You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
tailscale/util/deephash/deephash_test.go

597 lines
15 KiB
Go

// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package deephash
import (
"archive/tar"
"bufio"
"bytes"
"crypto/sha256"
"fmt"
"io"
"math"
"math/rand"
"reflect"
"runtime"
"testing"
"testing/quick"
"time"
"unsafe"
"go4.org/mem"
"inet.af/netaddr"
"tailscale.com/tailcfg"
"tailscale.com/types/dnstype"
"tailscale.com/types/ipproto"
"tailscale.com/types/key"
"tailscale.com/types/structs"
"tailscale.com/util/dnsname"
"tailscale.com/version"
"tailscale.com/wgengine/filter"
"tailscale.com/wgengine/router"
"tailscale.com/wgengine/wgcfg"
)
type appendBytes []byte
func (p appendBytes) AppendTo(b []byte) []byte {
return append(b, p...)
}
func TestHash(t *testing.T) {
type tuple [2]any
type iface struct{ X any }
type scalars struct {
I8 int8
I16 int16
I32 int32
I64 int64
I int
U8 uint8
U16 uint16
U32 uint32
U64 uint64
U uint
UP uintptr
F32 float32
F64 float64
C64 complex64
C128 complex128
}
type MyBool bool
type MyHeader tar.Header
var zeroFloat64 float64
tests := []struct {
in tuple
wantEq bool
}{
{in: tuple{false, true}, wantEq: false},
{in: tuple{true, true}, wantEq: true},
{in: tuple{false, false}, wantEq: true},
{
in: tuple{
scalars{-8, -16, -32, -64, -1234, 8, 16, 32, 64, 1234, 5678, 32.32, 64.64, 32 + 32i, 64 + 64i},
scalars{-8, -16, -32, -64, -1234, 8, 16, 32, 64, 1234, 5678, 32.32, 64.64, 32 + 32i, 64 + 64i},
},
wantEq: true,
},
{in: tuple{scalars{I8: math.MinInt8}, scalars{I8: math.MinInt8 / 2}}, wantEq: false},
{in: tuple{scalars{I16: math.MinInt16}, scalars{I16: math.MinInt16 / 2}}, wantEq: false},
{in: tuple{scalars{I32: math.MinInt32}, scalars{I32: math.MinInt32 / 2}}, wantEq: false},
{in: tuple{scalars{I64: math.MinInt64}, scalars{I64: math.MinInt64 / 2}}, wantEq: false},
{in: tuple{scalars{I: -1234}, scalars{I: -1234 / 2}}, wantEq: false},
{in: tuple{scalars{U8: math.MaxUint8}, scalars{U8: math.MaxUint8 / 2}}, wantEq: false},
{in: tuple{scalars{U16: math.MaxUint16}, scalars{U16: math.MaxUint16 / 2}}, wantEq: false},
{in: tuple{scalars{U32: math.MaxUint32}, scalars{U32: math.MaxUint32 / 2}}, wantEq: false},
{in: tuple{scalars{U64: math.MaxUint64}, scalars{U64: math.MaxUint64 / 2}}, wantEq: false},
{in: tuple{scalars{U: 1234}, scalars{U: 1234 / 2}}, wantEq: false},
{in: tuple{scalars{UP: 5678}, scalars{UP: 5678 / 2}}, wantEq: false},
{in: tuple{scalars{F32: 32.32}, scalars{F32: math.Nextafter32(32.32, 0)}}, wantEq: false},
{in: tuple{scalars{F64: 64.64}, scalars{F64: math.Nextafter(64.64, 0)}}, wantEq: false},
{in: tuple{scalars{F32: float32(math.NaN())}, scalars{F32: float32(math.NaN())}}, wantEq: true},
{in: tuple{scalars{F64: float64(math.NaN())}, scalars{F64: float64(math.NaN())}}, wantEq: true},
{in: tuple{scalars{C64: 32 + 32i}, scalars{C64: complex(math.Nextafter32(32, 0), 32)}}, wantEq: false},
{in: tuple{scalars{C128: 64 + 64i}, scalars{C128: complex(math.Nextafter(64, 0), 64)}}, wantEq: false},
{in: tuple{[]appendBytes{{}, {0, 0, 0, 0, 0, 0, 0, 1}}, []appendBytes{{}, {0, 0, 0, 0, 0, 0, 0, 1}}}, wantEq: true},
{in: tuple{[]appendBytes{{}, {0, 0, 0, 0, 0, 0, 0, 1}}, []appendBytes{{0, 0, 0, 0, 0, 0, 0, 1}, {}}}, wantEq: false},
{in: tuple{iface{MyBool(true)}, iface{MyBool(true)}}, wantEq: true},
{in: tuple{iface{true}, iface{MyBool(true)}}, wantEq: false},
{in: tuple{iface{MyHeader{}}, iface{MyHeader{}}}, wantEq: true},
{in: tuple{iface{MyHeader{}}, iface{tar.Header{}}}, wantEq: false},
{in: tuple{iface{&MyHeader{}}, iface{&MyHeader{}}}, wantEq: true},
{in: tuple{iface{&MyHeader{}}, iface{&tar.Header{}}}, wantEq: false},
{in: tuple{iface{[]map[string]MyBool{}}, iface{[]map[string]MyBool{}}}, wantEq: true},
{in: tuple{iface{[]map[string]bool{}}, iface{[]map[string]MyBool{}}}, wantEq: false},
{in: tuple{zeroFloat64, -zeroFloat64}, wantEq: false}, // Issue 4883 (false alarm)
{in: tuple{[]any(nil), 0.0}, wantEq: false}, // Issue 4883
{in: tuple{[]any(nil), uint8(0)}, wantEq: false}, // Issue 4883
{in: tuple{nil, nil}, wantEq: true}, // Issue 4883
util/deephash: improve cycle detection (#2470) The previous algorithm used a map of all visited pointers. The strength of this approach is that it quickly prunes any nodes that we have ever visited before. The detriment of the approach is that pruning is heavily dependent on the order that pointers were visited. This is especially relevant for hashing a map where map entries are visited in a non-deterministic manner, which would cause the map hash to be non-deterministic (which defeats the point of a hash). This new algorithm uses a stack of all visited pointers, similar to how github.com/google/go-cmp performs cycle detection. When we visit a pointer, we push it onto the stack, and when we leave a pointer, we pop it from the stack. Before visiting a pointer, we first check whether the pointer exists anywhere in the stack. If yes, then we prune the node. The detriment of this approach is that we may hash a node more often than before since we do not prune as aggressively. The set of visited pointers up until any node is only the path of nodes up to that node and not any other pointers that may have been visited elsewhere. This provides us deterministic hashing regardless of visit order. We can now delete hashMapFallback and associated complexity, which only exists because the previous approach was non-deterministic in the presence of cycles. This fixes a failure of the old algorithm where obviously different values are treated as equal because the pruning was too aggresive. See https://github.com/tailscale/tailscale/issues/2443#issuecomment-883653534 The new algorithm is slightly slower since it prunes less aggresively: name old time/op new time/op delta Hash-8 66.1µs ± 1% 68.8µs ± 1% +4.09% (p=0.000 n=19+19) HashMapAcyclic-8 63.0µs ± 1% 62.5µs ± 1% -0.76% (p=0.000 n=18+19) TailcfgNode-8 9.79µs ± 2% 9.88µs ± 1% +0.95% (p=0.000 n=19+17) HashArray-8 643ns ± 1% 653ns ± 1% +1.64% (p=0.000 n=19+19) However, a slower but more correct algorithm seems more favorable than a faster but incorrect algorithm. Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
{
in: func() tuple {
i1 := 1
i2 := 2
v1 := [3]*int{&i1, &i2, &i1}
v2 := [3]*int{&i1, &i2, &i2}
return tuple{v1, v2}
}(),
wantEq: false,
},
}
for _, tt := range tests {
gotEq := Hash(tt.in[0]) == Hash(tt.in[1])
if gotEq != tt.wantEq {
t.Errorf("(Hash(%T %v) == Hash(%T %v)) = %v, want %v", tt.in[0], tt.in[0], tt.in[1], tt.in[1], gotEq, tt.wantEq)
}
}
}
func TestDeepHash(t *testing.T) {
// v contains the types of values we care about for our current callers.
// Mostly we're just testing that we don't panic on handled types.
v := getVal()
hash1 := Hash(v)
t.Logf("hash: %v", hash1)
for i := 0; i < 20; i++ {
hash2 := Hash(getVal())
if hash1 != hash2 {
t.Error("second hash didn't match")
}
}
}
// Tests that we actually hash map elements. Whoops.
func TestIssue4868(t *testing.T) {
m1 := map[int]string{1: "foo"}
m2 := map[int]string{1: "bar"}
if Hash(m1) == Hash(m2) {
t.Error("bogus")
}
}
func TestIssue4871(t *testing.T) {
m1 := map[string]string{"": "", "x": "foo"}
m2 := map[string]string{}
if h1, h2 := Hash(m1), Hash(m2); h1 == h2 {
t.Errorf("bogus: h1=%x, h2=%x", h1, h2)
}
}
func TestNilVsEmptymap(t *testing.T) {
m1 := map[string]string(nil)
m2 := map[string]string{}
if h1, h2 := Hash(m1), Hash(m2); h1 == h2 {
t.Errorf("bogus: h1=%x, h2=%x", h1, h2)
}
}
func TestMapFraming(t *testing.T) {
m1 := map[string]string{"foo": "", "fo": "o"}
m2 := map[string]string{}
if h1, h2 := Hash(m1), Hash(m2); h1 == h2 {
t.Errorf("bogus: h1=%x, h2=%x", h1, h2)
}
}
func TestQuick(t *testing.T) {
initSeed()
err := quick.Check(func(v, w map[string]string) bool {
return (Hash(v) == Hash(w)) == reflect.DeepEqual(v, w)
}, &quick.Config{MaxCount: 1000, Rand: rand.New(rand.NewSource(int64(seed)))})
if err != nil {
t.Fatalf("seed=%v, err=%v", seed, err)
}
}
func getVal() []any {
return []any{
&wgcfg.Config{
Name: "foo",
Addresses: []netaddr.IPPrefix{netaddr.IPPrefixFrom(netaddr.IPFrom16([16]byte{3: 3}), 5)},
Peers: []wgcfg.Peer{
{
PublicKey: key.NodePublic{},
},
},
},
&router.Config{
Routes: []netaddr.IPPrefix{
netaddr.MustParseIPPrefix("1.2.3.0/24"),
netaddr.MustParseIPPrefix("1234::/64"),
},
},
map[dnsname.FQDN][]netaddr.IP{
dnsname.FQDN("a."): {netaddr.MustParseIP("1.2.3.4"), netaddr.MustParseIP("4.3.2.1")},
dnsname.FQDN("b."): {netaddr.MustParseIP("8.8.8.8"), netaddr.MustParseIP("9.9.9.9")},
dnsname.FQDN("c."): {netaddr.MustParseIP("6.6.6.6"), netaddr.MustParseIP("7.7.7.7")},
dnsname.FQDN("d."): {netaddr.MustParseIP("6.7.6.6"), netaddr.MustParseIP("7.7.7.8")},
dnsname.FQDN("e."): {netaddr.MustParseIP("6.8.6.6"), netaddr.MustParseIP("7.7.7.9")},
dnsname.FQDN("f."): {netaddr.MustParseIP("6.9.6.6"), netaddr.MustParseIP("7.7.7.0")},
},
map[dnsname.FQDN][]netaddr.IPPort{
dnsname.FQDN("a."): {netaddr.MustParseIPPort("1.2.3.4:11"), netaddr.MustParseIPPort("4.3.2.1:22")},
dnsname.FQDN("b."): {netaddr.MustParseIPPort("8.8.8.8:11"), netaddr.MustParseIPPort("9.9.9.9:22")},
dnsname.FQDN("c."): {netaddr.MustParseIPPort("8.8.8.8:12"), netaddr.MustParseIPPort("9.9.9.9:23")},
dnsname.FQDN("d."): {netaddr.MustParseIPPort("8.8.8.8:13"), netaddr.MustParseIPPort("9.9.9.9:24")},
dnsname.FQDN("e."): {netaddr.MustParseIPPort("8.8.8.8:14"), netaddr.MustParseIPPort("9.9.9.9:25")},
},
map[key.DiscoPublic]bool{
key.DiscoPublicFromRaw32(mem.B([]byte{1: 1, 31: 0})): true,
key.DiscoPublicFromRaw32(mem.B([]byte{1: 2, 31: 0})): false,
key.DiscoPublicFromRaw32(mem.B([]byte{1: 3, 31: 0})): true,
key.DiscoPublicFromRaw32(mem.B([]byte{1: 4, 31: 0})): false,
},
&tailcfg.MapResponse{
DERPMap: &tailcfg.DERPMap{
Regions: map[int]*tailcfg.DERPRegion{
1: {
RegionID: 1,
RegionCode: "foo",
Nodes: []*tailcfg.DERPNode{
{
Name: "n1",
RegionID: 1,
HostName: "foo.com",
},
{
Name: "n2",
RegionID: 1,
HostName: "bar.com",
},
},
},
},
},
DNSConfig: &tailcfg.DNSConfig{
Resolvers: []*dnstype.Resolver{
{Addr: "10.0.0.1"},
},
},
PacketFilter: []tailcfg.FilterRule{
{
SrcIPs: []string{"1.2.3.4"},
DstPorts: []tailcfg.NetPortRange{
{
IP: "1.2.3.4/32",
Ports: tailcfg.PortRange{First: 1, Last: 2},
},
},
},
},
Peers: []*tailcfg.Node{
{
ID: 1,
},
{
ID: 2,
},
},
UserProfiles: []tailcfg.UserProfile{
{ID: 1, LoginName: "foo@bar.com"},
{ID: 2, LoginName: "bar@foo.com"},
},
},
filter.Match{
IPProto: []ipproto.Proto{1, 2, 3},
},
}
}
func TestTypeIsRecursive(t *testing.T) {
type RecursiveStruct struct {
v *RecursiveStruct
}
type RecursiveChan chan *RecursiveChan
tests := []struct {
val any
want bool
}{
{val: 42, want: false},
{val: "string", want: false},
{val: 1 + 2i, want: false},
{val: struct{}{}, want: false},
{val: (*RecursiveStruct)(nil), want: true},
{val: RecursiveStruct{}, want: true},
{val: time.Unix(0, 0), want: false},
{val: structs.Incomparable{}, want: false}, // ignore its [0]func()
{val: tailcfg.NetPortRange{}, want: false}, // uses structs.Incomparable
{val: (*tailcfg.Node)(nil), want: false},
{val: map[string]bool{}, want: false},
{val: func() {}, want: false},
{val: make(chan int), want: false},
{val: unsafe.Pointer(nil), want: false},
{val: make(RecursiveChan), want: true},
{val: make(chan int), want: false},
}
for _, tt := range tests {
got := typeIsRecursive(reflect.TypeOf(tt.val))
if got != tt.want {
t.Errorf("for type %T: got %v, want %v", tt.val, got, tt.want)
}
}
}
type IntThenByte struct {
i int
b byte
}
type TwoInts struct{ a, b int }
type IntIntByteInt struct {
i1, i2 int32
b byte // padding after
i3 int32
}
func TestCanMemHash(t *testing.T) {
tests := []struct {
val any
want bool
}{
{true, true},
{uint(1), true},
{uint8(1), true},
{uint16(1), true},
{uint32(1), true},
{uint64(1), true},
{uintptr(1), true},
{int(1), true},
{int8(1), true},
{int16(1), true},
{int32(1), true},
{int64(1), true},
{float32(1), true},
{float64(1), true},
{complex64(1), true},
{complex128(1), true},
{[32]byte{}, true},
{func() {}, false},
{make(chan int), false},
{struct{ io.Writer }{nil}, false},
{unsafe.Pointer(nil), false},
{new(int), false},
{TwoInts{}, true},
{[4]TwoInts{}, true},
{IntThenByte{}, false},
{[4]IntThenByte{}, false},
{tailcfg.PortRange{}, true},
{int16(0), true},
{struct {
_ int
_ int
}{}, true},
{struct {
_ int
_ uint8
_ int
}{}, false}, // gap
{
struct {
_ structs.Incomparable // if not last, zero-width
x int
}{},
true,
},
{
struct {
x int
_ structs.Incomparable // zero-width last: has space, can't memhash
}{},
false,
}}
for _, tt := range tests {
got := canMemHash(reflect.TypeOf(tt.val))
if got != tt.want {
t.Errorf("for type %T: got %v, want %v", tt.val, got, tt.want)
}
}
}
var sink = Hash("foo")
func BenchmarkHash(b *testing.B) {
b.ReportAllocs()
v := getVal()
for i := 0; i < b.N; i++ {
sink = Hash(v)
}
}
func ptrTo[T any](v T) *T { return &v }
// filterRules is a packet filter that has both everything populated (in its
// first element) and also a few entries that are the typical shape for regular
// packet filters as sent to clients.
var filterRules = []tailcfg.FilterRule{
{
SrcIPs: []string{"*", "10.1.3.4/32", "10.0.0.0/24"},
SrcBits: []int{1, 2, 3},
DstPorts: []tailcfg.NetPortRange{{
IP: "1.2.3.4/32",
Bits: ptrTo(32),
Ports: tailcfg.PortRange{First: 1, Last: 2},
}},
IPProto: []int{1, 2, 3, 4},
CapGrant: []tailcfg.CapGrant{{
Dsts: []netaddr.IPPrefix{netaddr.MustParseIPPrefix("1.2.3.4/32")},
Caps: []string{"foo"},
}},
},
{
SrcIPs: []string{"foooooooooo"},
DstPorts: []tailcfg.NetPortRange{{
IP: "baaaaaarrrrr",
Ports: tailcfg.PortRange{First: 1, Last: 2},
}},
},
{
SrcIPs: []string{"foooooooooo"},
DstPorts: []tailcfg.NetPortRange{{
IP: "baaaaaarrrrr",
Ports: tailcfg.PortRange{First: 1, Last: 2},
}},
},
{
SrcIPs: []string{"foooooooooo"},
DstPorts: []tailcfg.NetPortRange{{
IP: "baaaaaarrrrr",
Ports: tailcfg.PortRange{First: 1, Last: 2},
}},
},
}
func BenchmarkHashPacketFilter(b *testing.B) {
b.ReportAllocs()
for i := 0; i < b.N; i++ {
sink = Hash(filterRules)
}
}
func TestHashMapAcyclic(t *testing.T) {
m := map[int]string{}
for i := 0; i < 100; i++ {
m[i] = fmt.Sprint(i)
}
got := map[string]bool{}
var buf bytes.Buffer
bw := bufio.NewWriter(&buf)
ti := getTypeInfo(reflect.TypeOf(m))
for i := 0; i < 20; i++ {
v := reflect.ValueOf(m)
buf.Reset()
bw.Reset(&buf)
util/deephash: improve cycle detection (#2470) The previous algorithm used a map of all visited pointers. The strength of this approach is that it quickly prunes any nodes that we have ever visited before. The detriment of the approach is that pruning is heavily dependent on the order that pointers were visited. This is especially relevant for hashing a map where map entries are visited in a non-deterministic manner, which would cause the map hash to be non-deterministic (which defeats the point of a hash). This new algorithm uses a stack of all visited pointers, similar to how github.com/google/go-cmp performs cycle detection. When we visit a pointer, we push it onto the stack, and when we leave a pointer, we pop it from the stack. Before visiting a pointer, we first check whether the pointer exists anywhere in the stack. If yes, then we prune the node. The detriment of this approach is that we may hash a node more often than before since we do not prune as aggressively. The set of visited pointers up until any node is only the path of nodes up to that node and not any other pointers that may have been visited elsewhere. This provides us deterministic hashing regardless of visit order. We can now delete hashMapFallback and associated complexity, which only exists because the previous approach was non-deterministic in the presence of cycles. This fixes a failure of the old algorithm where obviously different values are treated as equal because the pruning was too aggresive. See https://github.com/tailscale/tailscale/issues/2443#issuecomment-883653534 The new algorithm is slightly slower since it prunes less aggresively: name old time/op new time/op delta Hash-8 66.1µs ± 1% 68.8µs ± 1% +4.09% (p=0.000 n=19+19) HashMapAcyclic-8 63.0µs ± 1% 62.5µs ± 1% -0.76% (p=0.000 n=18+19) TailcfgNode-8 9.79µs ± 2% 9.88µs ± 1% +0.95% (p=0.000 n=19+17) HashArray-8 643ns ± 1% 653ns ± 1% +1.64% (p=0.000 n=19+19) However, a slower but more correct algorithm seems more favorable than a faster but incorrect algorithm. Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
h := &hasher{bw: bw}
h.hashMap(v, ti, false)
if got[string(buf.Bytes())] {
continue
}
got[string(buf.Bytes())] = true
}
if len(got) != 1 {
t.Errorf("got %d results; want 1", len(got))
}
}
func TestPrintArray(t *testing.T) {
type T struct {
X [32]byte
}
util/deephash: improve cycle detection (#2470) The previous algorithm used a map of all visited pointers. The strength of this approach is that it quickly prunes any nodes that we have ever visited before. The detriment of the approach is that pruning is heavily dependent on the order that pointers were visited. This is especially relevant for hashing a map where map entries are visited in a non-deterministic manner, which would cause the map hash to be non-deterministic (which defeats the point of a hash). This new algorithm uses a stack of all visited pointers, similar to how github.com/google/go-cmp performs cycle detection. When we visit a pointer, we push it onto the stack, and when we leave a pointer, we pop it from the stack. Before visiting a pointer, we first check whether the pointer exists anywhere in the stack. If yes, then we prune the node. The detriment of this approach is that we may hash a node more often than before since we do not prune as aggressively. The set of visited pointers up until any node is only the path of nodes up to that node and not any other pointers that may have been visited elsewhere. This provides us deterministic hashing regardless of visit order. We can now delete hashMapFallback and associated complexity, which only exists because the previous approach was non-deterministic in the presence of cycles. This fixes a failure of the old algorithm where obviously different values are treated as equal because the pruning was too aggresive. See https://github.com/tailscale/tailscale/issues/2443#issuecomment-883653534 The new algorithm is slightly slower since it prunes less aggresively: name old time/op new time/op delta Hash-8 66.1µs ± 1% 68.8µs ± 1% +4.09% (p=0.000 n=19+19) HashMapAcyclic-8 63.0µs ± 1% 62.5µs ± 1% -0.76% (p=0.000 n=18+19) TailcfgNode-8 9.79µs ± 2% 9.88µs ± 1% +0.95% (p=0.000 n=19+17) HashArray-8 643ns ± 1% 653ns ± 1% +1.64% (p=0.000 n=19+19) However, a slower but more correct algorithm seems more favorable than a faster but incorrect algorithm. Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
x := T{X: [32]byte{1: 1, 31: 31}}
var got bytes.Buffer
bw := bufio.NewWriter(&got)
util/deephash: improve cycle detection (#2470) The previous algorithm used a map of all visited pointers. The strength of this approach is that it quickly prunes any nodes that we have ever visited before. The detriment of the approach is that pruning is heavily dependent on the order that pointers were visited. This is especially relevant for hashing a map where map entries are visited in a non-deterministic manner, which would cause the map hash to be non-deterministic (which defeats the point of a hash). This new algorithm uses a stack of all visited pointers, similar to how github.com/google/go-cmp performs cycle detection. When we visit a pointer, we push it onto the stack, and when we leave a pointer, we pop it from the stack. Before visiting a pointer, we first check whether the pointer exists anywhere in the stack. If yes, then we prune the node. The detriment of this approach is that we may hash a node more often than before since we do not prune as aggressively. The set of visited pointers up until any node is only the path of nodes up to that node and not any other pointers that may have been visited elsewhere. This provides us deterministic hashing regardless of visit order. We can now delete hashMapFallback and associated complexity, which only exists because the previous approach was non-deterministic in the presence of cycles. This fixes a failure of the old algorithm where obviously different values are treated as equal because the pruning was too aggresive. See https://github.com/tailscale/tailscale/issues/2443#issuecomment-883653534 The new algorithm is slightly slower since it prunes less aggresively: name old time/op new time/op delta Hash-8 66.1µs ± 1% 68.8µs ± 1% +4.09% (p=0.000 n=19+19) HashMapAcyclic-8 63.0µs ± 1% 62.5µs ± 1% -0.76% (p=0.000 n=18+19) TailcfgNode-8 9.79µs ± 2% 9.88µs ± 1% +0.95% (p=0.000 n=19+17) HashArray-8 643ns ± 1% 653ns ± 1% +1.64% (p=0.000 n=19+19) However, a slower but more correct algorithm seems more favorable than a faster but incorrect algorithm. Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
h := &hasher{bw: bw}
h.hashValue(reflect.ValueOf(x), false)
bw.Flush()
util/deephash: remove unnecessary formatting for structs and slices (#2571) The index for every struct field or slice element and the number of fields for the struct is unncessary. The hashing of Go values is unambiguous because every type (except maps) encodes in a parsable manner. So long as we know the type information, we could theoretically decode every value (except for maps). At a high level: * numbers are encoded as fixed-width records according to precision. * strings (and AppendTo output) are encoded with a fixed-width length, followed by the contents of the buffer. * slices are prefixed by a fixed-width length, followed by the encoding of each value. So long as we know the type of each element, we could theoretically decode each element. * arrays are encoded just like slices, but elide the length since it is determined from the Go type. * maps are encoded first with a byte indicating whether it is a cycle. If a cycle, it is followed by a fixed-width index for the pointer, otherwise followed by the SHA-256 hash of its contents. The encoding of maps is not decodeable, but a SHA-256 hash is sufficient to avoid ambiguities. * interfaces are encoded first with a byte indicating whether it is nil. If not nil, it is followed by a fixed-width index for the type, and then the encoding for the underlying value. Having the type be encoded first ensures that the value could theoretically be decoded next. * pointers are encoded first with a byte indicating whether it is 1) nil, 2) a cycle, or 3) newly seen. If a cycle, it is followed by a fixed-width index for the pointer. If newly seen, it is followed by the encoding for the pointed-at value. Removing unnecessary details speeds up hashing: name old time/op new time/op delta Hash-8 76.0µs ± 1% 55.8µs ± 2% -26.62% (p=0.000 n=10+10) HashMapAcyclic-8 61.9µs ± 0% 62.0µs ± 0% ~ (p=0.666 n=9+9) TailcfgNode-8 10.2µs ± 1% 7.5µs ± 1% -26.90% (p=0.000 n=10+9) HashArray-8 1.07µs ± 1% 0.70µs ± 1% -34.67% (p=0.000 n=10+9) Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
const want = "\x00\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1f"
if got := got.Bytes(); string(got) != want {
t.Errorf("wrong:\n got: %q\nwant: %q\n", got, want)
}
}
func BenchmarkHashMapAcyclic(b *testing.B) {
b.ReportAllocs()
m := map[int]string{}
for i := 0; i < 100; i++ {
m[i] = fmt.Sprint(i)
}
var buf bytes.Buffer
bw := bufio.NewWriter(&buf)
v := reflect.ValueOf(m)
ti := getTypeInfo(v.Type())
util/deephash: improve cycle detection (#2470) The previous algorithm used a map of all visited pointers. The strength of this approach is that it quickly prunes any nodes that we have ever visited before. The detriment of the approach is that pruning is heavily dependent on the order that pointers were visited. This is especially relevant for hashing a map where map entries are visited in a non-deterministic manner, which would cause the map hash to be non-deterministic (which defeats the point of a hash). This new algorithm uses a stack of all visited pointers, similar to how github.com/google/go-cmp performs cycle detection. When we visit a pointer, we push it onto the stack, and when we leave a pointer, we pop it from the stack. Before visiting a pointer, we first check whether the pointer exists anywhere in the stack. If yes, then we prune the node. The detriment of this approach is that we may hash a node more often than before since we do not prune as aggressively. The set of visited pointers up until any node is only the path of nodes up to that node and not any other pointers that may have been visited elsewhere. This provides us deterministic hashing regardless of visit order. We can now delete hashMapFallback and associated complexity, which only exists because the previous approach was non-deterministic in the presence of cycles. This fixes a failure of the old algorithm where obviously different values are treated as equal because the pruning was too aggresive. See https://github.com/tailscale/tailscale/issues/2443#issuecomment-883653534 The new algorithm is slightly slower since it prunes less aggresively: name old time/op new time/op delta Hash-8 66.1µs ± 1% 68.8µs ± 1% +4.09% (p=0.000 n=19+19) HashMapAcyclic-8 63.0µs ± 1% 62.5µs ± 1% -0.76% (p=0.000 n=18+19) TailcfgNode-8 9.79µs ± 2% 9.88µs ± 1% +0.95% (p=0.000 n=19+17) HashArray-8 643ns ± 1% 653ns ± 1% +1.64% (p=0.000 n=19+19) However, a slower but more correct algorithm seems more favorable than a faster but incorrect algorithm. Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
h := &hasher{bw: bw}
for i := 0; i < b.N; i++ {
buf.Reset()
bw.Reset(&buf)
h.hashMap(v, ti, false)
}
}
func BenchmarkTailcfgNode(b *testing.B) {
b.ReportAllocs()
node := new(tailcfg.Node)
for i := 0; i < b.N; i++ {
sink = Hash(node)
}
}
func TestExhaustive(t *testing.T) {
seen := make(map[Sum]bool)
for i := 0; i < 100000; i++ {
s := Hash(i)
if seen[s] {
t.Fatalf("hash collision %v", i)
}
seen[s] = true
}
}
// verify this doesn't loop forever, as it used to (Issue 2340)
func TestMapCyclicFallback(t *testing.T) {
type T struct {
M map[string]any
}
v := &T{
M: map[string]any{},
}
v.M["m"] = v.M
Hash(v)
}
func TestArrayAllocs(t *testing.T) {
if version.IsRace() {
t.Skip("skipping test under race detector")
}
// In theory, there should be no allocations. However, escape analysis on
// certain architectures fails to detect that certain cases do not escape.
// This discrepency currently affects sha256.digest.Sum.
// Measure the number of allocations in sha256 to ensure that Hash does
// not allocate on top of its usage of sha256.
// See https://golang.org/issue/48055.
var b []byte
h := sha256.New()
want := int(testing.AllocsPerRun(1000, func() {
b = h.Sum(b[:0])
}))
switch runtime.GOARCH {
case "amd64", "arm64":
want = 0 // ensure no allocations on popular architectures
}
type T struct {
X [32]byte
}
x := &T{X: [32]byte{1: 1, 2: 2, 3: 3, 4: 4}}
got := int(testing.AllocsPerRun(1000, func() {
sink = Hash(x)
}))
if got > want {
t.Errorf("allocs = %v; want %v", got, want)
}
}
func BenchmarkHashArray(b *testing.B) {
b.ReportAllocs()
type T struct {
X [32]byte
}
x := &T{X: [32]byte{1: 1, 2: 2, 3: 3, 4: 4}}
for i := 0; i < b.N; i++ {
sink = Hash(x)
}
}