remNum, remDen *big.Int) *Dec {
+	return r.round(z, quo, remNum, remDen)
+}
+
+var intSign = []*big.Int{big.NewInt(-1), big.NewInt(0), big.NewInt(1)}
+
+func roundHalf(f func(c int, odd uint) (roundUp bool)) func(z, q *Dec, rA, rB *big.Int) *Dec {
+	return func(z, q *Dec, rA, rB *big.Int) *Dec {
+		z.Set(q)
+		brA, brB := rA.BitLen(), rB.BitLen()
+		if brA < brB-1 {
+			// brA < brB-1 => |rA| < |rB/2|
+			return z
+		}
+		roundUp := false
+		srA, srB := rA.Sign(), rB.Sign()
+		s := srA * srB
+		if brA == brB-1 {
+			rA2 := new(big.Int).Lsh(rA, 1)
+			if s < 0 {
+				rA2.Neg(rA2)
+			}
+			roundUp = f(rA2.Cmp(rB)*srB, z.UnscaledBig().Bit(0))
+		} else {
+			// brA > brB-1 => |rA| > |rB/2|
+			roundUp = true
+		}
+		if roundUp {
+			z.UnscaledBig().Add(z.UnscaledBig(), intSign[s+1])
+		}
+		return z
+	}
+}
+
+func init() {
+	RoundExact = rndr{true,
+		func(z, q *Dec, rA, rB *big.Int) *Dec {
+			if rA.Sign() != 0 {
+				return nil
+			}
+			return z.Set(q)
+		}}
+	RoundDown = rndr{false,
+		func(z, q *Dec, rA, rB *big.Int) *Dec {
+			return z.Set(q)
+		}}
+	RoundUp = rndr{true,
+		func(z, q *Dec, rA, rB *big.Int) *Dec {
+			z.Set(q)
+			if rA.Sign() != 0 {
+				z.UnscaledBig().Add(z.UnscaledBig(), intSign[rA.Sign()*rB.Sign()+1])
+			}
+			return z
+		}}
+	RoundFloor = rndr{true,
+		func(z, q *Dec, rA, rB *big.Int) *Dec {
+			z.Set(q)
+			if rA.Sign()*rB.Sign() < 0 {
+				z.UnscaledBig().Add(z.UnscaledBig(), intSign[0])
+			}
+			return z
+		}}
+	RoundCeil = rndr{true,
+		func(z, q *Dec, rA, rB *big.Int) *Dec {
+			z.Set(q)
+			if rA.Sign()*rB.Sign() > 0 {
+				z.UnscaledBig().Add(z.UnscaledBig(), intSign[2])
+			}
+			return z
+		}}
+	RoundHalfDown = rndr{true, roundHalf(
+		func(c int, odd uint) bool {
+			return c > 0
+		})}
+	RoundHalfUp = rndr{true, roundHalf(
+		func(c int, odd uint) bool {
+			return c >= 0
+		})}
+	RoundHalfEven = rndr{true, roundHalf(
+		func(c int, odd uint) bool {
+			return c > 0 || c == 0 && odd == 1
+		})}
+}