minio/internal/grid/benchmark_test.go

573 lines
16 KiB
Go
Raw Normal View History

perf: websocket grid connectivity for all internode communication (#18461) This PR adds a WebSocket grid feature that allows servers to communicate via a single two-way connection. There are two request types: * Single requests, which are `[]byte => ([]byte, error)`. This is for efficient small roundtrips with small payloads. * Streaming requests which are `[]byte, chan []byte => chan []byte (and error)`, which allows for different combinations of full two-way streams with an initial payload. Only a single stream is created between two machines - and there is, as such, no server/client relation since both sides can initiate and handle requests. Which server initiates the request is decided deterministically on the server names. Requests are made through a mux client and server, which handles message passing, congestion, cancelation, timeouts, etc. If a connection is lost, all requests are canceled, and the calling server will try to reconnect. Registered handlers can operate directly on byte slices or use a higher-level generics abstraction. There is no versioning of handlers/clients, and incompatible changes should be handled by adding new handlers. The request path can be changed to a new one for any protocol changes. First, all servers create a "Manager." The manager must know its address as well as all remote addresses. This will manage all connections. To get a connection to any remote, ask the manager to provide it given the remote address using. ``` func (m *Manager) Connection(host string) *Connection ``` All serverside handlers must also be registered on the manager. This will make sure that all incoming requests are served. The number of in-flight requests and responses must also be given for streaming requests. The "Connection" returned manages the mux-clients. Requests issued to the connection will be sent to the remote. * `func (c *Connection) Request(ctx context.Context, h HandlerID, req []byte) ([]byte, error)` performs a single request and returns the result. Any deadline provided on the request is forwarded to the server, and canceling the context will make the function return at once. * `func (c *Connection) NewStream(ctx context.Context, h HandlerID, payload []byte) (st *Stream, err error)` will initiate a remote call and send the initial payload. ```Go // A Stream is a two-way stream. // All responses *must* be read by the caller. // If the call is canceled through the context, //The appropriate error will be returned. type Stream struct { // Responses from the remote server. // Channel will be closed after an error or when the remote closes. // All responses *must* be read by the caller until either an error is returned or the channel is closed. // Canceling the context will cause the context cancellation error to be returned. Responses <-chan Response // Requests sent to the server. // If the handler is defined with 0 incoming capacity this will be nil. // Channel *must* be closed to signal the end of the stream. // If the request context is canceled, the stream will no longer process requests. Requests chan<- []byte } type Response struct { Msg []byte Err error } ``` There are generic versions of the server/client handlers that allow the use of type safe implementations for data types that support msgpack marshal/unmarshal.
2023-11-20 20:09:35 -05:00
// Copyright (c) 2015-2023 MinIO, Inc.
//
// This file is part of MinIO Object Storage stack
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Affero General Public License for more details.
//
// You should have received a copy of the GNU Affero General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
package grid
import (
"context"
"fmt"
"math/rand"
"runtime"
"strconv"
"sync/atomic"
"testing"
"time"
"github.com/minio/minio/internal/logger/target/testlogger"
)
func BenchmarkRequests(b *testing.B) {
for n := 2; n <= 32; n *= 2 {
b.Run("servers="+strconv.Itoa(n), func(b *testing.B) {
benchmarkGridRequests(b, n)
})
}
}
func benchmarkGridRequests(b *testing.B, n int) {
defer testlogger.T.SetErrorTB(b)()
errFatal := func(err error) {
b.Helper()
if err != nil {
b.Fatal(err)
}
}
rpc := NewSingleHandler[*testRequest, *testResponse](handlerTest2, newTestRequest, newTestResponse)
grid, err := SetupTestGrid(n)
errFatal(err)
b.Cleanup(grid.Cleanup)
// Create n managers.
for _, remote := range grid.Managers {
// Register a single handler which echos the payload.
errFatal(remote.RegisterSingleHandler(handlerTest, func(payload []byte) ([]byte, *RemoteErr) {
defer PutByteBuffer(payload)
return append(GetByteBuffer()[:0], payload...), nil
}))
errFatal(rpc.Register(remote, func(req *testRequest) (resp *testResponse, err *RemoteErr) {
return &testResponse{
OrgNum: req.Num,
OrgString: req.String,
Embedded: *req,
}, nil
}))
errFatal(err)
}
const payloadSize = 512
rng := rand.New(rand.NewSource(time.Now().UnixNano()))
payload := make([]byte, payloadSize)
_, err = rng.Read(payload)
errFatal(err)
// Wait for all to connect
// Parallel writes per server.
b.Run("bytes", func(b *testing.B) {
for par := 1; par <= 32; par *= 2 {
b.Run("par="+strconv.Itoa(par*runtime.GOMAXPROCS(0)), func(b *testing.B) {
defer timeout(60 * time.Second)()
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer cancel()
b.ReportAllocs()
b.SetBytes(int64(len(payload) * 2))
b.ResetTimer()
t := time.Now()
var ops int64
var lat int64
b.SetParallelism(par)
b.RunParallel(func(pb *testing.PB) {
rng := rand.New(rand.NewSource(time.Now().UnixNano()))
n := 0
var latency int64
managers := grid.Managers
hosts := grid.Hosts
for pb.Next() {
// Pick a random manager.
src, dst := rng.Intn(len(managers)), rng.Intn(len(managers))
if src == dst {
dst = (dst + 1) % len(managers)
}
local := managers[src]
conn := local.Connection(hosts[dst])
if conn == nil {
b.Fatal("No connection")
}
// Send the payload.
t := time.Now()
resp, err := conn.Request(ctx, handlerTest, payload)
latency += time.Since(t).Nanoseconds()
if err != nil {
if debugReqs {
fmt.Println(err.Error())
}
b.Fatal(err.Error())
}
PutByteBuffer(resp)
n++
}
atomic.AddInt64(&ops, int64(n))
atomic.AddInt64(&lat, latency)
})
spent := time.Since(t)
if spent > 0 && n > 0 {
// Since we are benchmarking n parallel servers we need to multiply by n.
// This will give an estimate of the total ops/s.
latency := float64(atomic.LoadInt64(&lat)) / float64(time.Millisecond)
b.ReportMetric(float64(n)*float64(ops)/spent.Seconds(), "vops/s")
b.ReportMetric(latency/float64(ops), "ms/op")
}
})
}
})
b.Run("rpc", func(b *testing.B) {
for par := 1; par <= 32; par *= 2 {
b.Run("par="+strconv.Itoa(par*runtime.GOMAXPROCS(0)), func(b *testing.B) {
defer timeout(60 * time.Second)()
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer cancel()
b.ReportAllocs()
b.ResetTimer()
t := time.Now()
var ops int64
var lat int64
b.SetParallelism(par)
b.RunParallel(func(pb *testing.PB) {
rng := rand.New(rand.NewSource(time.Now().UnixNano()))
n := 0
var latency int64
managers := grid.Managers
hosts := grid.Hosts
req := testRequest{
Num: rng.Int(),
String: "hello",
}
for pb.Next() {
// Pick a random manager.
src, dst := rng.Intn(len(managers)), rng.Intn(len(managers))
if src == dst {
dst = (dst + 1) % len(managers)
}
local := managers[src]
conn := local.Connection(hosts[dst])
if conn == nil {
b.Fatal("No connection")
}
// Send the payload.
t := time.Now()
resp, err := rpc.Call(ctx, conn, &req)
latency += time.Since(t).Nanoseconds()
if err != nil {
if debugReqs {
fmt.Println(err.Error())
}
b.Fatal(err.Error())
}
rpc.PutResponse(resp)
n++
}
atomic.AddInt64(&ops, int64(n))
atomic.AddInt64(&lat, latency)
})
spent := time.Since(t)
if spent > 0 && n > 0 {
// Since we are benchmarking n parallel servers we need to multiply by n.
// This will give an estimate of the total ops/s.
latency := float64(atomic.LoadInt64(&lat)) / float64(time.Millisecond)
b.ReportMetric(float64(n)*float64(ops)/spent.Seconds(), "vops/s")
b.ReportMetric(latency/float64(ops), "ms/op")
}
})
}
})
}
func BenchmarkStream(b *testing.B) {
tests := []struct {
name string
fn func(b *testing.B, n int)
}{
{name: "request", fn: benchmarkGridStreamReqOnly},
{name: "responses", fn: benchmarkGridStreamRespOnly},
{name: "twoway", fn: benchmarkGridStreamTwoway},
perf: websocket grid connectivity for all internode communication (#18461) This PR adds a WebSocket grid feature that allows servers to communicate via a single two-way connection. There are two request types: * Single requests, which are `[]byte => ([]byte, error)`. This is for efficient small roundtrips with small payloads. * Streaming requests which are `[]byte, chan []byte => chan []byte (and error)`, which allows for different combinations of full two-way streams with an initial payload. Only a single stream is created between two machines - and there is, as such, no server/client relation since both sides can initiate and handle requests. Which server initiates the request is decided deterministically on the server names. Requests are made through a mux client and server, which handles message passing, congestion, cancelation, timeouts, etc. If a connection is lost, all requests are canceled, and the calling server will try to reconnect. Registered handlers can operate directly on byte slices or use a higher-level generics abstraction. There is no versioning of handlers/clients, and incompatible changes should be handled by adding new handlers. The request path can be changed to a new one for any protocol changes. First, all servers create a "Manager." The manager must know its address as well as all remote addresses. This will manage all connections. To get a connection to any remote, ask the manager to provide it given the remote address using. ``` func (m *Manager) Connection(host string) *Connection ``` All serverside handlers must also be registered on the manager. This will make sure that all incoming requests are served. The number of in-flight requests and responses must also be given for streaming requests. The "Connection" returned manages the mux-clients. Requests issued to the connection will be sent to the remote. * `func (c *Connection) Request(ctx context.Context, h HandlerID, req []byte) ([]byte, error)` performs a single request and returns the result. Any deadline provided on the request is forwarded to the server, and canceling the context will make the function return at once. * `func (c *Connection) NewStream(ctx context.Context, h HandlerID, payload []byte) (st *Stream, err error)` will initiate a remote call and send the initial payload. ```Go // A Stream is a two-way stream. // All responses *must* be read by the caller. // If the call is canceled through the context, //The appropriate error will be returned. type Stream struct { // Responses from the remote server. // Channel will be closed after an error or when the remote closes. // All responses *must* be read by the caller until either an error is returned or the channel is closed. // Canceling the context will cause the context cancellation error to be returned. Responses <-chan Response // Requests sent to the server. // If the handler is defined with 0 incoming capacity this will be nil. // Channel *must* be closed to signal the end of the stream. // If the request context is canceled, the stream will no longer process requests. Requests chan<- []byte } type Response struct { Msg []byte Err error } ``` There are generic versions of the server/client handlers that allow the use of type safe implementations for data types that support msgpack marshal/unmarshal.
2023-11-20 20:09:35 -05:00
}
for _, test := range tests {
b.Run(test.name, func(b *testing.B) {
for n := 2; n <= 32; n *= 2 {
b.Run("servers="+strconv.Itoa(n), func(b *testing.B) {
test.fn(b, n)
})
}
})
}
}
func benchmarkGridStreamRespOnly(b *testing.B, n int) {
defer testlogger.T.SetErrorTB(b)()
errFatal := func(err error) {
b.Helper()
if err != nil {
b.Fatal(err)
}
}
grid, err := SetupTestGrid(n)
errFatal(err)
b.Cleanup(grid.Cleanup)
const responses = 10
// Create n managers.
for _, remote := range grid.Managers {
// Register a single handler which echos the payload.
errFatal(remote.RegisterStreamingHandler(handlerTest, StreamHandler{
// Send 10x response.
Handle: func(ctx context.Context, payload []byte, _ <-chan []byte, out chan<- []byte) *RemoteErr {
for i := 0; i < responses; i++ {
toSend := GetByteBuffer()[:0]
toSend = append(toSend, byte(i))
toSend = append(toSend, payload...)
select {
case <-ctx.Done():
return nil
case out <- toSend:
}
}
return nil
},
Subroute: "some-subroute",
OutCapacity: 1, // Only one message buffered.
InCapacity: 0,
}))
errFatal(err)
}
const payloadSize = 512
rng := rand.New(rand.NewSource(time.Now().UnixNano()))
payload := make([]byte, payloadSize)
_, err = rng.Read(payload)
errFatal(err)
// Wait for all to connect
// Parallel writes per server.
for par := 1; par <= 32; par *= 2 {
b.Run("par="+strconv.Itoa(par*runtime.GOMAXPROCS(0)), func(b *testing.B) {
defer timeout(30 * time.Second)()
b.ReportAllocs()
b.SetBytes(int64(len(payload) * (responses + 1)))
b.ResetTimer()
t := time.Now()
var ops int64
var lat int64
b.SetParallelism(par)
b.RunParallel(func(pb *testing.PB) {
rng := rand.New(rand.NewSource(time.Now().UnixNano()))
n := 0
var latency int64
managers := grid.Managers
hosts := grid.Hosts
for pb.Next() {
// Pick a random manager.
src, dst := rng.Intn(len(managers)), rng.Intn(len(managers))
if src == dst {
dst = (dst + 1) % len(managers)
}
local := managers[src]
conn := local.Connection(hosts[dst]).Subroute("some-subroute")
if conn == nil {
b.Fatal("No connection")
}
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
// Send the payload.
t := time.Now()
st, err := conn.NewStream(ctx, handlerTest, payload)
if err != nil {
if debugReqs {
fmt.Println(err.Error())
}
b.Fatal(err.Error())
}
got := 0
err = st.Results(func(b []byte) error {
got++
PutByteBuffer(b)
return nil
})
if err != nil {
if debugReqs {
fmt.Println(err.Error())
}
b.Fatal(err.Error())
}
latency += time.Since(t).Nanoseconds()
cancel()
n += got
}
atomic.AddInt64(&ops, int64(n))
atomic.AddInt64(&lat, latency)
})
spent := time.Since(t)
if spent > 0 && n > 0 {
// Since we are benchmarking n parallel servers we need to multiply by n.
// This will give an estimate of the total ops/s.
latency := float64(atomic.LoadInt64(&lat)) / float64(time.Millisecond)
b.ReportMetric(float64(n)*float64(ops)/spent.Seconds(), "vops/s")
b.ReportMetric(latency/float64(ops), "ms/op")
}
})
}
}
func benchmarkGridStreamReqOnly(b *testing.B, n int) {
defer testlogger.T.SetErrorTB(b)()
errFatal := func(err error) {
b.Helper()
if err != nil {
b.Fatal(err)
}
}
grid, err := SetupTestGrid(n)
errFatal(err)
b.Cleanup(grid.Cleanup)
const requests = 10
// Create n managers.
for _, remote := range grid.Managers {
// Register a single handler which echos the payload.
errFatal(remote.RegisterStreamingHandler(handlerTest, StreamHandler{
// Send 10x requests.
Handle: func(ctx context.Context, payload []byte, in <-chan []byte, out chan<- []byte) *RemoteErr {
got := 0
for b := range in {
PutByteBuffer(b)
got++
}
if got != requests {
return NewRemoteErrf("wrong number of requests. want %d, got %d", requests, got)
}
return nil
},
Subroute: "some-subroute",
OutCapacity: 1,
InCapacity: 1, // Only one message buffered.
}))
errFatal(err)
}
const payloadSize = 512
rng := rand.New(rand.NewSource(time.Now().UnixNano()))
payload := make([]byte, payloadSize)
_, err = rng.Read(payload)
errFatal(err)
// Wait for all to connect
// Parallel writes per server.
for par := 1; par <= 32; par *= 2 {
b.Run("par="+strconv.Itoa(par*runtime.GOMAXPROCS(0)), func(b *testing.B) {
defer timeout(30 * time.Second)()
b.ReportAllocs()
b.SetBytes(int64(len(payload) * (requests + 1)))
b.ResetTimer()
t := time.Now()
var ops int64
var lat int64
b.SetParallelism(par)
b.RunParallel(func(pb *testing.PB) {
rng := rand.New(rand.NewSource(time.Now().UnixNano()))
n := 0
var latency int64
managers := grid.Managers
hosts := grid.Hosts
for pb.Next() {
// Pick a random manager.
src, dst := rng.Intn(len(managers)), rng.Intn(len(managers))
if src == dst {
dst = (dst + 1) % len(managers)
}
local := managers[src]
conn := local.Connection(hosts[dst]).Subroute("some-subroute")
if conn == nil {
b.Fatal("No connection")
}
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
// Send the payload.
t := time.Now()
st, err := conn.NewStream(ctx, handlerTest, payload)
if err != nil {
if debugReqs {
fmt.Println(err.Error())
}
b.Fatal(err.Error())
}
got := 0
for i := 0; i < requests; i++ {
got++
st.Requests <- append(GetByteBuffer()[:0], payload...)
}
close(st.Requests)
err = st.Results(func(b []byte) error {
return nil
})
if err != nil {
if debugReqs {
fmt.Println(err.Error())
}
b.Fatal(err.Error())
}
latency += time.Since(t).Nanoseconds()
cancel()
n += got
}
atomic.AddInt64(&ops, int64(n))
atomic.AddInt64(&lat, latency)
})
spent := time.Since(t)
if spent > 0 && n > 0 {
// Since we are benchmarking n parallel servers we need to multiply by n.
// This will give an estimate of the total ops/s.
latency := float64(atomic.LoadInt64(&lat)) / float64(time.Millisecond)
b.ReportMetric(float64(n)*float64(ops)/spent.Seconds(), "vops/s")
b.ReportMetric(latency/float64(ops), "ms/op")
}
})
}
}
func benchmarkGridStreamTwoway(b *testing.B, n int) {
defer testlogger.T.SetErrorTB(b)()
errFatal := func(err error) {
b.Helper()
if err != nil {
b.Fatal(err)
}
}
grid, err := SetupTestGrid(n)
errFatal(err)
b.Cleanup(grid.Cleanup)
const messages = 10
// Create n managers.
const payloadSize = 512
rng := rand.New(rand.NewSource(time.Now().UnixNano()))
payload := make([]byte, payloadSize)
_, err = rng.Read(payload)
errFatal(err)
for _, remote := range grid.Managers {
// Register a single handler which echos the payload.
errFatal(remote.RegisterStreamingHandler(handlerTest, StreamHandler{
// Send 10x requests.
Handle: func(ctx context.Context, payload []byte, in <-chan []byte, out chan<- []byte) *RemoteErr {
got := 0
for {
select {
case b, ok := <-in:
if !ok {
if got != messages {
return NewRemoteErrf("wrong number of requests. want %d, got %d", messages, got)
}
return nil
}
out <- b
got++
}
}
},
Subroute: "some-subroute",
OutCapacity: 1,
InCapacity: 1, // Only one message buffered.
}))
errFatal(err)
}
// Wait for all to connect
// Parallel writes per server.
for par := 1; par <= 32; par *= 2 {
b.Run("par="+strconv.Itoa(par*runtime.GOMAXPROCS(0)), func(b *testing.B) {
defer timeout(30 * time.Second)()
b.ReportAllocs()
b.SetBytes(int64(len(payload) * (2*messages + 1)))
b.ResetTimer()
t := time.Now()
var ops int64
var lat int64
b.SetParallelism(par)
b.RunParallel(func(pb *testing.PB) {
rng := rand.New(rand.NewSource(time.Now().UnixNano()))
n := 0
var latency int64
managers := grid.Managers
hosts := grid.Hosts
for pb.Next() {
// Pick a random manager.
src, dst := rng.Intn(len(managers)), rng.Intn(len(managers))
if src == dst {
dst = (dst + 1) % len(managers)
}
local := managers[src]
conn := local.Connection(hosts[dst]).Subroute("some-subroute")
if conn == nil {
b.Fatal("No connection")
}
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
// Send the payload.
t := time.Now()
st, err := conn.NewStream(ctx, handlerTest, payload)
if err != nil {
if debugReqs {
fmt.Println(err.Error())
}
b.Fatal(err.Error())
}
got := 0
sent := 0
go func() {
for i := 0; i < messages; i++ {
st.Requests <- append(GetByteBuffer()[:0], payload...)
if sent++; sent == messages {
close(st.Requests)
return
}
}
}()
err = st.Results(func(b []byte) error {
got++
PutByteBuffer(b)
return nil
})
if err != nil {
if debugReqs {
fmt.Println(err.Error())
}
b.Fatal(err.Error())
}
if got != messages {
b.Fatalf("wrong number of responses. want %d, got %d", messages, got)
}
latency += time.Since(t).Nanoseconds()
cancel()
n += got
}
atomic.AddInt64(&ops, int64(n*2))
atomic.AddInt64(&lat, latency)
})
spent := time.Since(t)
if spent > 0 && n > 0 {
// Since we are benchmarking n parallel servers we need to multiply by n.
// This will give an estimate of the total ops/s.
latency := float64(atomic.LoadInt64(&lat)) / float64(time.Millisecond)
b.ReportMetric(float64(n)*float64(ops)/spent.Seconds(), "vops/s")
b.ReportMetric(latency/float64(ops), "ms/op")
}
})
}
}