This commit is contained in:
Xiang Li 2013-08-06 19:57:03 -07:00
commit e5aa6256f0
97 changed files with 1184 additions and 5448 deletions

6
.gitignore vendored
View File

@ -1,2 +1,4 @@
src
etcd
src/
pkg/
./etcd
release_version.go

8
.travis.yml Normal file
View File

@ -0,0 +1,8 @@
language: go
go: 1.1
install:
- echo "Skip install"
script:
- ./test

View File

@ -1,5 +1,7 @@
# etcd
[![Build Status](https://travis-ci.org/coreos/etcd.png)](https://travis-ci.org/coreos/etcd)
A highly-available key value store for shared configuration and service discovery. etcd is inspired by zookeeper and doozer, with a focus on:
* Simple: curl'able user facing API (HTTP+JSON)
@ -18,14 +20,10 @@ See [go-etcd][go-etcd] for a native go client. Or feel free to just use curl, as
### Building
etcd is installed like any other Go (golang >= 1.1) binary. The steps below will put everything into a directory called etcd.
To build etcd run the build script. This will generate a binary in the base directory called `./etcd`.
```
mkdir etcd
cd etcd
export GOPATH=`pwd`
go get github.com/coreos/etcd
go install github.com/coreos/etcd
./build
```
### Running a single node
@ -33,7 +31,7 @@ go install github.com/coreos/etcd
These examples will use a single node cluster to show you the basics of the etcd REST API. Lets start etcd:
```sh
./bin/etcd
./etcd
```
This will bring up a node, which will be listening on internal port 7001 (for server communication) and external port 4001 (for client communication)

1
build
View File

@ -21,4 +21,5 @@ for i in third_party/*; do
cp -R $i src/
done
./scripts/release-version > release_version.go
go build ${ETCD_PACKAGE}

View File

@ -4,6 +4,7 @@ import (
"github.com/coreos/etcd/store"
"net/http"
"strconv"
"fmt"
"time"
)
@ -14,13 +15,14 @@ import (
// Multiplex GET/POST/DELETE request to corresponding handlers
func Multiplexer(w http.ResponseWriter, req *http.Request) {
if req.Method == "GET" {
switch req.Method {
case "GET":
GetHttpHandler(&w, req)
} else if req.Method == "POST" {
case "POST":
SetHttpHandler(&w, req)
} else if req.Method == "DELETE" {
case "DELETE":
DeleteHttpHandler(&w, req)
} else {
default:
w.WriteHeader(http.StatusMethodNotAllowed)
return
}
@ -69,18 +71,22 @@ func SetHttpHandler(w *http.ResponseWriter, req *http.Request) {
}
if len(prevValue) != 0 {
command := &TestAndSetCommand{}
command.Key = key
command.Value = value
command.PrevValue = prevValue
command.ExpireTime = expireTime
command := &TestAndSetCommand{
Key: key,
Value: value,
PrevValue: prevValue,
ExpireTime: expireTime,
}
dispatch(command, w, req, true)
} else {
command := &SetCommand{}
command.Key = key
command.Value = value
command.ExpireTime = expireTime
command := &SetCommand{
Key: key,
Value: value,
ExpireTime: expireTime,
}
dispatch(command, w, req, true)
}
@ -92,8 +98,9 @@ func DeleteHttpHandler(w *http.ResponseWriter, req *http.Request) {
debugf("[recv] DELETE http://%v/v1/keys/%s", raftServer.Name(), key)
command := &DeleteCommand{}
command.Key = key
command := &DeleteCommand{
Key: key,
}
dispatch(command, w, req, true)
}
@ -228,7 +235,7 @@ func MachinesHttpHandler(w http.ResponseWriter, req *http.Request) {
// Handler to return the current version of etcd
func VersionHttpHandler(w http.ResponseWriter, req *http.Request) {
w.WriteHeader(http.StatusOK)
w.Write([]byte(releaseVersion))
w.Write([]byte(fmt.Sprintf("etcd %s", releaseVersion)))
}
// Handler to return the basic stats of etcd
@ -243,8 +250,9 @@ func GetHttpHandler(w *http.ResponseWriter, req *http.Request) {
debugf("[recv] GET http://%v/v1/keys/%s", raftServer.Name(), key)
command := &GetCommand{}
command.Key = key
command := &GetCommand{
Key: key,
}
if body, err := command.Apply(raftServer); err != nil {
@ -274,8 +282,9 @@ func GetHttpHandler(w *http.ResponseWriter, req *http.Request) {
func WatchHttpHandler(w http.ResponseWriter, req *http.Request) {
key := req.URL.Path[len("/v1/watch/"):]
command := &WatchCommand{}
command.Key = key
command := &WatchCommand{
Key: key,
}
if req.Method == "GET" {
debugf("[recv] GET http://%v/watch/%s", raftServer.Name(), key)

58
etcd.go
View File

@ -74,7 +74,7 @@ func init() {
flag.StringVar(&hostname, "h", "0.0.0.0", "the hostname of the local machine")
flag.IntVar(&clientPort, "c", 4001, "the port to communicate with clients")
flag.IntVar(&raftPort, "s", 7001, "the port to communicate with servers")
flag.IntVar(&webPort, "w", -1, "the port of web interface")
flag.IntVar(&webPort, "w", -1, "the port of web interface (-1 means do not start web interface)")
flag.StringVar(&serverCAFile, "serverCAFile", "", "the path of the CAFile")
flag.StringVar(&serverCertFile, "serverCert", "", "the cert file of the server")
@ -278,11 +278,12 @@ func startRaft(securityType int) {
// leader need to join self as a peer
for {
command := &JoinCommand{}
command.Name = raftServer.Name()
command.Hostname = hostname
command.RaftPort = raftPort
command.ClientPort = clientPort
command := &JoinCommand{
Name: raftServer.Name(),
Hostname: hostname,
RaftPort: raftPort,
ClientPort: clientPort,
}
_, err := raftServer.Do(command)
if err == nil {
break
@ -513,7 +514,6 @@ func securityType(source int) int {
// Get the server info from previous conf file
// or from the user
func getInfo(path string) *Info {
info := &Info{}
// Read in the server info if available.
infoPath := fmt.Sprintf("%s/info", path)
@ -532,6 +532,7 @@ func getInfo(path string) *Info {
}
if file, err := os.Open(infoPath); err == nil {
info := &Info{}
if content, err := ioutil.ReadAll(file); err != nil {
fatalf("Unable to read info: %v", err)
} else {
@ -540,29 +541,32 @@ func getInfo(path string) *Info {
}
}
file.Close()
return info
} else {
// Otherwise ask user for info and write it to file.
hostname = strings.TrimSpace(hostname)
if hostname == "" {
fatal("Please give the address of the local machine")
}
info.Hostname = hostname
info.Hostname = strings.TrimSpace(info.Hostname)
fmt.Println("address ", info.Hostname)
fmt.Println("address ", hostname)
info := &Info{
Hostname: hostname,
info.RaftPort = raftPort
info.ClientPort = clientPort
info.WebPort = webPort
RaftPort: raftPort,
ClientPort: clientPort,
WebPort: webPort,
info.ClientCAFile = clientCAFile
info.ClientCertFile = clientCertFile
info.ClientKeyFile = clientKeyFile
ClientCAFile: clientCAFile,
ClientCertFile: clientCertFile,
ClientKeyFile: clientKeyFile,
info.ServerCAFile = serverCAFile
info.ServerKeyFile = serverKeyFile
info.ServerCertFile = serverCertFile
ServerCAFile: serverCAFile,
ServerKeyFile: serverKeyFile,
ServerCertFile: serverCertFile,
}
// Write to file.
content, _ := json.Marshal(info)
@ -570,9 +574,8 @@ func getInfo(path string) *Info {
if err := ioutil.WriteFile(infoPath, content, 0644); err != nil {
fatalf("Unable to write info to file: %v", err)
}
return info
}
return info
}
// Create client auth certpool
@ -598,11 +601,12 @@ func createCertPool(CAFile string) *x509.CertPool {
func joinCluster(s *raft.Server, serverName string) error {
var b bytes.Buffer
command := &JoinCommand{}
command.Name = s.Name()
command.Hostname = info.Hostname
command.RaftPort = info.RaftPort
command.ClientPort = info.ClientPort
command := &JoinCommand{
Name: s.Name(),
Hostname: info.Hostname,
RaftPort: info.RaftPort,
ClientPort: info.ClientPort,
}
json.NewEncoder(&b).Encode(command)

8
scripts/release-version Executable file
View File

@ -0,0 +1,8 @@
#!/bin/sh
VER=$(git describe --tags HEAD)
cat <<EOF
package main
const releaseVersion = "$VER"
EOF

7
test
View File

@ -1,3 +1,8 @@
#!/bin/sh
go build
# Get GOPATH, etc from build
. ./build
# Run the tests!
go test -i
go test -v

View File

@ -377,12 +377,17 @@ func (es enumSymbol) GenerateAlias(g *Generator, pkg string) {
}
type constOrVarSymbol struct {
sym string
typ string // either "const" or "var"
sym string
typ string // either "const" or "var"
cast string // if non-empty, a type cast is required (used for enums)
}
func (cs constOrVarSymbol) GenerateAlias(g *Generator, pkg string) {
g.P(cs.typ, " ", cs.sym, " = ", pkg, ".", cs.sym)
v := pkg + "." + cs.sym
if cs.cast != "" {
v = cs.cast + "(" + v + ")"
}
g.P(cs.typ, " ", cs.sym, " = ", v)
}
// Object is an interface abstracting the abilities shared by enums, messages, extensions and imported objects.
@ -1157,7 +1162,7 @@ func (g *Generator) generateEnum(enum *EnumDescriptor) {
name := ccPrefix + *e.Name
g.P(name, " ", ccTypeName, " = ", e.Number)
g.file.addExport(enum, constOrVarSymbol{name, "const"})
g.file.addExport(enum, constOrVarSymbol{name, "const", ccTypeName})
}
g.Out()
g.P(")")
@ -1255,9 +1260,18 @@ func (g *Generator) goTag(field *descriptor.FieldDescriptorProto, wiretype strin
case descriptor.FieldDescriptorProto_TYPE_ENUM:
// For enums we need to provide the integer constant.
obj := g.ObjectNamed(field.GetTypeName())
if id, ok := obj.(*ImportedDescriptor); ok {
// It is an enum that was publicly imported.
// We need the underlying type.
obj = id.o
}
enum, ok := obj.(*EnumDescriptor)
if !ok {
g.Fail("enum type inconsistent for", CamelCaseSlice(obj.TypeName()))
log.Printf("obj is a %T", obj)
if id, ok := obj.(*ImportedDescriptor); ok {
log.Printf("id.o is a %T", id.o)
}
g.Fail("unknown enum type", CamelCaseSlice(obj.TypeName()))
}
defaultValue = enum.integerValueAsString(defaultValue)
}
@ -1268,6 +1282,9 @@ func (g *Generator) goTag(field *descriptor.FieldDescriptorProto, wiretype strin
// We avoid using obj.PackageName(), because we want to use the
// original (proto-world) package name.
obj := g.ObjectNamed(field.GetTypeName())
if id, ok := obj.(*ImportedDescriptor); ok {
obj = id.o
}
enum = ",enum="
if pkg := obj.File().GetPackage(); pkg != "" {
enum += pkg + "."
@ -1541,15 +1558,21 @@ func (g *Generator) generateMessage(message *Descriptor) {
case *field.Type == descriptor.FieldDescriptorProto_TYPE_ENUM:
// Must be an enum. Need to construct the prefixed name.
obj := g.ObjectNamed(field.GetTypeName())
enum, ok := obj.(*EnumDescriptor)
if !ok {
log.Print("don't know how to generate constant for", fieldname)
var enum *EnumDescriptor
if id, ok := obj.(*ImportedDescriptor); ok {
// The enum type has been publicly imported.
enum, _ = id.o.(*EnumDescriptor)
} else {
enum, _ = obj.(*EnumDescriptor)
}
if enum == nil {
log.Printf("don't know how to generate constant for %s", fieldname)
continue
}
def = g.DefaultPackageName(enum) + enum.prefix() + def
def = g.DefaultPackageName(obj) + enum.prefix() + def
}
g.P(kind, fieldname, " ", typename, " = ", def)
g.file.addExport(message, constOrVarSymbol{fieldname, kind})
g.file.addExport(message, constOrVarSymbol{fieldname, kind, ""})
}
g.P()
@ -1701,7 +1724,7 @@ func (g *Generator) generateExtension(ext *ExtensionDescriptor) {
g.P("}")
g.P()
g.file.addExport(ext, constOrVarSymbol{ccTypeName, "var"})
g.file.addExport(ext, constOrVarSymbol{ccTypeName, "var", ""})
}
func (g *Generator) generateInitFunction() {

View File

@ -33,3 +33,8 @@ package imp;
message PubliclyImportedMessage {
optional int64 field = 1;
}
enum PubliclyImportedEnum {
GLASSES = 1;
HAIR = 2;
}

View File

@ -66,9 +66,10 @@ message Request {
optional int32 group_field = 9;
}
// This foreign message type is in imp2.proto,
// These foreign types are in imp2.proto,
// which is publicly imported by imp.proto.
// optional imp.PubliclyImportedMessage pub = 10;
// optional imp.PubliclyImportedEnum pub_enum = 13 [default=HAIR];
optional int32 reset = 12;

View File

@ -1,24 +0,0 @@
# Compiled Object files, Static and Dynamic libs (Shared Objects)
*.o
*.a
*.so
# Folders
_obj
_test
# Architecture specific extensions/prefixes
*.[568vq]
[568vq].out
*.cgo1.go
*.cgo2.c
_cgo_defun.c
_cgo_gotypes.go
_cgo_export.*
_testmain.go
*.exe
coverage.html

View File

@ -1,8 +0,0 @@
language: go
go:
- 1.1
install:
- make dependencies

View File

@ -1,20 +0,0 @@
Copyright 2013 go-raft contributors
Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
"Software"), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:
The above copyright notice and this permission notice shall be
included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

View File

@ -1,13 +0,0 @@
all: test
coverage:
gocov test github.com/benbjohnson/go-raft | gocov-html > coverage.html
open coverage.html
dependencies:
go get -d .
test:
go test -v ./...
.PHONY: coverage dependencies test

View File

@ -1,69 +0,0 @@
[![Build Status](https://travis-ci.org/benbjohnson/go-raft.png?branch=master)](https://travis-ci.org/benbjohnson/go-raft)
go-raft
=======
## Overview
This is an Go implementation of the Raft distributed consensus protocol.
Raft is a protocol by which a cluster of nodes can maintain a replicated state machine.
The state machine is kept in sync through the use of a replicated log.
For more details on Raft, you can read [In Search of an Understandable Consensus Algorithm](https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf) by Diego Ongaro and John Ousterhout.
## The Raft Protocol
### Overview
Maintaining state in a single process on a single server is easy.
Your process is a single point of authority so there are no conflicts when reading and writing state.
Even multi-threaded processes can rely on locks or coroutines to serialize access to the data.
However, in a distributed system there is no single point of authority.
Servers can crash or the network between two machines can become unavailable or any number of other problems can occur.
A distributed consensus protocol is used for maintaining a consistent state across multiple servers in a cluster.
Many distributed systems are built upon the Paxos protocol but Paxos can be difficult to understand and there are many gaps between Paxos and real world implementation.
An alternative is the [Raft distributed consensus protocol](https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf) by Diego Ongaro and John Ousterhout.
Raft is a protocol built with understandability as a primary tenant and it centers around two things:
1. Leader Election
2. Replicated Log
With these two constructs, you can build a system that can maintain state across multiple servers -- even in the event of multiple failures.
### Leader Election
The Raft protocol effectively works as a master-slave system whereby state changes are written to a single server in the cluster and are distributed out to the rest of the servers in the cluster.
This simplifies the protocol since there is only one data authority and conflicts will not have to be resolved.
Raft ensures that there is only one leader at a time.
It does this by performing elections among the nodes in the cluster and requiring that a node must receive a majority of the votes in order to become leader.
For example, if you have 3 nodes in your cluster then a single node would need 2 votes in order to become the leader.
For a 5 node cluster, a server would need 3 votes to become leader.
### Replicated Log
To maintain state, a log of commands is maintained.
Each command makes a change to the state of the server and the command is deterministic.
By ensuring that this log is replicated identically between all the nodes in the cluster we can replicate the state at any point in time in the log by running each command sequentially.
Replicating the log under normal conditions is done by sending an `AppendEntries` RPC from the leader to each of the other servers in the cluster (called Peers).
Each peer will append the entries from the leader through a 2-phase commit process which ensure that a majority of servers in the cluster have entries written to log.
For a more detailed explanation on the failover process and election terms please see the full paper describing the protocol: [In Search of an Understandable Consensus Algorithm](https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf)
## Project Status
The go-raft library is feature complete but in alpha.
There is a reference implementation called [raftd](https://github.com/benbjohnson/raftd) that demonstrates how to use the library
The library will be considered experimental until it has significant production usage.
I'm writing the library for the purpose of including distributed processing in my behavioral analytics database called [Sky](https://github.com/skydb/sky).
However, I hope other projects can benefit from having a distributed consensus protocol so the go-raft library is available under MIT license.
If you have a project that you're using go-raft in, please add it to this README and send a pull request so others can see implementation examples.
If you have any questions on implementing go-raft in your project, feel free to contact me on [GitHub](https://github.com/benbjohnson), [Twitter](https://twitter.com/benbjohnson) or by e-mail at [ben@skylandlabs.com](mailto:ben@skylandlabs.com).

View File

@ -1,98 +0,0 @@
package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"io"
"io/ioutil"
)
// The request sent to a server to append entries to the log.
type AppendEntriesRequest struct {
Term uint64
PrevLogIndex uint64
PrevLogTerm uint64
CommitIndex uint64
LeaderName string
Entries []*LogEntry
}
// Creates a new AppendEntries request.
func newAppendEntriesRequest(term uint64, prevLogIndex uint64, prevLogTerm uint64, commitIndex uint64, leaderName string, entries []*LogEntry) *AppendEntriesRequest {
return &AppendEntriesRequest{
Term: term,
PrevLogIndex: prevLogIndex,
PrevLogTerm: prevLogTerm,
CommitIndex: commitIndex,
LeaderName: leaderName,
Entries: entries,
}
}
// Encodes the AppendEntriesRequest to a buffer. Returns the number of bytes
// written and any error that may have occurred.
func (req *AppendEntriesRequest) encode(w io.Writer) (int, error) {
protoEntries := make([]*protobuf.ProtoAppendEntriesRequest_ProtoLogEntry, len(req.Entries))
for i, entry := range req.Entries {
protoEntries[i] = &protobuf.ProtoAppendEntriesRequest_ProtoLogEntry{
Index: proto.Uint64(entry.Index),
Term: proto.Uint64(entry.Term),
CommandName: proto.String(entry.CommandName),
Command: entry.Command,
}
}
pb := &protobuf.ProtoAppendEntriesRequest{
Term: proto.Uint64(req.Term),
PrevLogIndex: proto.Uint64(req.PrevLogIndex),
PrevLogTerm: proto.Uint64(req.PrevLogTerm),
CommitIndex: proto.Uint64(req.CommitIndex),
LeaderName: proto.String(req.LeaderName),
Entries: protoEntries,
}
p, err := proto.Marshal(pb)
if err != nil {
return -1, err
}
return w.Write(p)
}
// Decodes the AppendEntriesRequest from a buffer. Returns the number of bytes read and
// any error that occurs.
func (req *AppendEntriesRequest) decode(r io.Reader) (int, error) {
data, err := ioutil.ReadAll(r)
if err != nil {
return -1, err
}
totalBytes := len(data)
pb := &protobuf.ProtoAppendEntriesRequest{}
if err := proto.Unmarshal(data, pb); err != nil {
return -1, err
}
req.Term = pb.GetTerm()
req.PrevLogIndex = pb.GetPrevLogIndex()
req.PrevLogTerm = pb.GetPrevLogTerm()
req.CommitIndex = pb.GetCommitIndex()
req.LeaderName = pb.GetLeaderName()
req.Entries = make([]*LogEntry, len(pb.Entries))
for i, entry := range pb.Entries {
req.Entries[i] = &LogEntry{
Index: entry.GetIndex(),
Term: entry.GetTerm(),
CommandName: entry.GetCommandName(),
Command: entry.Command,
}
}
return totalBytes, nil
}

View File

@ -1,40 +0,0 @@
package raft
import (
"bytes"
"testing"
)
func BenchmarkAppendEntriesRequestEncoding(b *testing.B) {
req, tmp := createTestAppendEntriesRequest(2000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
var buf bytes.Buffer
req.encode(&buf)
}
b.SetBytes(int64(len(tmp)))
}
func BenchmarkAppendEntriesRequestDecoding(b *testing.B) {
req, buf := createTestAppendEntriesRequest(2000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
req.decode(bytes.NewReader(buf))
}
b.SetBytes(int64(len(buf)))
}
func createTestAppendEntriesRequest(entryCount int) (*AppendEntriesRequest, []byte) {
entries := make([]*LogEntry, 0)
for i := 0; i < entryCount; i++ {
command := &DefaultJoinCommand{Name: "localhost:1000"}
entry, _ := newLogEntry(nil, 1, 2, command)
entries = append(entries, entry)
}
req := newAppendEntriesRequest(1, 1, 1, 1, "leader", entries)
var buf bytes.Buffer
req.encode(&buf)
return req, buf.Bytes()
}

View File

@ -1,70 +0,0 @@
package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"io"
"io/ioutil"
)
// The response returned from a server appending entries to the log.
type AppendEntriesResponse struct {
Term uint64
// the current index of the server
Index uint64
Success bool
CommitIndex uint64
peer string
append bool
}
// Creates a new AppendEntries response.
func newAppendEntriesResponse(term uint64, success bool, index uint64, commitIndex uint64) *AppendEntriesResponse {
return &AppendEntriesResponse{
Term: term,
Success: success,
Index: index,
CommitIndex: commitIndex,
}
}
// Encodes the AppendEntriesResponse to a buffer. Returns the number of bytes
// written and any error that may have occurred.
func (resp *AppendEntriesResponse) encode(w io.Writer) (int, error) {
pb := &protobuf.ProtoAppendEntriesResponse{
Term: proto.Uint64(resp.Term),
Index: proto.Uint64(resp.Index),
CommitIndex: proto.Uint64(resp.CommitIndex),
Success: proto.Bool(resp.Success),
}
p, err := proto.Marshal(pb)
if err != nil {
return -1, err
}
return w.Write(p)
}
// Decodes the AppendEntriesResponse from a buffer. Returns the number of bytes read and
// any error that occurs.
func (resp *AppendEntriesResponse) decode(r io.Reader) (int, error) {
data, err := ioutil.ReadAll(r)
if err != nil {
return -1, err
}
totalBytes := len(data)
pb := &protobuf.ProtoAppendEntriesResponse{}
if err := proto.Unmarshal(data, pb); err != nil {
return -1, err
}
resp.Term = pb.GetTerm()
resp.Index = pb.GetIndex()
resp.CommitIndex = pb.GetCommitIndex()
resp.Success = pb.GetSuccess()
return totalBytes, nil
}

View File

@ -1,34 +0,0 @@
package raft
import (
"bytes"
"testing"
)
func BenchmarkAppendEntriesResponseEncoding(b *testing.B) {
req, tmp := createTestAppendEntriesResponse(2000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
var buf bytes.Buffer
req.encode(&buf)
}
b.SetBytes(int64(len(tmp)))
}
func BenchmarkAppendEntriesResponseDecoding(b *testing.B) {
req, buf := createTestAppendEntriesResponse(2000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
req.decode(bytes.NewReader(buf))
}
b.SetBytes(int64(len(buf)))
}
func createTestAppendEntriesResponse(entryCount int) (*AppendEntriesResponse, []byte) {
resp := newAppendEntriesResponse(1, true, 1, 1)
var buf bytes.Buffer
resp.encode(&buf)
return resp, buf.Bytes()
}

View File

@ -1,92 +0,0 @@
package raft
import (
"bytes"
"encoding/json"
"fmt"
"io"
"reflect"
)
//------------------------------------------------------------------------------
//
// Globals
//
//------------------------------------------------------------------------------
var commandTypes map[string]Command
func init() {
commandTypes = map[string]Command{}
}
//------------------------------------------------------------------------------
//
// Typedefs
//
//------------------------------------------------------------------------------
// A command represents an action to be taken on the replicated state machine.
type Command interface {
CommandName() string
Apply(server *Server) (interface{}, error)
}
type CommandEncoder interface {
Encode(w io.Writer) error
Decode(r io.Reader) error
}
//------------------------------------------------------------------------------
//
// Functions
//
//------------------------------------------------------------------------------
//--------------------------------------
// Instantiation
//--------------------------------------
// Creates a new instance of a command by name.
func newCommand(name string, data []byte) (Command, error) {
// Find the registered command.
command := commandTypes[name]
if command == nil {
return nil, fmt.Errorf("raft.Command: Unregistered command type: %s", name)
}
// Make a copy of the command.
v := reflect.New(reflect.Indirect(reflect.ValueOf(command)).Type()).Interface()
copy, ok := v.(Command)
if !ok {
panic(fmt.Sprintf("raft: Unable to copy command: %s (%v)", command.CommandName(), reflect.ValueOf(v).Kind().String()))
}
// If data for the command was passed in the decode it.
if data != nil {
if encoder, ok := copy.(CommandEncoder); ok {
if err := encoder.Decode(bytes.NewReader(data)); err != nil {
return nil, err
}
} else {
json.NewDecoder(bytes.NewReader(data)).Decode(copy)
}
}
return copy, nil
}
//--------------------------------------
// Registration
//--------------------------------------
// Registers a command by storing a reference to an instance of it.
func RegisterCommand(command Command) {
if command == nil {
panic(fmt.Sprintf("raft: Cannot register nil"))
} else if commandTypes[command.CommandName()] != nil {
panic(fmt.Sprintf("raft: Duplicate registration: %s", command.CommandName()))
return
}
commandTypes[command.CommandName()] = command
}

View File

@ -1,116 +0,0 @@
package raft
import (
"log"
"os"
)
//------------------------------------------------------------------------------
//
// Variables
//
//------------------------------------------------------------------------------
const (
Debug = 1
Trace = 2
)
var logLevel int = 0
var logger *log.Logger
func init() {
logger = log.New(os.Stdout, "[raft]", log.Lmicroseconds)
}
//------------------------------------------------------------------------------
//
// Functions
//
//------------------------------------------------------------------------------
func LogLevel() int {
return logLevel
}
func SetLogLevel(level int) {
logLevel = level
}
//--------------------------------------
// Warnings
//--------------------------------------
// Prints to the standard logger. Arguments are handled in the manner of
// fmt.Print.
func warn(v ...interface{}) {
logger.Print(v...)
}
// Prints to the standard logger. Arguments are handled in the manner of
// fmt.Printf.
func warnf(format string, v ...interface{}) {
logger.Printf(format, v...)
}
// Prints to the standard logger. Arguments are handled in the manner of
// fmt.Println.
func warnln(v ...interface{}) {
logger.Println(v...)
}
//--------------------------------------
// Basic debugging
//--------------------------------------
// Prints to the standard logger if debug mode is enabled. Arguments
// are handled in the manner of fmt.Print.
func debug(v ...interface{}) {
if logLevel >= Debug {
logger.Print(v...)
}
}
// Prints to the standard logger if debug mode is enabled. Arguments
// are handled in the manner of fmt.Printf.
func debugf(format string, v ...interface{}) {
if logLevel >= Debug {
logger.Printf(format, v...)
}
}
// Prints to the standard logger if debug mode is enabled. Arguments
// are handled in the manner of fmt.Println.
func debugln(v ...interface{}) {
if logLevel >= Debug {
logger.Println(v...)
}
}
//--------------------------------------
// Trace-level debugging
//--------------------------------------
// Prints to the standard logger if trace debugging is enabled. Arguments
// are handled in the manner of fmt.Print.
func trace(v ...interface{}) {
if logLevel >= Trace {
logger.Print(v...)
}
}
// Prints to the standard logger if trace debugging is enabled. Arguments
// are handled in the manner of fmt.Printf.
func tracef(format string, v ...interface{}) {
if logLevel >= Trace {
logger.Printf(format, v...)
}
}
// Prints to the standard logger if trace debugging is enabled. Arguments
// are handled in the manner of debugln.
func traceln(v ...interface{}) {
if logLevel >= Trace {
logger.Println(v...)
}
}

View File

@ -1,195 +0,0 @@
package raft
import (
"bytes"
"fmt"
"io"
"net/http"
)
// Parts from this transporter were heavily influenced by Peter Bougon's
// raft implementation: https://github.com/peterbourgon/raft
//------------------------------------------------------------------------------
//
// Typedefs
//
//------------------------------------------------------------------------------
// An HTTPTransporter is a default transport layer used to communicate between
// multiple servers.
type HTTPTransporter struct {
DisableKeepAlives bool
prefix string
appendEntriesPath string
requestVotePath string
}
type HTTPMuxer interface {
HandleFunc(string, func(http.ResponseWriter, *http.Request))
}
//------------------------------------------------------------------------------
//
// Constructor
//
//------------------------------------------------------------------------------
// Creates a new HTTP transporter with the given path prefix.
func NewHTTPTransporter(prefix string) *HTTPTransporter {
return &HTTPTransporter{
DisableKeepAlives: false,
prefix: prefix,
appendEntriesPath: fmt.Sprintf("%s%s", prefix, "/appendEntries"),
requestVotePath: fmt.Sprintf("%s%s", prefix, "/requestVote"),
}
}
//------------------------------------------------------------------------------
//
// Accessors
//
//------------------------------------------------------------------------------
// Retrieves the path prefix used by the transporter.
func (t *HTTPTransporter) Prefix() string {
return t.prefix
}
// Retrieves the AppendEntries path.
func (t *HTTPTransporter) AppendEntriesPath() string {
return t.appendEntriesPath
}
// Retrieves the RequestVote path.
func (t *HTTPTransporter) RequestVotePath() string {
return t.requestVotePath
}
//------------------------------------------------------------------------------
//
// Methods
//
//------------------------------------------------------------------------------
//--------------------------------------
// Installation
//--------------------------------------
// Applies Raft routes to an HTTP router for a given server.
func (t *HTTPTransporter) Install(server *Server, mux HTTPMuxer) {
mux.HandleFunc(t.AppendEntriesPath(), t.appendEntriesHandler(server))
mux.HandleFunc(t.RequestVotePath(), t.requestVoteHandler(server))
}
//--------------------------------------
// Outgoing
//--------------------------------------
// Sends an AppendEntries RPC to a peer.
func (t *HTTPTransporter) SendAppendEntriesRequest(server *Server, peer *Peer, req *AppendEntriesRequest) *AppendEntriesResponse {
var b bytes.Buffer
if _, err := req.encode(&b); err != nil {
traceln("transporter.ae.encoding.error:", err)
return nil
}
url := fmt.Sprintf("http://%s%s", peer.Name(), t.AppendEntriesPath())
traceln(server.Name(), "POST", url)
client := &http.Client{Transport: &http.Transport{DisableKeepAlives: t.DisableKeepAlives}}
httpResp, err := client.Post(url, "application/protobuf", &b)
if httpResp == nil || err != nil {
traceln("transporter.ae.response.error:", err)
return nil
}
defer httpResp.Body.Close()
resp := &AppendEntriesResponse{}
if _, err = resp.decode(httpResp.Body); err != nil && err != io.EOF {
traceln("transporter.ae.decoding.error:", err)
return nil
}
return resp
}
// Sends a RequestVote RPC to a peer.
func (t *HTTPTransporter) SendVoteRequest(server *Server, peer *Peer, req *RequestVoteRequest) *RequestVoteResponse {
var b bytes.Buffer
if _, err := req.encode(&b); err != nil {
traceln("transporter.rv.encoding.error:", err)
return nil
}
url := fmt.Sprintf("http://%s%s", peer.Name(), t.RequestVotePath())
traceln(server.Name(), "POST", url)
client := &http.Client{Transport: &http.Transport{DisableKeepAlives: t.DisableKeepAlives}}
httpResp, err := client.Post(url, "application/protobuf", &b)
if httpResp == nil || err != nil {
traceln("transporter.rv.response.error:", err)
return nil
}
defer httpResp.Body.Close()
resp := &RequestVoteResponse{}
if _, err = resp.decode(httpResp.Body); err != nil && err != io.EOF {
traceln("transporter.rv.decoding.error:", err)
return nil
}
return resp
}
// Sends a SnapshotRequest RPC to a peer.
func (t *HTTPTransporter) SendSnapshotRequest(server *Server, peer *Peer, req *SnapshotRequest) *SnapshotResponse {
return nil
}
// Sends a SnapshotRequest RPC to a peer.
func (t *HTTPTransporter) SendSnapshotRecoveryRequest(server *Server, peer *Peer, req *SnapshotRecoveryRequest) *SnapshotRecoveryResponse {
return nil
}
//--------------------------------------
// Incoming
//--------------------------------------
// Handles incoming AppendEntries requests.
func (t *HTTPTransporter) appendEntriesHandler(server *Server) http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
traceln(server.Name(), "RECV /appendEntries")
req := &AppendEntriesRequest{}
if _, err := req.decode(r.Body); err != nil {
http.Error(w, "", http.StatusBadRequest)
return
}
resp := server.AppendEntries(req)
if _, err := resp.encode(w); err != nil {
http.Error(w, "", http.StatusInternalServerError)
return
}
}
}
// Handles incoming RequestVote requests.
func (t *HTTPTransporter) requestVoteHandler(server *Server) http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
traceln(server.Name(), "RECV /requestVote")
req := &RequestVoteRequest{}
if _, err := req.decode(r.Body); err != nil {
http.Error(w, "", http.StatusBadRequest)
return
}
resp := server.RequestVote(req)
if _, err := resp.encode(w); err != nil {
http.Error(w, "", http.StatusInternalServerError)
return
}
}
}

View File

@ -1,153 +0,0 @@
package raft
import (
"fmt"
"net"
"net/http"
"sync"
"testing"
"time"
)
// Ensure that we can start several servers and have them communicate.
func TestHTTPTransporter(t *testing.T) {
transporter := NewHTTPTransporter("/raft")
transporter.DisableKeepAlives = true
servers := []*Server{}
f0 := func(server *Server, httpServer *http.Server) {
// Stop the leader and wait for an election.
server.Stop()
time.Sleep(testElectionTimeout * 2)
if servers[1].State() != Leader && servers[2].State() != Leader {
t.Fatal("Expected re-election:", servers[1].State(), servers[2].State())
}
server.Start()
}
f1 := func(server *Server, httpServer *http.Server) {
}
f2 := func(server *Server, httpServer *http.Server) {
}
runTestHttpServers(t, &servers, transporter, f0, f1, f2)
}
// Starts multiple independent Raft servers wrapped with HTTP servers.
func runTestHttpServers(t *testing.T, servers *[]*Server, transporter *HTTPTransporter, callbacks ...func(*Server, *http.Server)) {
var wg sync.WaitGroup
httpServers := []*http.Server{}
listeners := []net.Listener{}
for i := range callbacks {
wg.Add(1)
port := 9000 + i
// Create raft server.
server := newTestServer(fmt.Sprintf("localhost:%d", port), transporter)
server.SetHeartbeatTimeout(testHeartbeatTimeout)
server.SetElectionTimeout(testElectionTimeout)
server.Start()
defer server.Stop()
*servers = append(*servers, server)
// Create listener for HTTP server and start it.
listener, err := net.Listen("tcp", fmt.Sprintf(":%d", port))
if err != nil {
panic(err)
}
defer listener.Close()
listeners = append(listeners, listener)
// Create wrapping HTTP server.
mux := http.NewServeMux()
transporter.Install(server, mux)
httpServer := &http.Server{Addr: fmt.Sprintf(":%d", port), Handler: mux}
httpServers = append(httpServers, httpServer)
go func() { httpServer.Serve(listener) }()
}
// Setup configuration.
for _, server := range *servers {
if _, err := (*servers)[0].Do(&DefaultJoinCommand{Name: server.Name()}); err != nil {
t.Fatalf("Server %s unable to join: %v", server.Name(), err)
}
}
// Wait for configuration to propagate.
time.Sleep(testHeartbeatTimeout * 2)
// Execute all the callbacks at the same time.
for _i, _f := range callbacks {
i, f := _i, _f
go func() {
defer wg.Done()
f((*servers)[i], httpServers[i])
}()
}
// Wait until everything is done.
wg.Wait()
}
func BenchmarkSpeed(b *testing.B) {
transporter := NewHTTPTransporter("/raft")
transporter.DisableKeepAlives = true
servers := []*Server{}
for i := 0; i < 3; i++ {
port := 9000 + i
// Create raft server.
server := newTestServer(fmt.Sprintf("localhost:%d", port), transporter)
server.SetHeartbeatTimeout(testHeartbeatTimeout)
server.SetElectionTimeout(testElectionTimeout)
server.Start()
defer server.Stop()
servers = append(servers, server)
// Create listener for HTTP server and start it.
listener, err := net.Listen("tcp", fmt.Sprintf(":%d", port))
if err != nil {
panic(err)
}
defer listener.Close()
// Create wrapping HTTP server.
mux := http.NewServeMux()
transporter.Install(server, mux)
httpServer := &http.Server{Addr: fmt.Sprintf(":%d", port), Handler: mux}
go func() { httpServer.Serve(listener) }()
}
// Setup configuration.
for _, server := range servers {
(servers)[0].Do(&DefaultJoinCommand{Name: server.Name()})
}
c := make(chan bool)
// Wait for configuration to propagate.
time.Sleep(testHeartbeatTimeout * 2)
b.ResetTimer()
for n := 0; n < b.N; n++ {
for i := 0; i < 1000; i++ {
go send(c, servers[0])
}
for i := 0; i < 1000; i++ {
<-c
}
}
}
func send(c chan bool, s *Server) {
for i := 0; i < 20; i++ {
s.Do(&NOPCommand{})
}
c <- true
}

View File

@ -1,28 +0,0 @@
package raft
// Join command interface
type JoinCommand interface {
CommandName() string
Apply(server *Server) (interface{}, error)
NodeName() string
}
// Join command
type DefaultJoinCommand struct {
Name string `json:"name"`
}
// The name of the Join command in the log
func (c *DefaultJoinCommand) CommandName() string {
return "raft:join"
}
func (c *DefaultJoinCommand) Apply(server *Server) (interface{}, error) {
err := server.AddPeer(c.Name)
return []byte("join"), err
}
func (c *DefaultJoinCommand) NodeName() string {
return c.Name
}

View File

@ -1,27 +0,0 @@
package raft
// Leave command interface
type LeaveCommand interface {
CommandName() string
Apply(server *Server) (interface{}, error)
NodeName() string
}
// Leave command
type DefaultLeaveCommand struct {
Name string `json:"name"`
}
// The name of the Leave command in the log
func (c *DefaultLeaveCommand) CommandName() string {
return "raft:leave"
}
func (c *DefaultLeaveCommand) Apply(server *Server) (interface{}, error) {
err := server.RemovePeer(c.Name)
return []byte("leave"), err
}
func (c *DefaultLeaveCommand) NodeName() string {
return c.Name
}

View File

@ -1,616 +0,0 @@
package raft
import (
"bufio"
"code.google.com/p/goprotobuf/proto"
"errors"
"fmt"
"github.com/benbjohnson/go-raft/protobuf"
"io"
"os"
"sync"
)
//------------------------------------------------------------------------------
//
// Typedefs
//
//------------------------------------------------------------------------------
// A log is a collection of log entries that are persisted to durable storage.
type Log struct {
ApplyFunc func(Command) (interface{}, error)
file *os.File
path string
entries []*LogEntry
results []*logResult
commitIndex uint64
mutex sync.RWMutex
startIndex uint64 // the index before the first entry in the Log entries
startTerm uint64
pBuffer *proto.Buffer
pLogEntry *protobuf.ProtoLogEntry
}
// The results of the applying a log entry.
type logResult struct {
returnValue interface{}
err error
}
//------------------------------------------------------------------------------
//
// Constructor
//
//------------------------------------------------------------------------------
// Creates a new log.
func newLog() *Log {
return &Log{
entries: make([]*LogEntry, 0),
pBuffer: proto.NewBuffer(nil),
pLogEntry: &protobuf.ProtoLogEntry{},
}
}
//------------------------------------------------------------------------------
//
// Accessors
//
//------------------------------------------------------------------------------
//--------------------------------------
// Log Indices
//--------------------------------------
// The last committed index in the log.
func (l *Log) CommitIndex() uint64 {
l.mutex.RLock()
defer l.mutex.RUnlock()
return l.commitIndex
}
// The current index in the log.
func (l *Log) currentIndex() uint64 {
l.mutex.RLock()
defer l.mutex.RUnlock()
if len(l.entries) == 0 {
return l.startIndex
}
return l.entries[len(l.entries)-1].Index
}
// The current index in the log without locking
func (l *Log) internalCurrentIndex() uint64 {
if len(l.entries) == 0 {
return l.startIndex
}
return l.entries[len(l.entries)-1].Index
}
// The next index in the log.
func (l *Log) nextIndex() uint64 {
return l.currentIndex() + 1
}
// Determines if the log contains zero entries.
func (l *Log) isEmpty() bool {
l.mutex.RLock()
defer l.mutex.RUnlock()
return (len(l.entries) == 0) && (l.startIndex == 0)
}
// The name of the last command in the log.
func (l *Log) lastCommandName() string {
l.mutex.RLock()
defer l.mutex.RUnlock()
if len(l.entries) > 0 {
if entry := l.entries[len(l.entries)-1]; entry != nil {
return entry.CommandName
}
}
return ""
}
//--------------------------------------
// Log Terms
//--------------------------------------
// The current term in the log.
func (l *Log) currentTerm() uint64 {
l.mutex.RLock()
defer l.mutex.RUnlock()
if len(l.entries) == 0 {
return l.startTerm
}
return l.entries[len(l.entries)-1].Term
}
//------------------------------------------------------------------------------
//
// Methods
//
//------------------------------------------------------------------------------
//--------------------------------------
// State
//--------------------------------------
// Opens the log file and reads existing entries. The log can remain open and
// continue to append entries to the end of the log.
func (l *Log) open(path string) error {
// Read all the entries from the log if one exists.
var readBytes int64
var err error
debugln("log.open.open ", path)
// open log file
l.file, err = os.OpenFile(path, os.O_RDWR, 0600)
l.path = path
if err != nil {
// if the log file does not exist before
// we create the log file and set commitIndex to 0
if os.IsNotExist(err) {
l.file, err = os.OpenFile(path, os.O_WRONLY|os.O_CREATE, 0600)
debugln("log.open.create ", path)
return err
}
return err
}
debugln("log.open.exist ", path)
// Read the file and decode entries.
for {
// Instantiate log entry and decode into it.
entry, _ := newLogEntry(l, 0, 0, nil)
entry.Position, _ = l.file.Seek(0, os.SEEK_CUR)
n, err := entry.decode(l.file)
if err != nil {
if err == io.EOF {
debugln("open.log.append: finish ")
} else {
if err = os.Truncate(path, readBytes); err != nil {
return fmt.Errorf("raft.Log: Unable to recover: %v", err)
}
}
break
}
// Append entry.
l.entries = append(l.entries, entry)
debugln("open.log.append log index ", entry.Index)
readBytes += int64(n)
}
l.results = make([]*logResult, len(l.entries))
l.compact(l.startIndex, l.startTerm)
debugln("open.log.recovery number of log ", len(l.entries))
return nil
}
// Closes the log file.
func (l *Log) close() {
l.mutex.Lock()
defer l.mutex.Unlock()
if l.file != nil {
l.file.Close()
l.file = nil
}
l.entries = make([]*LogEntry, 0)
l.results = make([]*logResult, 0)
}
//--------------------------------------
// Entries
//--------------------------------------
// Creates a log entry associated with this log.
func (l *Log) createEntry(term uint64, command Command) (*LogEntry, error) {
return newLogEntry(l, l.nextIndex(), term, command)
}
// Retrieves an entry from the log. If the entry has been eliminated because
// of a snapshot then nil is returned.
func (l *Log) getEntry(index uint64) *LogEntry {
l.mutex.RLock()
defer l.mutex.RUnlock()
if index <= l.startIndex || index > (l.startIndex+uint64(len(l.entries))) {
return nil
}
return l.entries[index-l.startIndex-1]
}
// Checks if the log contains a given index/term combination.
func (l *Log) containsEntry(index uint64, term uint64) bool {
entry := l.getEntry(index)
return (entry != nil && entry.Term == term)
}
// Retrieves a list of entries after a given index as well as the term of the
// index provided. A nil list of entries is returned if the index no longer
// exists because a snapshot was made.
func (l *Log) getEntriesAfter(index uint64, maxLogEntriesPerRequest uint64) ([]*LogEntry, uint64) {
l.mutex.Lock()
defer l.mutex.Unlock()
// Return nil if index is before the start of the log.
if index < l.startIndex {
traceln("log.entriesAfter.before: ", index, " ", l.startIndex)
return nil, 0
}
// Return an error if the index doesn't exist.
if index > (uint64(len(l.entries)) + l.startIndex) {
panic(fmt.Sprintf("raft: Index is beyond end of log: %v %v", len(l.entries), index))
}
// If we're going from the beginning of the log then return the whole log.
if index == l.startIndex {
traceln("log.entriesAfter.beginning: ", index, " ", l.startIndex)
return l.entries, l.startTerm
}
traceln("log.entriesAfter.partial: ", index, " ", l.entries[len(l.entries)-1].Index)
entries := l.entries[index-l.startIndex:]
length := len(entries)
if uint64(length) < maxLogEntriesPerRequest {
// Determine the term at the given entry and return a subslice.
return entries, l.entries[index-1-l.startIndex].Term
} else {
return entries[:maxLogEntriesPerRequest], l.entries[index-1-l.startIndex].Term
}
}
// Retrieves the return value and error for an entry. The result can only exist
// after the entry has been committed.
func (l *Log) getEntryResult(entry *LogEntry, clear bool) (interface{}, error) {
l.mutex.RLock()
defer l.mutex.RUnlock()
if entry == nil {
panic("raft: Log entry required for error retrieval")
}
debugln("getEntryResult.result index: ", entry.Index-l.startIndex-1)
// If a result exists for the entry then return it with its error.
if entry.Index > l.startIndex && entry.Index <= l.startIndex+uint64(len(l.results)) {
if result := l.results[entry.Index-l.startIndex-1]; result != nil {
// keep the records before remove it
returnValue, err := result.returnValue, result.err
// Remove reference to result if it's being cleared after retrieval.
if clear {
result.returnValue = nil
}
return returnValue, err
}
}
return nil, nil
}
//--------------------------------------
// Commit
//--------------------------------------
// Retrieves the last index and term that has been committed to the log.
func (l *Log) commitInfo() (index uint64, term uint64) {
l.mutex.RLock()
defer l.mutex.RUnlock()
// If we don't have any committed entries then just return zeros.
if l.commitIndex == 0 {
return 0, 0
}
// No new commit log after snapshot
if l.commitIndex == l.startIndex {
return l.startIndex, l.startTerm
}
// Return the last index & term from the last committed entry.
debugln("commitInfo.get.[", l.commitIndex, "/", l.startIndex, "]")
entry := l.entries[l.commitIndex-1-l.startIndex]
return entry.Index, entry.Term
}
// Retrieves the last index and term that has been committed to the log.
func (l *Log) lastInfo() (index uint64, term uint64) {
l.mutex.RLock()
defer l.mutex.RUnlock()
// If we don't have any entries then just return zeros.
if len(l.entries) == 0 {
return l.startIndex, l.startTerm
}
// Return the last index & term
entry := l.entries[len(l.entries)-1]
return entry.Index, entry.Term
}
// Updates the commit index
func (l *Log) updateCommitIndex(index uint64) {
l.mutex.Lock()
defer l.mutex.Unlock()
l.commitIndex = index
}
// Updates the commit index and writes entries after that index to the stable storage.
func (l *Log) setCommitIndex(index uint64) error {
l.mutex.Lock()
defer l.mutex.Unlock()
// this is not error any more after limited the number of sending entries
// commit up to what we already have
if index > l.startIndex+uint64(len(l.entries)) {
debugln("raft.Log: Commit index", index, "set back to ", len(l.entries))
index = l.startIndex + uint64(len(l.entries))
}
// Do not allow previous indices to be committed again.
// This could happens, since the guarantee is that the new leader has up-to-dated
// log entires rather than has most up-to-dated committed index
// For example, Leader 1 send log 80 to follower 2 and follower 3
// follower 2 and follow 3 all got the new entries and reply
// leader 1 committed entry 80 and send reply to follower 2 and follower3
// follower 2 receive the new committed index and update committed index to 80
// leader 1 fail to send the committed index to follower 3
// follower 3 promote to leader (server 1 and server 2 will vote, since leader 3
// has up-to-dated the entries)
// when new leader 3 send heartbeat with committed index = 0 to follower 2,
// follower 2 should reply success and let leader 3 update the committed index to 80
if index < l.commitIndex {
return nil
}
// Find all entries whose index is between the previous index and the current index.
for i := l.commitIndex + 1; i <= index; i++ {
entryIndex := i - 1 - l.startIndex
entry := l.entries[entryIndex]
// Update commit index.
l.commitIndex = entry.Index
// Decode the command.
command, err := newCommand(entry.CommandName, entry.Command)
if err != nil {
return err
}
// Apply the changes to the state machine and store the error code.
returnValue, err := l.ApplyFunc(command)
debugln("setCommitIndex.set.result index: ", entryIndex)
l.results[entryIndex] = &logResult{returnValue: returnValue, err: err}
}
return nil
}
// Set the commitIndex at the head of the log file to the current
// commit Index. This should be called after obtained a log lock
func (l *Log) flushCommitIndex() {
l.file.Seek(0, os.SEEK_SET)
fmt.Fprintf(l.file, "%8x\n", l.commitIndex)
l.file.Seek(0, os.SEEK_END)
}
//--------------------------------------
// Truncation
//--------------------------------------
// Truncates the log to the given index and term. This only works if the log
// at the index has not been committed.
func (l *Log) truncate(index uint64, term uint64) error {
l.mutex.Lock()
defer l.mutex.Unlock()
debugln("log.truncate: ", index)
// Do not allow committed entries to be truncated.
if index < l.commitIndex {
debugln("log.truncate.before")
return fmt.Errorf("raft.Log: Index is already committed (%v): (IDX=%v, TERM=%v)", l.commitIndex, index, term)
}
// Do not truncate past end of entries.
if index > l.startIndex+uint64(len(l.entries)) {
debugln("log.truncate.after")
return fmt.Errorf("raft.Log: Entry index does not exist (MAX=%v): (IDX=%v, TERM=%v)", len(l.entries), index, term)
}
// If we're truncating everything then just clear the entries.
if index == l.startIndex {
debugln("log.truncate.clear")
l.file.Truncate(0)
l.file.Seek(0, os.SEEK_SET)
l.entries = []*LogEntry{}
} else {
// Do not truncate if the entry at index does not have the matching term.
entry := l.entries[index-l.startIndex-1]
if len(l.entries) > 0 && entry.Term != term {
debugln("log.truncate.termMismatch")
return fmt.Errorf("raft.Log: Entry at index does not have matching term (%v): (IDX=%v, TERM=%v)", entry.Term, index, term)
}
// Otherwise truncate up to the desired entry.
if index < l.startIndex+uint64(len(l.entries)) {
debugln("log.truncate.finish")
position := l.entries[index-l.startIndex].Position
l.file.Truncate(position)
l.file.Seek(position, os.SEEK_SET)
l.entries = l.entries[0 : index-l.startIndex]
}
}
return nil
}
//--------------------------------------
// Append
//--------------------------------------
// Appends a series of entries to the log. These entries are not written to
// disk until setCommitIndex() is called.
func (l *Log) appendEntries(entries []*LogEntry) error {
l.mutex.Lock()
defer l.mutex.Unlock()
startPosition, _ := l.file.Seek(0, os.SEEK_CUR)
w := bufio.NewWriter(l.file)
var size int64
var err error
// Append each entry but exit if we hit an error.
for _, entry := range entries {
entry.log = l
if size, err = l.writeEntry(entry, w); err != nil {
return err
}
entry.Position = startPosition
startPosition += size
}
w.Flush()
return nil
}
// Writes a single log entry to the end of the log. This function does not
// obtain a lock and should only be used internally. Use AppendEntries() and
// AppendEntry() to use it externally.
func (l *Log) appendEntry(entry *LogEntry) error {
if l.file == nil {
return errors.New("raft.Log: Log is not open")
}
// Make sure the term and index are greater than the previous.
if len(l.entries) > 0 {
lastEntry := l.entries[len(l.entries)-1]
if entry.Term < lastEntry.Term {
return fmt.Errorf("raft.Log: Cannot append entry with earlier term (%x:%x <= %x:%x)", entry.Term, entry.Index, lastEntry.Term, lastEntry.Index)
} else if entry.Term == lastEntry.Term && entry.Index <= lastEntry.Index {
return fmt.Errorf("raft.Log: Cannot append entry with earlier index in the same term (%x:%x <= %x:%x)", entry.Term, entry.Index, lastEntry.Term, lastEntry.Index)
}
}
position, _ := l.file.Seek(0, os.SEEK_CUR)
entry.Position = position
// Write to storage.
if _, err := entry.encode(l.file); err != nil {
return err
}
// Append to entries list if stored on disk.
l.entries = append(l.entries, entry)
l.results = append(l.results, nil)
return nil
}
// appendEntry with Buffered io
func (l *Log) writeEntry(entry *LogEntry, w io.Writer) (int64, error) {
if l.file == nil {
return -1, errors.New("raft.Log: Log is not open")
}
// Make sure the term and index are greater than the previous.
if len(l.entries) > 0 {
lastEntry := l.entries[len(l.entries)-1]
if entry.Term < lastEntry.Term {
return -1, fmt.Errorf("raft.Log: Cannot append entry with earlier term (%x:%x <= %x:%x)", entry.Term, entry.Index, lastEntry.Term, lastEntry.Index)
} else if entry.Term == lastEntry.Term && entry.Index <= lastEntry.Index {
return -1, fmt.Errorf("raft.Log: Cannot append entry with earlier index in the same term (%x:%x <= %x:%x)", entry.Term, entry.Index, lastEntry.Term, lastEntry.Index)
}
}
// Write to storage.
size, err := entry.encode(w)
if err != nil {
return -1, err
}
// Append to entries list if stored on disk.
l.entries = append(l.entries, entry)
l.results = append(l.results, nil)
return int64(size), nil
}
//--------------------------------------
// Log compaction
//--------------------------------------
// compact the log before index (including index)
func (l *Log) compact(index uint64, term uint64) error {
var entries []*LogEntry
var results []*logResult
l.mutex.Lock()
defer l.mutex.Unlock()
if index == 0 {
return nil
}
// nothing to compaction
// the index may be greater than the current index if
// we just recovery from on snapshot
if index >= l.internalCurrentIndex() {
entries = make([]*LogEntry, 0)
results = make([]*logResult, 0)
} else {
// get all log entries after index
entries = l.entries[index-l.startIndex:]
results = l.results[index-l.startIndex:]
}
// create a new log file and add all the entries
file, err := os.OpenFile(l.path+".new", os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0600)
if err != nil {
return err
}
for _, entry := range entries {
position, _ := l.file.Seek(0, os.SEEK_CUR)
entry.Position = position
if _, err = entry.encode(file); err != nil {
return err
}
}
// close the current log file
l.file.Close()
// remove the current log file to .bak
err = os.Remove(l.path)
if err != nil {
return err
}
// rename the new log file
err = os.Rename(l.path+".new", l.path)
if err != nil {
return err
}
l.file = file
// compaction the in memory log
l.entries = entries
l.results = results
l.startIndex = index
l.startTerm = term
return nil
}

View File

@ -1,99 +0,0 @@
package raft
import (
"bytes"
"code.google.com/p/goprotobuf/proto"
"encoding/json"
"fmt"
"github.com/benbjohnson/go-raft/protobuf"
"io"
)
// A log entry stores a single item in the log.
type LogEntry struct {
log *Log
Index uint64
Term uint64
CommandName string
Command []byte
Position int64 // position in the log file
commit chan bool
}
// Creates a new log entry associated with a log.
func newLogEntry(log *Log, index uint64, term uint64, command Command) (*LogEntry, error) {
var buf bytes.Buffer
var commandName string
if command != nil {
commandName = command.CommandName()
if encoder, ok := command.(CommandEncoder); ok {
if err := encoder.Encode(&buf); err != nil {
return nil, err
}
} else {
json.NewEncoder(&buf).Encode(command)
}
}
e := &LogEntry{
log: log,
Index: index,
Term: term,
CommandName: commandName,
Command: buf.Bytes(),
commit: make(chan bool, 5),
}
return e, nil
}
// Encodes the log entry to a buffer. Returns the number of bytes
// written and any error that may have occurred.
func (e *LogEntry) encode(w io.Writer) (int, error) {
defer e.log.pBuffer.Reset()
e.log.pLogEntry.Index = proto.Uint64(e.Index)
e.log.pLogEntry.Term = proto.Uint64(e.Term)
e.log.pLogEntry.CommandName = proto.String(e.CommandName)
e.log.pLogEntry.Command = e.Command
err := e.log.pBuffer.Marshal(e.log.pLogEntry)
if err != nil {
return -1, err
}
if _, err = fmt.Fprintf(w, "%8x\n", len(e.log.pBuffer.Bytes())); err != nil {
return -1, err
}
return w.Write(e.log.pBuffer.Bytes())
}
// Decodes the log entry from a buffer. Returns the number of bytes read and
// any error that occurs.
func (e *LogEntry) decode(r io.Reader) (int, error) {
var length int
_, err := fmt.Fscanf(r, "%8x\n", &length)
if err != nil {
return -1, err
}
data := make([]byte, length)
_, err = r.Read(data)
if err != nil {
return -1, err
}
pb := &protobuf.ProtoLogEntry{}
if err = proto.Unmarshal(data, pb); err != nil {
return -1, err
}
e.Term = pb.GetTerm()
e.Index = pb.GetIndex()
e.CommandName = pb.GetCommandName()
e.Command = pb.Command
return length, nil
}

View File

@ -1,232 +0,0 @@
package raft
import (
"io/ioutil"
"os"
"reflect"
"testing"
)
//------------------------------------------------------------------------------
//
// Tests
//
//------------------------------------------------------------------------------
//--------------------------------------
// Append
//--------------------------------------
// Ensure that we can append to a new log.
func TestLogNewLog(t *testing.T) {
path := getLogPath()
log := newLog()
log.ApplyFunc = func(c Command) (interface{}, error) {
return nil, nil
}
if err := log.open(path); err != nil {
t.Fatalf("Unable to open log: %v", err)
}
defer log.close()
defer os.Remove(path)
e, _ := newLogEntry(log, 1, 1, &testCommand1{Val: "foo", I: 20})
if err := log.appendEntry(e); err != nil {
t.Fatalf("Unable to append: %v", err)
}
e, _ = newLogEntry(log, 2, 1, &testCommand2{X: 100})
if err := log.appendEntry(e); err != nil {
t.Fatalf("Unable to append: %v", err)
}
e, _ = newLogEntry(log, 3, 2, &testCommand1{Val: "bar", I: 0})
if err := log.appendEntry(e); err != nil {
t.Fatalf("Unable to append: %v", err)
}
// Partial commit.
if err := log.setCommitIndex(2); err != nil {
t.Fatalf("Unable to partially commit: %v", err)
}
if index, term := log.commitInfo(); index != 2 || term != 1 {
t.Fatalf("Invalid commit info [IDX=%v, TERM=%v]", index, term)
}
// Full commit.
if err := log.setCommitIndex(3); err != nil {
t.Fatalf("Unable to commit: %v", err)
}
if index, term := log.commitInfo(); index != 3 || term != 2 {
t.Fatalf("Invalid commit info [IDX=%v, TERM=%v]", index, term)
}
}
// Ensure that we can decode and encode to an existing log.
func TestLogExistingLog(t *testing.T) {
tmpLog := newLog()
e0, _ := newLogEntry(tmpLog, 1, 1, &testCommand1{Val: "foo", I: 20})
e1, _ := newLogEntry(tmpLog, 2, 1, &testCommand2{X: 100})
e2, _ := newLogEntry(tmpLog, 3, 2, &testCommand1{Val: "bar", I: 0})
log, path := setupLog([]*LogEntry{e0, e1, e2})
defer log.close()
defer os.Remove(path)
// Validate existing log entries.
if len(log.entries) != 3 {
t.Fatalf("Expected 3 entries, got %d", len(log.entries))
}
if log.entries[0].Index != 1 || log.entries[0].Term != 1 {
t.Fatalf("Unexpected entry[0]: %v", log.entries[0])
}
if log.entries[1].Index != 2 || log.entries[1].Term != 1 {
t.Fatalf("Unexpected entry[1]: %v", log.entries[1])
}
if log.entries[2].Index != 3 || log.entries[2].Term != 2 {
t.Fatalf("Unexpected entry[2]: %v", log.entries[2])
}
}
// Ensure that we can check the contents of the log by index/term.
func TestLogContainsEntries(t *testing.T) {
tmpLog := newLog()
e0, _ := newLogEntry(tmpLog, 1, 1, &testCommand1{Val: "foo", I: 20})
e1, _ := newLogEntry(tmpLog, 2, 1, &testCommand2{X: 100})
e2, _ := newLogEntry(tmpLog, 3, 2, &testCommand1{Val: "bar", I: 0})
log, path := setupLog([]*LogEntry{e0, e1, e2})
defer log.close()
defer os.Remove(path)
if log.containsEntry(0, 0) {
t.Fatalf("Zero-index entry should not exist in log.")
}
if log.containsEntry(1, 0) {
t.Fatalf("Entry with mismatched term should not exist")
}
if log.containsEntry(4, 0) {
t.Fatalf("Out-of-range entry should not exist")
}
if !log.containsEntry(2, 1) {
t.Fatalf("Entry 2/1 should exist")
}
if !log.containsEntry(3, 2) {
t.Fatalf("Entry 2/1 should exist")
}
}
// Ensure that we can recover from an incomplete/corrupt log and continue logging.
func TestLogRecovery(t *testing.T) {
tmpLog := newLog()
e0, _ := newLogEntry(tmpLog, 1, 1, &testCommand1{Val: "foo", I: 20})
e1, _ := newLogEntry(tmpLog, 2, 1, &testCommand2{X: 100})
f, _ := ioutil.TempFile("", "raft-log-")
e0.encode(f)
e1.encode(f)
f.WriteString("CORRUPT!")
f.Close()
log := newLog()
log.ApplyFunc = func(c Command) (interface{}, error) {
return nil, nil
}
if err := log.open(f.Name()); err != nil {
t.Fatalf("Unable to open log: %v", err)
}
defer log.close()
defer os.Remove(f.Name())
e, _ := newLogEntry(log, 3, 2, &testCommand1{Val: "bat", I: -5})
if err := log.appendEntry(e); err != nil {
t.Fatalf("Unable to append: %v", err)
}
// Validate existing log entries.
if len(log.entries) != 3 {
t.Fatalf("Expected 3 entries, got %d", len(log.entries))
}
if log.entries[0].Index != 1 || log.entries[0].Term != 1 {
t.Fatalf("Unexpected entry[0]: %v", log.entries[0])
}
if log.entries[1].Index != 2 || log.entries[1].Term != 1 {
t.Fatalf("Unexpected entry[1]: %v", log.entries[1])
}
if log.entries[2].Index != 3 || log.entries[2].Term != 2 {
t.Fatalf("Unexpected entry[2]: %v", log.entries[2])
}
}
//--------------------------------------
// Append
//--------------------------------------
// Ensure that we can truncate uncommitted entries in the log.
func TestLogTruncate(t *testing.T) {
log, path := setupLog(nil)
if err := log.open(path); err != nil {
t.Fatalf("Unable to open log: %v", err)
}
defer os.Remove(path)
entry1, _ := newLogEntry(log, 1, 1, &testCommand1{Val: "foo", I: 20})
if err := log.appendEntry(entry1); err != nil {
t.Fatalf("Unable to append: %v", err)
}
entry2, _ := newLogEntry(log, 2, 1, &testCommand2{X: 100})
if err := log.appendEntry(entry2); err != nil {
t.Fatalf("Unable to append: %v", err)
}
entry3, _ := newLogEntry(log, 3, 2, &testCommand1{Val: "bar", I: 0})
if err := log.appendEntry(entry3); err != nil {
t.Fatalf("Unable to append: %v", err)
}
if err := log.setCommitIndex(2); err != nil {
t.Fatalf("Unable to partially commit: %v", err)
}
// Truncate committed entry.
if err := log.truncate(1, 1); err == nil || err.Error() != "raft.Log: Index is already committed (2): (IDX=1, TERM=1)" {
t.Fatalf("Truncating committed entries shouldn't work: %v", err)
}
// Truncate past end of log.
if err := log.truncate(4, 2); err == nil || err.Error() != "raft.Log: Entry index does not exist (MAX=3): (IDX=4, TERM=2)" {
t.Fatalf("Truncating past end-of-log shouldn't work: %v", err)
}
// Truncate entry with mismatched term.
if err := log.truncate(2, 2); err == nil || err.Error() != "raft.Log: Entry at index does not have matching term (1): (IDX=2, TERM=2)" {
t.Fatalf("Truncating mismatched entries shouldn't work: %v", err)
}
// Truncate end of log.
if err := log.truncate(3, 2); !(err == nil && reflect.DeepEqual(log.entries, []*LogEntry{entry1, entry2, entry3})) {
t.Fatalf("Truncating end of log should work: %v\n\nEntries:\nActual: %v\nExpected: %v", err, log.entries, []*LogEntry{entry1, entry2, entry3})
}
// Truncate at last commit.
if err := log.truncate(2, 1); !(err == nil && reflect.DeepEqual(log.entries, []*LogEntry{entry1, entry2})) {
t.Fatalf("Truncating at last commit should work: %v\n\nEntries:\nActual: %v\nExpected: %v", err, log.entries, []*LogEntry{entry1, entry2})
}
// Append after truncate
if err := log.appendEntry(entry3); err != nil {
t.Fatalf("Unable to append after truncate: %v", err)
}
log.close()
// Recovery the truncated log
log = newLog()
if err := log.open(path); err != nil {
t.Fatalf("Unable to open log: %v", err)
}
// Validate existing log entries.
if len(log.entries) != 3 {
t.Fatalf("Expected 3 entries, got %d", len(log.entries))
}
if log.entries[0].Index != 1 || log.entries[0].Term != 1 {
t.Fatalf("Unexpected entry[0]: %v", log.entries[0])
}
if log.entries[1].Index != 2 || log.entries[1].Term != 1 {
t.Fatalf("Unexpected entry[1]: %v", log.entries[1])
}
if log.entries[2].Index != 3 || log.entries[2].Term != 2 {
t.Fatalf("Unexpected entry[2]: %v", log.entries[2])
}
}

View File

@ -1,26 +0,0 @@
package raft
import (
"io"
)
// NOP command
type NOPCommand struct {
}
// The name of the NOP command in the log
func (c NOPCommand) CommandName() string {
return "raft:nop"
}
func (c NOPCommand) Apply(server *Server) (interface{}, error) {
return nil, nil
}
func (c NOPCommand) Encode(w io.Writer) error {
return nil
}
func (c NOPCommand) Decode(r io.Reader) error {
return nil
}

View File

@ -1,271 +0,0 @@
package raft
import (
"sync"
"time"
)
//------------------------------------------------------------------------------
//
// Typedefs
//
//------------------------------------------------------------------------------
// A peer is a reference to another server involved in the consensus protocol.
type Peer struct {
server *Server
name string
prevLogIndex uint64
mutex sync.RWMutex
stopChan chan bool
heartbeatTimeout time.Duration
}
//------------------------------------------------------------------------------
//
// Constructor
//
//------------------------------------------------------------------------------
// Creates a new peer.
func newPeer(server *Server, name string, heartbeatTimeout time.Duration) *Peer {
return &Peer{
server: server,
name: name,
heartbeatTimeout: heartbeatTimeout,
}
}
//------------------------------------------------------------------------------
//
// Accessors
//
//------------------------------------------------------------------------------
// Retrieves the name of the peer.
func (p *Peer) Name() string {
return p.name
}
// Sets the heartbeat timeout.
func (p *Peer) setHeartbeatTimeout(duration time.Duration) {
p.heartbeatTimeout = duration
}
//--------------------------------------
// Prev log index
//--------------------------------------
// Retrieves the previous log index.
func (p *Peer) getPrevLogIndex() uint64 {
p.mutex.RLock()
defer p.mutex.RUnlock()
return p.prevLogIndex
}
// Sets the previous log index.
func (p *Peer) setPrevLogIndex(value uint64) {
p.mutex.Lock()
defer p.mutex.Unlock()
p.prevLogIndex = value
}
//------------------------------------------------------------------------------
//
// Methods
//
//------------------------------------------------------------------------------
//--------------------------------------
// Heartbeat
//--------------------------------------
// Starts the peer heartbeat.
func (p *Peer) startHeartbeat() {
p.stopChan = make(chan bool, 1)
c := make(chan bool)
go p.heartbeat(c)
<-c
}
// Stops the peer heartbeat.
func (p *Peer) stopHeartbeat() {
// here is a problem
// the previous stop is no buffer leader may get blocked
// when heartbeat returns at line 132
// I make the channel with 1 buffer
// and try to panic here
select {
case p.stopChan <- true:
default:
panic("[" + p.server.Name() + "] cannot stop [" + p.Name() + "] heartbeat")
}
}
//--------------------------------------
// Copying
//--------------------------------------
// Clones the state of the peer. The clone is not attached to a server and
// the heartbeat timer will not exist.
func (p *Peer) clone() *Peer {
p.mutex.Lock()
defer p.mutex.Unlock()
return &Peer{
name: p.name,
prevLogIndex: p.prevLogIndex,
}
}
//--------------------------------------
// Heartbeat
//--------------------------------------
// Listens to the heartbeat timeout and flushes an AppendEntries RPC.
func (p *Peer) heartbeat(c chan bool) {
stopChan := p.stopChan
c <- true
debugln("peer.heartbeat: ", p.Name(), p.heartbeatTimeout)
for {
select {
case <-stopChan:
debugln("peer.heartbeat.stop: ", p.Name())
return
case <-time.After(p.heartbeatTimeout):
debugln("peer.heartbeat.run: ", p.Name())
prevLogIndex := p.getPrevLogIndex()
entries, prevLogTerm := p.server.log.getEntriesAfter(prevLogIndex, p.server.maxLogEntriesPerRequest)
if p.server.State() != Leader {
return
}
if entries != nil {
p.sendAppendEntriesRequest(newAppendEntriesRequest(p.server.currentTerm, prevLogIndex, prevLogTerm, p.server.log.CommitIndex(), p.server.name, entries))
} else {
p.sendSnapshotRequest(newSnapshotRequest(p.server.name, p.server.lastSnapshot))
}
}
}
}
//--------------------------------------
// Append Entries
//--------------------------------------
// Sends an AppendEntries request to the peer through the transport.
func (p *Peer) sendAppendEntriesRequest(req *AppendEntriesRequest) {
traceln("peer.flush.send: ", p.server.Name(), "->", p.Name(), " ", len(req.Entries))
resp := p.server.Transporter().SendAppendEntriesRequest(p.server, p, req)
if resp == nil {
debugln("peer.flush.timeout: ", p.server.Name(), "->", p.Name())
return
}
traceln("peer.flush.recv: ", p.Name())
// If successful then update the previous log index.
p.mutex.Lock()
if resp.Success {
if len(req.Entries) > 0 {
p.prevLogIndex = req.Entries[len(req.Entries)-1].Index
// if peer append a log entry from the current term
// we set append to true
if req.Entries[len(req.Entries)-1].Term == p.server.currentTerm {
resp.append = true
}
}
traceln("peer.flush.success: ", p.server.Name(), "->", p.Name(), "; idx =", p.prevLogIndex)
// If it was unsuccessful then decrement the previous log index and
// we'll try again next time.
} else {
if resp.CommitIndex >= p.prevLogIndex {
// we may miss a response from peer
// so maybe the peer has commited the logs we sent
// but we did not receive the success reply and did not increase
// the prevLogIndex
p.prevLogIndex = resp.CommitIndex
debugln("peer.flush.commitIndex: ", p.server.Name(), "->", p.Name(), " idx =", p.prevLogIndex)
} else if p.prevLogIndex > 0 {
// Decrement the previous log index down until we find a match. Don't
// let it go below where the peer's commit index is though. That's a
// problem.
p.prevLogIndex--
// if it not enough, we directly decrease to the index of the
if p.prevLogIndex > resp.Index {
p.prevLogIndex = resp.Index
}
debugln("peer.flush.decrement: ", p.server.Name(), "->", p.Name(), " idx =", p.prevLogIndex)
}
}
p.mutex.Unlock()
// Attach the peer to resp, thus server can know where it comes from
resp.peer = p.Name()
// Send response to server for processing.
p.server.send(resp)
}
// Sends an Snapshot request to the peer through the transport.
func (p *Peer) sendSnapshotRequest(req *SnapshotRequest) {
debugln("peer.snap.send: ", p.name)
resp := p.server.Transporter().SendSnapshotRequest(p.server, p, req)
if resp == nil {
debugln("peer.snap.timeout: ", p.name)
return
}
debugln("peer.snap.recv: ", p.name)
// If successful, the peer should have been to snapshot state
// Send it the snapshot!
if resp.Success {
p.sendSnapshotRecoveryRequest()
} else {
debugln("peer.snap.failed: ", p.name)
return
}
}
// Sends an Snapshot Recovery request to the peer through the transport.
func (p *Peer) sendSnapshotRecoveryRequest() {
req := newSnapshotRecoveryRequest(p.server.name, p.server.lastSnapshot)
debugln("peer.snap.recovery.send: ", p.name)
resp := p.server.Transporter().SendSnapshotRecoveryRequest(p.server, p, req)
if resp.Success {
p.prevLogIndex = req.LastIndex
} else {
debugln("peer.snap.recovery.failed: ", p.name)
return
}
// Send response to server for processing.
p.server.send(&AppendEntriesResponse{Term: resp.Term, Success: resp.Success, append: (resp.Term == p.server.currentTerm)})
}
//--------------------------------------
// Vote Requests
//--------------------------------------
// send VoteRequest Request
func (p *Peer) sendVoteRequest(req *RequestVoteRequest, c chan *RequestVoteResponse) {
debugln("peer.vote: ", p.server.Name(), "->", p.Name())
req.peer = p
if resp := p.server.Transporter().SendVoteRequest(p.server, p, req); resp != nil {
debugln("peer.vote: recv", p.server.Name(), "<-", p.Name())
resp.peer = p
c <- resp
}
}

View File

@ -1,115 +0,0 @@
// Code generated by protoc-gen-go.
// source: append_entries_request.proto
// DO NOT EDIT!
package protobuf
import proto "code.google.com/p/goprotobuf/proto"
import json "encoding/json"
import math "math"
// Reference proto, json, and math imports to suppress error if they are not otherwise used.
var _ = proto.Marshal
var _ = &json.SyntaxError{}
var _ = math.Inf
type ProtoAppendEntriesRequest struct {
Term *uint64 `protobuf:"varint,1,req" json:"Term,omitempty"`
PrevLogIndex *uint64 `protobuf:"varint,2,req" json:"PrevLogIndex,omitempty"`
PrevLogTerm *uint64 `protobuf:"varint,3,req" json:"PrevLogTerm,omitempty"`
CommitIndex *uint64 `protobuf:"varint,4,req" json:"CommitIndex,omitempty"`
LeaderName *string `protobuf:"bytes,5,req" json:"LeaderName,omitempty"`
Entries []*ProtoAppendEntriesRequest_ProtoLogEntry `protobuf:"bytes,6,rep" json:"Entries,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoAppendEntriesRequest) Reset() { *m = ProtoAppendEntriesRequest{} }
func (m *ProtoAppendEntriesRequest) String() string { return proto.CompactTextString(m) }
func (*ProtoAppendEntriesRequest) ProtoMessage() {}
func (m *ProtoAppendEntriesRequest) GetTerm() uint64 {
if m != nil && m.Term != nil {
return *m.Term
}
return 0
}
func (m *ProtoAppendEntriesRequest) GetPrevLogIndex() uint64 {
if m != nil && m.PrevLogIndex != nil {
return *m.PrevLogIndex
}
return 0
}
func (m *ProtoAppendEntriesRequest) GetPrevLogTerm() uint64 {
if m != nil && m.PrevLogTerm != nil {
return *m.PrevLogTerm
}
return 0
}
func (m *ProtoAppendEntriesRequest) GetCommitIndex() uint64 {
if m != nil && m.CommitIndex != nil {
return *m.CommitIndex
}
return 0
}
func (m *ProtoAppendEntriesRequest) GetLeaderName() string {
if m != nil && m.LeaderName != nil {
return *m.LeaderName
}
return ""
}
func (m *ProtoAppendEntriesRequest) GetEntries() []*ProtoAppendEntriesRequest_ProtoLogEntry {
if m != nil {
return m.Entries
}
return nil
}
type ProtoAppendEntriesRequest_ProtoLogEntry struct {
Index *uint64 `protobuf:"varint,1,req" json:"Index,omitempty"`
Term *uint64 `protobuf:"varint,2,req" json:"Term,omitempty"`
CommandName *string `protobuf:"bytes,3,req" json:"CommandName,omitempty"`
Command []byte `protobuf:"bytes,4,opt" json:"Command,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoAppendEntriesRequest_ProtoLogEntry) Reset() {
*m = ProtoAppendEntriesRequest_ProtoLogEntry{}
}
func (m *ProtoAppendEntriesRequest_ProtoLogEntry) String() string { return proto.CompactTextString(m) }
func (*ProtoAppendEntriesRequest_ProtoLogEntry) ProtoMessage() {}
func (m *ProtoAppendEntriesRequest_ProtoLogEntry) GetIndex() uint64 {
if m != nil && m.Index != nil {
return *m.Index
}
return 0
}
func (m *ProtoAppendEntriesRequest_ProtoLogEntry) GetTerm() uint64 {
if m != nil && m.Term != nil {
return *m.Term
}
return 0
}
func (m *ProtoAppendEntriesRequest_ProtoLogEntry) GetCommandName() string {
if m != nil && m.CommandName != nil {
return *m.CommandName
}
return ""
}
func (m *ProtoAppendEntriesRequest_ProtoLogEntry) GetCommand() []byte {
if m != nil {
return m.Command
}
return nil
}
func init() {
}

View File

@ -1,18 +0,0 @@
package protobuf;
message ProtoAppendEntriesRequest {
required uint64 Term=1;
required uint64 PrevLogIndex=2;
required uint64 PrevLogTerm=3;
required uint64 CommitIndex=4;
required string LeaderName=5;
message ProtoLogEntry {
required uint64 Index=1;
required uint64 Term=2;
required string CommandName=3;
optional bytes Command=4;
}
repeated ProtoLogEntry Entries=6;
}

View File

@ -1,57 +0,0 @@
// Code generated by protoc-gen-go.
// source: append_entries_responses.proto
// DO NOT EDIT!
package protobuf
import proto "code.google.com/p/goprotobuf/proto"
import json "encoding/json"
import math "math"
// Reference proto, json, and math imports to suppress error if they are not otherwise used.
var _ = proto.Marshal
var _ = &json.SyntaxError{}
var _ = math.Inf
type ProtoAppendEntriesResponse struct {
Term *uint64 `protobuf:"varint,1,req" json:"Term,omitempty"`
Index *uint64 `protobuf:"varint,2,req" json:"Index,omitempty"`
CommitIndex *uint64 `protobuf:"varint,3,req" json:"CommitIndex,omitempty"`
Success *bool `protobuf:"varint,4,req" json:"Success,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoAppendEntriesResponse) Reset() { *m = ProtoAppendEntriesResponse{} }
func (m *ProtoAppendEntriesResponse) String() string { return proto.CompactTextString(m) }
func (*ProtoAppendEntriesResponse) ProtoMessage() {}
func (m *ProtoAppendEntriesResponse) GetTerm() uint64 {
if m != nil && m.Term != nil {
return *m.Term
}
return 0
}
func (m *ProtoAppendEntriesResponse) GetIndex() uint64 {
if m != nil && m.Index != nil {
return *m.Index
}
return 0
}
func (m *ProtoAppendEntriesResponse) GetCommitIndex() uint64 {
if m != nil && m.CommitIndex != nil {
return *m.CommitIndex
}
return 0
}
func (m *ProtoAppendEntriesResponse) GetSuccess() bool {
if m != nil && m.Success != nil {
return *m.Success
}
return false
}
func init() {
}

View File

@ -1,8 +0,0 @@
package protobuf;
message ProtoAppendEntriesResponse {
required uint64 Term=1;
required uint64 Index=2;
required uint64 CommitIndex=3;
required bool Success=4;
}

View File

@ -1,57 +0,0 @@
// Code generated by protoc-gen-go.
// source: log_entry.proto
// DO NOT EDIT!
package protobuf
import proto "code.google.com/p/goprotobuf/proto"
import json "encoding/json"
import math "math"
// Reference proto, json, and math imports to suppress error if they are not otherwise used.
var _ = proto.Marshal
var _ = &json.SyntaxError{}
var _ = math.Inf
type ProtoLogEntry struct {
Index *uint64 `protobuf:"varint,1,req" json:"Index,omitempty"`
Term *uint64 `protobuf:"varint,2,req" json:"Term,omitempty"`
CommandName *string `protobuf:"bytes,3,req" json:"CommandName,omitempty"`
Command []byte `protobuf:"bytes,4,opt" json:"Command,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoLogEntry) Reset() { *m = ProtoLogEntry{} }
func (m *ProtoLogEntry) String() string { return proto.CompactTextString(m) }
func (*ProtoLogEntry) ProtoMessage() {}
func (m *ProtoLogEntry) GetIndex() uint64 {
if m != nil && m.Index != nil {
return *m.Index
}
return 0
}
func (m *ProtoLogEntry) GetTerm() uint64 {
if m != nil && m.Term != nil {
return *m.Term
}
return 0
}
func (m *ProtoLogEntry) GetCommandName() string {
if m != nil && m.CommandName != nil {
return *m.CommandName
}
return ""
}
func (m *ProtoLogEntry) GetCommand() []byte {
if m != nil {
return m.Command
}
return nil
}
func init() {
}

View File

@ -1,8 +0,0 @@
package protobuf;
message ProtoLogEntry {
required uint64 Index=1;
required uint64 Term=2;
required string CommandName=3;
optional bytes Command=4; // for nop-command
}

View File

@ -1,57 +0,0 @@
// Code generated by protoc-gen-go.
// source: request_vote_request.proto
// DO NOT EDIT!
package protobuf
import proto "code.google.com/p/goprotobuf/proto"
import json "encoding/json"
import math "math"
// Reference proto, json, and math imports to suppress error if they are not otherwise used.
var _ = proto.Marshal
var _ = &json.SyntaxError{}
var _ = math.Inf
type ProtoRequestVoteRequest struct {
Term *uint64 `protobuf:"varint,1,req" json:"Term,omitempty"`
LastLogIndex *uint64 `protobuf:"varint,2,req" json:"LastLogIndex,omitempty"`
LastLogTerm *uint64 `protobuf:"varint,3,req" json:"LastLogTerm,omitempty"`
CandidateName *string `protobuf:"bytes,4,req" json:"CandidateName,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoRequestVoteRequest) Reset() { *m = ProtoRequestVoteRequest{} }
func (m *ProtoRequestVoteRequest) String() string { return proto.CompactTextString(m) }
func (*ProtoRequestVoteRequest) ProtoMessage() {}
func (m *ProtoRequestVoteRequest) GetTerm() uint64 {
if m != nil && m.Term != nil {
return *m.Term
}
return 0
}
func (m *ProtoRequestVoteRequest) GetLastLogIndex() uint64 {
if m != nil && m.LastLogIndex != nil {
return *m.LastLogIndex
}
return 0
}
func (m *ProtoRequestVoteRequest) GetLastLogTerm() uint64 {
if m != nil && m.LastLogTerm != nil {
return *m.LastLogTerm
}
return 0
}
func (m *ProtoRequestVoteRequest) GetCandidateName() string {
if m != nil && m.CandidateName != nil {
return *m.CandidateName
}
return ""
}
func init() {
}

View File

@ -1,8 +0,0 @@
package protobuf;
message ProtoRequestVoteRequest {
required uint64 Term=1;
required uint64 LastLogIndex=2;
required uint64 LastLogTerm=3;
required string CandidateName=4;
}

View File

@ -1,41 +0,0 @@
// Code generated by protoc-gen-go.
// source: request_vote_responses.proto
// DO NOT EDIT!
package protobuf
import proto "code.google.com/p/goprotobuf/proto"
import json "encoding/json"
import math "math"
// Reference proto, json, and math imports to suppress error if they are not otherwise used.
var _ = proto.Marshal
var _ = &json.SyntaxError{}
var _ = math.Inf
type ProtoRequestVoteResponse struct {
Term *uint64 `protobuf:"varint,1,req" json:"Term,omitempty"`
VoteGranted *bool `protobuf:"varint,2,req" json:"VoteGranted,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoRequestVoteResponse) Reset() { *m = ProtoRequestVoteResponse{} }
func (m *ProtoRequestVoteResponse) String() string { return proto.CompactTextString(m) }
func (*ProtoRequestVoteResponse) ProtoMessage() {}
func (m *ProtoRequestVoteResponse) GetTerm() uint64 {
if m != nil && m.Term != nil {
return *m.Term
}
return 0
}
func (m *ProtoRequestVoteResponse) GetVoteGranted() bool {
if m != nil && m.VoteGranted != nil {
return *m.VoteGranted
}
return false
}
func init() {
}

View File

@ -1,6 +0,0 @@
package protobuf;
message ProtoRequestVoteResponse {
required uint64 Term=1;
required bool VoteGranted=2;
}

View File

@ -1,65 +0,0 @@
// Code generated by protoc-gen-go.
// source: snapshot_recovery_request.proto
// DO NOT EDIT!
package protobuf
import proto "code.google.com/p/goprotobuf/proto"
import json "encoding/json"
import math "math"
// Reference proto, json, and math imports to suppress error if they are not otherwise used.
var _ = proto.Marshal
var _ = &json.SyntaxError{}
var _ = math.Inf
type ProtoSnapshotRecoveryRequest struct {
LeaderName *string `protobuf:"bytes,1,req" json:"LeaderName,omitempty"`
LastIndex *uint64 `protobuf:"varint,2,req" json:"LastIndex,omitempty"`
LastTerm *uint64 `protobuf:"varint,3,req" json:"LastTerm,omitempty"`
Peers []string `protobuf:"bytes,4,rep" json:"Peers,omitempty"`
State []byte `protobuf:"bytes,5,req" json:"State,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoSnapshotRecoveryRequest) Reset() { *m = ProtoSnapshotRecoveryRequest{} }
func (m *ProtoSnapshotRecoveryRequest) String() string { return proto.CompactTextString(m) }
func (*ProtoSnapshotRecoveryRequest) ProtoMessage() {}
func (m *ProtoSnapshotRecoveryRequest) GetLeaderName() string {
if m != nil && m.LeaderName != nil {
return *m.LeaderName
}
return ""
}
func (m *ProtoSnapshotRecoveryRequest) GetLastIndex() uint64 {
if m != nil && m.LastIndex != nil {
return *m.LastIndex
}
return 0
}
func (m *ProtoSnapshotRecoveryRequest) GetLastTerm() uint64 {
if m != nil && m.LastTerm != nil {
return *m.LastTerm
}
return 0
}
func (m *ProtoSnapshotRecoveryRequest) GetPeers() []string {
if m != nil {
return m.Peers
}
return nil
}
func (m *ProtoSnapshotRecoveryRequest) GetState() []byte {
if m != nil {
return m.State
}
return nil
}
func init() {
}

View File

@ -1,9 +0,0 @@
package protobuf;
message ProtoSnapshotRecoveryRequest {
required string LeaderName=1;
required uint64 LastIndex=2;
required uint64 LastTerm=3;
repeated string Peers=4;
required bytes State=5;
}

View File

@ -1,49 +0,0 @@
// Code generated by protoc-gen-go.
// source: snapshot_recovery_response.proto
// DO NOT EDIT!
package protobuf
import proto "code.google.com/p/goprotobuf/proto"
import json "encoding/json"
import math "math"
// Reference proto, json, and math imports to suppress error if they are not otherwise used.
var _ = proto.Marshal
var _ = &json.SyntaxError{}
var _ = math.Inf
type ProtoSnapshotRecoveryResponse struct {
Term *uint64 `protobuf:"varint,1,req" json:"Term,omitempty"`
Success *bool `protobuf:"varint,2,req" json:"Success,omitempty"`
CommitIndex *uint64 `protobuf:"varint,3,req" json:"CommitIndex,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoSnapshotRecoveryResponse) Reset() { *m = ProtoSnapshotRecoveryResponse{} }
func (m *ProtoSnapshotRecoveryResponse) String() string { return proto.CompactTextString(m) }
func (*ProtoSnapshotRecoveryResponse) ProtoMessage() {}
func (m *ProtoSnapshotRecoveryResponse) GetTerm() uint64 {
if m != nil && m.Term != nil {
return *m.Term
}
return 0
}
func (m *ProtoSnapshotRecoveryResponse) GetSuccess() bool {
if m != nil && m.Success != nil {
return *m.Success
}
return false
}
func (m *ProtoSnapshotRecoveryResponse) GetCommitIndex() uint64 {
if m != nil && m.CommitIndex != nil {
return *m.CommitIndex
}
return 0
}
func init() {
}

View File

@ -1,7 +0,0 @@
package protobuf;
message ProtoSnapshotRecoveryResponse {
required uint64 Term=1;
required bool Success=2;
required uint64 CommitIndex=3;
}

View File

@ -1,49 +0,0 @@
// Code generated by protoc-gen-go.
// source: snapshot_request.proto
// DO NOT EDIT!
package protobuf
import proto "code.google.com/p/goprotobuf/proto"
import json "encoding/json"
import math "math"
// Reference proto, json, and math imports to suppress error if they are not otherwise used.
var _ = proto.Marshal
var _ = &json.SyntaxError{}
var _ = math.Inf
type ProtoSnapshotRequest struct {
LeaderName *string `protobuf:"bytes,1,req" json:"LeaderName,omitempty"`
LastIndex *uint64 `protobuf:"varint,2,req" json:"LastIndex,omitempty"`
LastTerm *uint64 `protobuf:"varint,3,req" json:"LastTerm,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoSnapshotRequest) Reset() { *m = ProtoSnapshotRequest{} }
func (m *ProtoSnapshotRequest) String() string { return proto.CompactTextString(m) }
func (*ProtoSnapshotRequest) ProtoMessage() {}
func (m *ProtoSnapshotRequest) GetLeaderName() string {
if m != nil && m.LeaderName != nil {
return *m.LeaderName
}
return ""
}
func (m *ProtoSnapshotRequest) GetLastIndex() uint64 {
if m != nil && m.LastIndex != nil {
return *m.LastIndex
}
return 0
}
func (m *ProtoSnapshotRequest) GetLastTerm() uint64 {
if m != nil && m.LastTerm != nil {
return *m.LastTerm
}
return 0
}
func init() {
}

View File

@ -1,7 +0,0 @@
package protobuf;
message ProtoSnapshotRequest {
required string LeaderName=1;
required uint64 LastIndex=2;
required uint64 LastTerm=3;
}

View File

@ -1,33 +0,0 @@
// Code generated by protoc-gen-go.
// source: snapshot_response.proto
// DO NOT EDIT!
package protobuf
import proto "code.google.com/p/goprotobuf/proto"
import json "encoding/json"
import math "math"
// Reference proto, json, and math imports to suppress error if they are not otherwise used.
var _ = proto.Marshal
var _ = &json.SyntaxError{}
var _ = math.Inf
type ProtoSnapshotResponse struct {
Success *bool `protobuf:"varint,1,req" json:"Success,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *ProtoSnapshotResponse) Reset() { *m = ProtoSnapshotResponse{} }
func (m *ProtoSnapshotResponse) String() string { return proto.CompactTextString(m) }
func (*ProtoSnapshotResponse) ProtoMessage() {}
func (m *ProtoSnapshotResponse) GetSuccess() bool {
if m != nil && m.Success != nil {
return *m.Success
}
return false
}
func init() {
}

View File

@ -1,5 +0,0 @@
package protobuf;
message ProtoSnapshotResponse {
required bool Success=1;
}

View File

@ -1,68 +0,0 @@
package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"io"
"io/ioutil"
)
// The request sent to a server to vote for a candidate to become a leader.
type RequestVoteRequest struct {
peer *Peer
Term uint64
LastLogIndex uint64
LastLogTerm uint64
CandidateName string
}
// Creates a new RequestVote request.
func newRequestVoteRequest(term uint64, candidateName string, lastLogIndex uint64, lastLogTerm uint64) *RequestVoteRequest {
return &RequestVoteRequest{
Term: term,
LastLogIndex: lastLogIndex,
LastLogTerm: lastLogTerm,
CandidateName: candidateName,
}
}
// Encodes the RequestVoteRequest to a buffer. Returns the number of bytes
// written and any error that may have occurred.
func (req *RequestVoteRequest) encode(w io.Writer) (int, error) {
pb := &protobuf.ProtoRequestVoteRequest{
Term: proto.Uint64(req.Term),
LastLogIndex: proto.Uint64(req.LastLogIndex),
LastLogTerm: proto.Uint64(req.LastLogTerm),
CandidateName: proto.String(req.CandidateName),
}
p, err := proto.Marshal(pb)
if err != nil {
return -1, err
}
return w.Write(p)
}
// Decodes the RequestVoteRequest from a buffer. Returns the number of bytes read and
// any error that occurs.
func (req *RequestVoteRequest) decode(r io.Reader) (int, error) {
data, err := ioutil.ReadAll(r)
if err != nil {
return -1, err
}
totalBytes := len(data)
pb := &protobuf.ProtoRequestVoteRequest{}
if err = proto.Unmarshal(data, pb); err != nil {
return -1, err
}
req.Term = pb.GetTerm()
req.LastLogIndex = pb.GetLastLogIndex()
req.LastLogTerm = pb.GetLastLogTerm()
req.CandidateName = pb.GetCandidateName()
return totalBytes, nil
}

View File

@ -1,61 +0,0 @@
package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"io"
"io/ioutil"
)
// The response returned from a server after a vote for a candidate to become a leader.
type RequestVoteResponse struct {
peer *Peer
Term uint64
VoteGranted bool
}
// Creates a new RequestVote response.
func newRequestVoteResponse(term uint64, voteGranted bool) *RequestVoteResponse {
return &RequestVoteResponse{
Term: term,
VoteGranted: voteGranted,
}
}
// Encodes the RequestVoteResponse to a buffer. Returns the number of bytes
// written and any error that may have occurred.
func (resp *RequestVoteResponse) encode(w io.Writer) (int, error) {
pb := &protobuf.ProtoRequestVoteResponse{
Term: proto.Uint64(resp.Term),
VoteGranted: proto.Bool(resp.VoteGranted),
}
p, err := proto.Marshal(pb)
if err != nil {
return -1, err
}
return w.Write(p)
}
// Decodes the RequestVoteResponse from a buffer. Returns the number of bytes read and
// any error that occurs.
func (resp *RequestVoteResponse) decode(r io.Reader) (int, error) {
data, err := ioutil.ReadAll(r)
if err != nil {
return 0, err
}
totalBytes := len(data)
pb := &protobuf.ProtoRequestVoteResponse{}
if err = proto.Unmarshal(data, pb); err != nil {
return -1, err
}
resp.Term = pb.GetTerm()
resp.VoteGranted = pb.GetVoteGranted()
return totalBytes, nil
}

File diff suppressed because it is too large Load Diff

View File

@ -1,504 +0,0 @@
package raft
import (
"fmt"
"reflect"
"strconv"
"sync"
"testing"
"time"
)
//------------------------------------------------------------------------------
//
// Tests
//
//------------------------------------------------------------------------------
//--------------------------------------
// Request Vote
//--------------------------------------
// Ensure that we can request a vote from a server that has not voted.
func TestServerRequestVote(t *testing.T) {
server := newTestServer("1", &testTransporter{})
server.Start()
if _, err := server.Do(&DefaultJoinCommand{Name: server.Name()}); err != nil {
t.Fatalf("Server %s unable to join: %v", server.Name(), err)
}
defer server.Stop()
resp := server.RequestVote(newRequestVoteRequest(1, "foo", 1, 0))
if resp.Term != 1 || !resp.VoteGranted {
t.Fatalf("Invalid request vote response: %v/%v", resp.Term, resp.VoteGranted)
}
}
// // Ensure that a vote request is denied if it comes from an old term.
func TestServerRequestVoteDeniedForStaleTerm(t *testing.T) {
server := newTestServer("1", &testTransporter{})
server.Start()
if _, err := server.Do(&DefaultJoinCommand{Name: server.Name()}); err != nil {
t.Fatalf("Server %s unable to join: %v", server.Name(), err)
}
server.currentTerm = 2
defer server.Stop()
resp := server.RequestVote(newRequestVoteRequest(1, "foo", 1, 0))
if resp.Term != 2 || resp.VoteGranted {
t.Fatalf("Invalid request vote response: %v/%v", resp.Term, resp.VoteGranted)
}
if server.currentTerm != 2 && server.State() != Follower {
t.Fatalf("Server did not update term and demote: %v / %v", server.currentTerm, server.State())
}
}
// Ensure that a vote request is denied if we've already voted for a different candidate.
func TestServerRequestVoteDeniedIfAlreadyVoted(t *testing.T) {
server := newTestServer("1", &testTransporter{})
server.Start()
if _, err := server.Do(&DefaultJoinCommand{Name: server.Name()}); err != nil {
t.Fatalf("Server %s unable to join: %v", server.Name(), err)
}
server.currentTerm = 2
defer server.Stop()
resp := server.RequestVote(newRequestVoteRequest(2, "foo", 1, 0))
if resp.Term != 2 || !resp.VoteGranted {
t.Fatalf("First vote should not have been denied")
}
resp = server.RequestVote(newRequestVoteRequest(2, "bar", 1, 0))
if resp.Term != 2 || resp.VoteGranted {
t.Fatalf("Second vote should have been denied")
}
}
// Ensure that a vote request is approved if vote occurs in a new term.
func TestServerRequestVoteApprovedIfAlreadyVotedInOlderTerm(t *testing.T) {
server := newTestServer("1", &testTransporter{})
server.Start()
if _, err := server.Do(&DefaultJoinCommand{Name: server.Name()}); err != nil {
t.Fatalf("Server %s unable to join: %v", server.Name(), err)
}
time.Sleep(time.Millisecond * 100)
server.currentTerm = 2
defer server.Stop()
resp := server.RequestVote(newRequestVoteRequest(2, "foo", 2, 1))
if resp.Term != 2 || !resp.VoteGranted || server.VotedFor() != "foo" {
t.Fatalf("First vote should not have been denied")
}
resp = server.RequestVote(newRequestVoteRequest(3, "bar", 2, 1))
if resp.Term != 3 || !resp.VoteGranted || server.VotedFor() != "bar" {
t.Fatalf("Second vote should have been approved")
}
}
// Ensure that a vote request is denied if the log is out of date.
func TestServerRequestVoteDenyIfCandidateLogIsBehind(t *testing.T) {
tmpLog := newLog()
e0, _ := newLogEntry(tmpLog, 1, 1, &testCommand1{Val: "foo", I: 20})
e1, _ := newLogEntry(tmpLog, 2, 1, &testCommand2{X: 100})
e2, _ := newLogEntry(tmpLog, 3, 2, &testCommand1{Val: "bar", I: 0})
server := newTestServerWithLog("1", &testTransporter{}, []*LogEntry{e0, e1, e2})
// start as a follower with term 2 and index 3
server.Start()
defer server.Stop()
// request vote from term 3 with last log entry 2, 2
resp := server.RequestVote(newRequestVoteRequest(3, "foo", 2, 2))
if resp.Term != 3 || resp.VoteGranted {
t.Fatalf("Stale index vote should have been denied [%v/%v]", resp.Term, resp.VoteGranted)
}
// request vote from term 2 with last log entry 2, 3
resp = server.RequestVote(newRequestVoteRequest(2, "foo", 3, 2))
if resp.Term != 3 || resp.VoteGranted {
t.Fatalf("Stale term vote should have been denied [%v/%v]", resp.Term, resp.VoteGranted)
}
// request vote from term 3 with last log entry 2, 3
resp = server.RequestVote(newRequestVoteRequest(3, "foo", 3, 2))
if resp.Term != 3 || !resp.VoteGranted {
t.Fatalf("Matching log vote should have been granted")
}
// request vote from term 3 with last log entry 2, 4
resp = server.RequestVote(newRequestVoteRequest(3, "foo", 4, 2))
if resp.Term != 3 || !resp.VoteGranted {
t.Fatalf("Ahead-of-log vote should have been granted")
}
}
// //--------------------------------------
// // Promotion
// //--------------------------------------
// // Ensure that we can self-promote a server to candidate, obtain votes and become a fearless leader.
func TestServerPromoteSelf(t *testing.T) {
e0, _ := newLogEntry(newLog(), 1, 1, &testCommand1{Val: "foo", I: 20})
server := newTestServerWithLog("1", &testTransporter{}, []*LogEntry{e0})
// start as a follower
server.Start()
defer server.Stop()
time.Sleep(2 * testElectionTimeout)
if server.State() != Leader {
t.Fatalf("Server self-promotion failed: %v", server.State())
}
}
//Ensure that we can promote a server within a cluster to a leader.
func TestServerPromote(t *testing.T) {
lookup := map[string]*Server{}
transporter := &testTransporter{}
transporter.sendVoteRequestFunc = func(server *Server, peer *Peer, req *RequestVoteRequest) *RequestVoteResponse {
return lookup[peer.Name()].RequestVote(req)
}
transporter.sendAppendEntriesRequestFunc = func(server *Server, peer *Peer, req *AppendEntriesRequest) *AppendEntriesResponse {
return lookup[peer.Name()].AppendEntries(req)
}
servers := newTestCluster([]string{"1", "2", "3"}, transporter, lookup)
servers[0].Start()
servers[1].Start()
servers[2].Start()
time.Sleep(2 * testElectionTimeout)
if servers[0].State() != Leader && servers[1].State() != Leader && servers[2].State() != Leader {
t.Fatalf("No leader elected: (%s, %s, %s)", servers[0].State(), servers[1].State(), servers[2].State())
}
for _, server := range servers {
server.Stop()
}
}
//--------------------------------------
// Append Entries
//--------------------------------------
// Ensure we can append entries to a server.
func TestServerAppendEntries(t *testing.T) {
server := newTestServer("1", &testTransporter{})
server.SetHeartbeatTimeout(time.Second * 10)
server.Start()
defer server.Stop()
// Append single entry.
e, _ := newLogEntry(nil, 1, 1, &testCommand1{Val: "foo", I: 10})
entries := []*LogEntry{e}
resp := server.AppendEntries(newAppendEntriesRequest(1, 0, 0, 0, "ldr", entries))
if resp.Term != 1 || !resp.Success {
t.Fatalf("AppendEntries failed: %v/%v", resp.Term, resp.Success)
}
if index, term := server.log.commitInfo(); index != 0 || term != 0 {
t.Fatalf("Invalid commit info [IDX=%v, TERM=%v]", index, term)
}
// Append multiple entries + commit the last one.
e1, _ := newLogEntry(nil, 2, 1, &testCommand1{Val: "bar", I: 20})
e2, _ := newLogEntry(nil, 3, 1, &testCommand1{Val: "baz", I: 30})
entries = []*LogEntry{e1, e2}
resp = server.AppendEntries(newAppendEntriesRequest(1, 1, 1, 1, "ldr", entries))
if resp.Term != 1 || !resp.Success {
t.Fatalf("AppendEntries failed: %v/%v", resp.Term, resp.Success)
}
if index, term := server.log.commitInfo(); index != 1 || term != 1 {
t.Fatalf("Invalid commit info [IDX=%v, TERM=%v]", index, term)
}
// Send zero entries and commit everything.
resp = server.AppendEntries(newAppendEntriesRequest(2, 3, 1, 3, "ldr", []*LogEntry{}))
if resp.Term != 2 || !resp.Success {
t.Fatalf("AppendEntries failed: %v/%v", resp.Term, resp.Success)
}
if index, term := server.log.commitInfo(); index != 3 || term != 1 {
t.Fatalf("Invalid commit info [IDX=%v, TERM=%v]", index, term)
}
}
//Ensure that entries with stale terms are rejected.
func TestServerAppendEntriesWithStaleTermsAreRejected(t *testing.T) {
server := newTestServer("1", &testTransporter{})
server.Start()
defer server.Stop()
server.currentTerm = 2
// Append single entry.
e, _ := newLogEntry(nil, 1, 1, &testCommand1{Val: "foo", I: 10})
entries := []*LogEntry{e}
resp := server.AppendEntries(newAppendEntriesRequest(1, 0, 0, 0, "ldr", entries))
if resp.Term != 2 || resp.Success {
t.Fatalf("AppendEntries should have failed: %v/%v", resp.Term, resp.Success)
}
if index, term := server.log.commitInfo(); index != 0 || term != 0 {
t.Fatalf("Invalid commit info [IDX=%v, TERM=%v]", index, term)
}
}
// Ensure that we reject entries if the commit log is different.
func TestServerAppendEntriesRejectedIfAlreadyCommitted(t *testing.T) {
server := newTestServer("1", &testTransporter{})
server.Start()
defer server.Stop()
// Append single entry + commit.
e1, _ := newLogEntry(nil, 1, 1, &testCommand1{Val: "foo", I: 10})
e2, _ := newLogEntry(nil, 2, 1, &testCommand1{Val: "foo", I: 15})
entries := []*LogEntry{e1, e2}
resp := server.AppendEntries(newAppendEntriesRequest(1, 0, 0, 2, "ldr", entries))
if resp.Term != 1 || !resp.Success {
t.Fatalf("AppendEntries failed: %v/%v", resp.Term, resp.Success)
}
// Append entry again (post-commit).
e, _ := newLogEntry(nil, 2, 1, &testCommand1{Val: "bar", I: 20})
entries = []*LogEntry{e}
resp = server.AppendEntries(newAppendEntriesRequest(1, 2, 1, 1, "ldr", entries))
if resp.Term != 1 || resp.Success {
t.Fatalf("AppendEntries should have failed: %v/%v", resp.Term, resp.Success)
}
}
// Ensure that we uncommitted entries are rolled back if new entries overwrite them.
func TestServerAppendEntriesOverwritesUncommittedEntries(t *testing.T) {
server := newTestServer("1", &testTransporter{})
server.Start()
defer server.Stop()
entry1, _ := newLogEntry(nil, 1, 1, &testCommand1{Val: "foo", I: 10})
entry2, _ := newLogEntry(nil, 2, 1, &testCommand1{Val: "foo", I: 15})
entry3, _ := newLogEntry(nil, 2, 2, &testCommand1{Val: "bar", I: 20})
// Append single entry + commit.
entries := []*LogEntry{entry1, entry2}
resp := server.AppendEntries(newAppendEntriesRequest(1, 0, 0, 1, "ldr", entries))
if resp.Term != 1 || !resp.Success || server.log.commitIndex != 1 || !reflect.DeepEqual(server.log.entries, []*LogEntry{entry1, entry2}) {
t.Fatalf("AppendEntries failed: %v/%v", resp.Term, resp.Success)
}
// Append entry that overwrites the second (uncommitted) entry.
entries = []*LogEntry{entry3}
resp = server.AppendEntries(newAppendEntriesRequest(2, 1, 1, 2, "ldr", entries))
if resp.Term != 2 || !resp.Success || server.log.commitIndex != 2 || !reflect.DeepEqual(server.log.entries, []*LogEntry{entry1, entry3}) {
t.Fatalf("AppendEntries should have succeeded: %v/%v", resp.Term, resp.Success)
}
}
//--------------------------------------
// Command Execution
//--------------------------------------
// Ensure that a follower cannot execute a command.
func TestServerDenyCommandExecutionWhenFollower(t *testing.T) {
server := newTestServer("1", &testTransporter{})
server.Start()
defer server.Stop()
var err error
if _, err = server.Do(&testCommand1{Val: "foo", I: 10}); err != NotLeaderError {
t.Fatalf("Expected error: %v, got: %v", NotLeaderError, err)
}
}
//--------------------------------------
// Membership
//--------------------------------------
// Ensure that we can start a single server and append to its log.
func TestServerSingleNode(t *testing.T) {
server := newTestServer("1", &testTransporter{})
if server.State() != Stopped {
t.Fatalf("Unexpected server state: %v", server.State())
}
server.Start()
time.Sleep(testHeartbeatTimeout)
// Join the server to itself.
if _, err := server.Do(&DefaultJoinCommand{Name: "1"}); err != nil {
t.Fatalf("Unable to join: %v", err)
}
debugln("finish command")
if server.State() != Leader {
t.Fatalf("Unexpected server state: %v", server.State())
}
server.Stop()
if server.State() != Stopped {
t.Fatalf("Unexpected server state: %v", server.State())
}
}
// Ensure that we can start multiple servers and determine a leader.
func TestServerMultiNode(t *testing.T) {
// Initialize the servers.
var mutex sync.RWMutex
servers := map[string]*Server{}
transporter := &testTransporter{}
transporter.sendVoteRequestFunc = func(server *Server, peer *Peer, req *RequestVoteRequest) *RequestVoteResponse {
mutex.RLock()
s := servers[peer.name]
mutex.RUnlock()
return s.RequestVote(req)
}
transporter.sendAppendEntriesRequestFunc = func(server *Server, peer *Peer, req *AppendEntriesRequest) *AppendEntriesResponse {
mutex.RLock()
s := servers[peer.name]
mutex.RUnlock()
return s.AppendEntries(req)
}
disTransporter := &testTransporter{}
disTransporter.sendVoteRequestFunc = func(server *Server, peer *Peer, req *RequestVoteRequest) *RequestVoteResponse {
return nil
}
disTransporter.sendAppendEntriesRequestFunc = func(server *Server, peer *Peer, req *AppendEntriesRequest) *AppendEntriesResponse {
return nil
}
var names []string
n := 5
// add n servers
for i := 1; i <= n; i++ {
names = append(names, strconv.Itoa(i))
}
var leader *Server
for _, name := range names {
server := newTestServer(name, transporter)
defer server.Stop()
mutex.Lock()
servers[name] = server
mutex.Unlock()
if name == "1" {
leader = server
server.SetHeartbeatTimeout(testHeartbeatTimeout)
server.Start()
time.Sleep(testHeartbeatTimeout)
} else {
server.SetElectionTimeout(testElectionTimeout)
server.SetHeartbeatTimeout(testHeartbeatTimeout)
server.Start()
time.Sleep(testHeartbeatTimeout)
}
if _, err := leader.Do(&DefaultJoinCommand{Name: name}); err != nil {
t.Fatalf("Unable to join server[%s]: %v", name, err)
}
}
time.Sleep(2 * testElectionTimeout)
// Check that two peers exist on leader.
mutex.RLock()
if leader.MemberCount() != n {
t.Fatalf("Expected member count to be %v, got %v", n, leader.MemberCount())
}
if servers["2"].State() == Leader || servers["3"].State() == Leader {
t.Fatalf("Expected leader should be 1: 2=%v, 3=%v\n", servers["2"].state, servers["3"].state)
}
mutex.RUnlock()
for i := 0; i < 20; i++ {
retry := 0
fmt.Println("Round ", i)
num := strconv.Itoa(i%(len(servers)) + 1)
num_1 := strconv.Itoa((i+3)%(len(servers)) + 1)
toStop := servers[num]
toStop_1 := servers[num_1]
// Stop the first server and wait for a re-election.
time.Sleep(2 * testElectionTimeout)
debugln("Disconnect ", toStop.Name())
debugln("disconnect ", num, " ", num_1)
toStop.SetTransporter(disTransporter)
toStop_1.SetTransporter(disTransporter)
time.Sleep(2 * testElectionTimeout)
// Check that either server 2 or 3 is the leader now.
//mutex.Lock()
leader := 0
for key, value := range servers {
debugln("Play begin")
if key != num && key != num_1 {
if value.State() == Leader {
debugln("Found leader")
for i := 0; i < 10; i++ {
debugln("[Test] do ", value.Name())
if _, err := value.Do(&testCommand2{X: 1}); err != nil {
break
}
debugln("[Test] Done")
}
debugln("Leader is ", value.Name(), " Index ", value.log.commitIndex)
}
debugln("Not Found leader")
}
}
for {
for key, value := range servers {
if key != num && key != num_1 {
if value.State() == Leader {
leader++
}
debugln(value.Name(), " ", value.currentTerm, " ", value.state)
}
}
if leader > 1 {
if retry < 300 {
debugln("retry")
retry++
leader = 0
time.Sleep(2 * testElectionTimeout)
continue
}
t.Fatalf("wrong leader number %v", leader)
}
if leader == 0 {
if retry < 300 {
retry++
fmt.Println("retry 0")
leader = 0
time.Sleep(2 * testElectionTimeout)
continue
}
t.Fatalf("wrong leader number %v", leader)
}
if leader == 1 {
break
}
}
//mutex.Unlock()
toStop.SetTransporter(transporter)
toStop_1.SetTransporter(transporter)
}
}

View File

@ -1,65 +0,0 @@
package raft
import (
//"bytes"
"encoding/json"
"fmt"
"hash/crc32"
"os"
"syscall"
)
//------------------------------------------------------------------------------
//
// Typedefs
//
//------------------------------------------------------------------------------
// the in memory SnapShot struct
// TODO add cluster configuration
type Snapshot struct {
LastIndex uint64 `json:"lastIndex"`
LastTerm uint64 `json:"lastTerm"`
// cluster configuration.
Peers []string `json: "peers"`
State []byte `json: "state"`
Path string `json: "path"`
}
// Save the snapshot to a file
func (ss *Snapshot) save() error {
// Write machine state to temporary buffer.
// open file
file, err := os.OpenFile(ss.Path, os.O_CREATE|os.O_WRONLY, 0600)
if err != nil {
return err
}
defer file.Close()
b, err := json.Marshal(ss)
// Generate checksum.
checksum := crc32.ChecksumIEEE(b)
// Write snapshot with checksum.
if _, err = fmt.Fprintf(file, "%08x\n", checksum); err != nil {
return err
}
if _, err = file.Write(b); err != nil {
return err
}
// force the change writting to disk
syscall.Fsync(int(file.Fd()))
return err
}
// remove the file of the snapshot
func (ss *Snapshot) remove() error {
err := os.Remove(ss.Path)
return err
}

View File

@ -1,77 +0,0 @@
package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"io"
"io/ioutil"
)
// The request sent to a server to start from the snapshot.
type SnapshotRecoveryRequest struct {
LeaderName string
LastIndex uint64
LastTerm uint64
Peers []string
State []byte
}
//------------------------------------------------------------------------------
//
// Constructors
//
//------------------------------------------------------------------------------
// Creates a new Snapshot request.
func newSnapshotRecoveryRequest(leaderName string, snapshot *Snapshot) *SnapshotRecoveryRequest {
return &SnapshotRecoveryRequest{
LeaderName: leaderName,
LastIndex: snapshot.LastIndex,
LastTerm: snapshot.LastTerm,
Peers: snapshot.Peers,
State: snapshot.State,
}
}
// Encodes the SnapshotRecoveryRequest to a buffer. Returns the number of bytes
// written and any error that may have occurred.
func (req *SnapshotRecoveryRequest) encode(w io.Writer) (int, error) {
pb := &protobuf.ProtoSnapshotRecoveryRequest{
LeaderName: proto.String(req.LeaderName),
LastIndex: proto.Uint64(req.LastIndex),
LastTerm: proto.Uint64(req.LastTerm),
Peers: req.Peers,
State: req.State,
}
p, err := proto.Marshal(pb)
if err != nil {
return -1, err
}
return w.Write(p)
}
// Decodes the SnapshotRecoveryRequest from a buffer. Returns the number of bytes read and
// any error that occurs.
func (req *SnapshotRecoveryRequest) decode(r io.Reader) (int, error) {
data, err := ioutil.ReadAll(r)
if err != nil {
return 0, err
}
totalBytes := len(data)
pb := &protobuf.ProtoSnapshotRequest{}
if err = proto.Unmarshal(data, pb); err != nil {
return -1, err
}
req.LeaderName = pb.GetLeaderName()
req.LastIndex = pb.GetLastIndex()
req.LastTerm = pb.GetLastTerm()
req.Peers = req.Peers
req.State = req.State
return totalBytes, nil
}

View File

@ -1,69 +0,0 @@
package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"io"
"io/ioutil"
)
// The response returned from a server appending entries to the log.
type SnapshotRecoveryResponse struct {
Term uint64
Success bool
CommitIndex uint64
}
//------------------------------------------------------------------------------
//
// Constructors
//
//------------------------------------------------------------------------------
// Creates a new Snapshot response.
func newSnapshotRecoveryResponse(term uint64, success bool, commitIndex uint64) *SnapshotRecoveryResponse {
return &SnapshotRecoveryResponse{
Term: term,
Success: success,
CommitIndex: commitIndex,
}
}
// Encodes the SnapshotRecoveryResponse to a buffer. Returns the number of bytes
// written and any error that may have occurred.
func (req *SnapshotRecoveryResponse) encode(w io.Writer) (int, error) {
pb := &protobuf.ProtoSnapshotRecoveryResponse{
Term: proto.Uint64(req.Term),
Success: proto.Bool(req.Success),
CommitIndex: proto.Uint64(req.CommitIndex),
}
p, err := proto.Marshal(pb)
if err != nil {
return -1, err
}
return w.Write(p)
}
// Decodes the SnapshotRecoveryResponse from a buffer. Returns the number of bytes read and
// any error that occurs.
func (req *SnapshotRecoveryResponse) decode(r io.Reader) (int, error) {
data, err := ioutil.ReadAll(r)
if err != nil {
return 0, err
}
totalBytes := len(data)
pb := &protobuf.ProtoSnapshotRecoveryResponse{}
if err := proto.Unmarshal(data, pb); err != nil {
return -1, err
}
req.Term = pb.GetTerm()
req.Success = pb.GetSuccess()
req.CommitIndex = pb.GetCommitIndex()
return totalBytes, nil
}

View File

@ -1,70 +0,0 @@
package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"io"
"io/ioutil"
)
// The request sent to a server to start from the snapshot.
type SnapshotRequest struct {
LeaderName string
LastIndex uint64
LastTerm uint64
}
//------------------------------------------------------------------------------
//
// Constructors
//
//------------------------------------------------------------------------------
// Creates a new Snapshot request.
func newSnapshotRequest(leaderName string, snapshot *Snapshot) *SnapshotRequest {
return &SnapshotRequest{
LeaderName: leaderName,
LastIndex: snapshot.LastIndex,
LastTerm: snapshot.LastTerm,
}
}
// Encodes the SnapshotRequest to a buffer. Returns the number of bytes
// written and any error that may have occurred.
func (req *SnapshotRequest) encode(w io.Writer) (int, error) {
pb := &protobuf.ProtoSnapshotRequest{
LeaderName: proto.String(req.LeaderName),
LastIndex: proto.Uint64(req.LastIndex),
LastTerm: proto.Uint64(req.LastTerm),
}
p, err := proto.Marshal(pb)
if err != nil {
return -1, err
}
return w.Write(p)
}
// Decodes the SnapshotRequest from a buffer. Returns the number of bytes read and
// any error that occurs.
func (req *SnapshotRequest) decode(r io.Reader) (int, error) {
data, err := ioutil.ReadAll(r)
if err != nil {
return 0, err
}
totalBytes := len(data)
pb := &protobuf.ProtoSnapshotRequest{}
if err := proto.Unmarshal(data, pb); err != nil {
return -1, err
}
req.LeaderName = pb.GetLeaderName()
req.LastIndex = pb.GetLastIndex()
req.LastTerm = pb.GetLastTerm()
return totalBytes, nil
}

View File

@ -1,61 +0,0 @@
package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"io"
"io/ioutil"
)
// The response returned if the follower entered snapshot state
type SnapshotResponse struct {
Success bool `json:"success"`
}
//------------------------------------------------------------------------------
//
// Constructors
//
//------------------------------------------------------------------------------
// Creates a new Snapshot response.
func newSnapshotResponse(success bool) *SnapshotResponse {
return &SnapshotResponse{
Success: success,
}
}
// Encodes the SnapshotResponse to a buffer. Returns the number of bytes
// written and any error that may have occurred.
func (resp *SnapshotResponse) encode(w io.Writer) (int, error) {
pb := &protobuf.ProtoSnapshotResponse{
Success: proto.Bool(resp.Success),
}
p, err := proto.Marshal(pb)
if err != nil {
return -1, err
}
return w.Write(p)
}
// Decodes the SnapshotResponse from a buffer. Returns the number of bytes read and
// any error that occurs.
func (resp *SnapshotResponse) decode(r io.Reader) (int, error) {
data, err := ioutil.ReadAll(r)
if err != nil {
return 0, err
}
totalBytes := len(data)
pb := &protobuf.ProtoSnapshotResponse{}
if err := proto.Unmarshal(data, pb); err != nil {
return -1, err
}
resp.Success = pb.GetSuccess()
return totalBytes, nil
}

View File

@ -1,23 +0,0 @@
package raft
//------------------------------------------------------------------------------
//
// Typedefs
//
//------------------------------------------------------------------------------
type uint64Slice []uint64
//------------------------------------------------------------------------------
//
// Functions
//
//------------------------------------------------------------------------------
//--------------------------------------
// uint64
//--------------------------------------
func (p uint64Slice) Len() int { return len(p) }
func (p uint64Slice) Less(i, j int) bool { return p[i] < p[j] }
func (p uint64Slice) Swap(i, j int) { p[i], p[j] = p[j], p[i] }

View File

@ -1,14 +0,0 @@
package raft
//------------------------------------------------------------------------------
//
// Typedefs
//
//------------------------------------------------------------------------------
// StateMachine is the interface for allowing the host application to save and
// recovery the state machine
type StateMachine interface {
Save() ([]byte, error)
Recovery([]byte) error
}

View File

@ -1,179 +0,0 @@
package raft
import (
"fmt"
"io/ioutil"
"os"
"time"
)
const (
testHeartbeatTimeout = 50 * time.Millisecond
testElectionTimeout = 200 * time.Millisecond
)
func init() {
RegisterCommand(&testCommand1{})
RegisterCommand(&testCommand2{})
}
//------------------------------------------------------------------------------
//
// Helpers
//
//------------------------------------------------------------------------------
//--------------------------------------
// Logs
//--------------------------------------
func getLogPath() string {
f, _ := ioutil.TempFile("", "raft-log-")
f.Close()
os.Remove(f.Name())
return f.Name()
}
func setupLog(entries []*LogEntry) (*Log, string) {
f, _ := ioutil.TempFile("", "raft-log-")
for _, entry := range entries {
entry.encode(f)
}
err := f.Close()
if err != nil {
panic(err)
}
log := newLog()
log.ApplyFunc = func(c Command) (interface{}, error) {
return nil, nil
}
if err := log.open(f.Name()); err != nil {
panic(err)
}
return log, f.Name()
}
//--------------------------------------
// Servers
//--------------------------------------
func newTestServer(name string, transporter Transporter) *Server {
p, _ := ioutil.TempDir("", "raft-server-")
if err := os.MkdirAll(p, 0644); err != nil {
panic(err.Error())
}
server, _ := NewServer(name, p, transporter, nil, nil)
return server
}
func newTestServerWithLog(name string, transporter Transporter, entries []*LogEntry) *Server {
server := newTestServer(name, transporter)
f, err := os.Create(server.LogPath())
if err != nil {
panic(err)
}
for _, entry := range entries {
entry.encode(f)
}
f.Close()
return server
}
func newTestCluster(names []string, transporter Transporter, lookup map[string]*Server) []*Server {
servers := []*Server{}
e0, _ := newLogEntry(newLog(), 1, 1, &testCommand1{Val: "foo", I: 20})
for _, name := range names {
if lookup[name] != nil {
panic(fmt.Sprintf("raft: Duplicate server in test cluster! %v", name))
}
server := newTestServerWithLog("1", transporter, []*LogEntry{e0})
server.SetElectionTimeout(testElectionTimeout)
servers = append(servers, server)
lookup[name] = server
}
for _, server := range servers {
server.SetHeartbeatTimeout(testHeartbeatTimeout)
server.Start()
for _, peer := range servers {
server.AddPeer(peer.Name())
}
}
return servers
}
//--------------------------------------
// Transporter
//--------------------------------------
type testTransporter struct {
sendVoteRequestFunc func(server *Server, peer *Peer, req *RequestVoteRequest) *RequestVoteResponse
sendAppendEntriesRequestFunc func(server *Server, peer *Peer, req *AppendEntriesRequest) *AppendEntriesResponse
sendSnapshotRequestFunc func(server *Server, peer *Peer, req *SnapshotRequest) *SnapshotResponse
}
func (t *testTransporter) SendVoteRequest(server *Server, peer *Peer, req *RequestVoteRequest) *RequestVoteResponse {
return t.sendVoteRequestFunc(server, peer, req)
}
func (t *testTransporter) SendAppendEntriesRequest(server *Server, peer *Peer, req *AppendEntriesRequest) *AppendEntriesResponse {
return t.sendAppendEntriesRequestFunc(server, peer, req)
}
func (t *testTransporter) SendSnapshotRequest(server *Server, peer *Peer, req *SnapshotRequest) *SnapshotResponse {
return t.sendSnapshotRequestFunc(server, peer, req)
}
func (t *testTransporter) SendSnapshotRecoveryRequest(server *Server, peer *Peer, req *SnapshotRecoveryRequest) *SnapshotRecoveryResponse {
return t.SendSnapshotRecoveryRequest(server, peer, req)
}
type testStateMachine struct {
saveFunc func() ([]byte, error)
recoveryFunc func([]byte) error
}
func (sm *testStateMachine) Save() ([]byte, error) {
return sm.saveFunc()
}
func (sm *testStateMachine) Recovery(state []byte) error {
return sm.recoveryFunc(state)
}
//--------------------------------------
// Command1
//--------------------------------------
type testCommand1 struct {
Val string `json:"val"`
I int `json:"i"`
}
func (c *testCommand1) CommandName() string {
return "cmd_1"
}
func (c *testCommand1) Apply(server *Server) (interface{}, error) {
return nil, nil
}
//--------------------------------------
// Command2
//--------------------------------------
type testCommand2 struct {
X int `json:"x"`
}
func (c *testCommand2) CommandName() string {
return "cmd_2"
}
func (c *testCommand2) Apply(server *Server) (interface{}, error) {
return nil, nil
}

View File

@ -1,17 +0,0 @@
package raft
import (
"math/rand"
"time"
)
// Waits for a random time between two durations and sends the current time on
// the returned channel.
func afterBetween(min time.Duration, max time.Duration) <-chan time.Time {
rand := rand.New(rand.NewSource(time.Now().UnixNano()))
d, delta := min, (max - min)
if delta > 0 {
d += time.Duration(rand.Int63n(int64(delta)))
}
return time.After(d)
}

View File

@ -1,16 +0,0 @@
package raft
//------------------------------------------------------------------------------
//
// Typedefs
//
//------------------------------------------------------------------------------
// Transporter is the interface for allowing the host application to transport
// requests to other nodes.
type Transporter interface {
SendVoteRequest(server *Server, peer *Peer, req *RequestVoteRequest) *RequestVoteResponse
SendAppendEntriesRequest(server *Server, peer *Peer, req *AppendEntriesRequest) *AppendEntriesResponse
SendSnapshotRequest(server *Server, peer *Peer, req *SnapshotRequest) *SnapshotResponse
SendSnapshotRecoveryRequest(server *Server, peer *Peer, req *SnapshotRecoveryRequest) *SnapshotRecoveryResponse
}

View File

@ -1,13 +0,0 @@
package raft
/*
import (
"testing"
"time"
)
func TestGC(t *testing.T) {
<-time.After(500 * time.Millisecond)
panic("Oh god no!")
}
*/

View File

@ -18,10 +18,13 @@ package config
import (
"bufio"
"errors"
"os"
"strings"
)
var commentPrefix = []string{"//", "#", ";"}
func Read(filename string) (map[string]string, error) {
var res = map[string]string{}
in, err := os.Open(filename)
@ -30,11 +33,19 @@ func Read(filename string) (map[string]string, error) {
}
scanner := bufio.NewScanner(in)
line := ""
section := ""
for scanner.Scan() {
if strings.HasPrefix(scanner.Text(), "//") {
if scanner.Text() == "" {
continue
}
if strings.HasPrefix(scanner.Text(), "#") {
if line == "" {
sec := checkSection(scanner.Text())
if sec != "" {
section = sec + "."
continue
}
}
if checkComment(scanner.Text()) {
continue
}
line += scanner.Text()
@ -42,13 +53,47 @@ func Read(filename string) (map[string]string, error) {
line = line[:len(line)-1]
continue
}
sp := strings.SplitN(line, "=", 2)
if len(sp) != 2 {
continue
key, value, err := checkLine(line)
if err != nil {
return res, errors.New("WRONG: " + line)
}
res[strings.TrimSpace(sp[0])] = strings.TrimSpace(sp[1])
res[section+key] = value
line = ""
}
in.Close()
return res, nil
}
func checkSection(line string) string {
line = strings.TrimSpace(line)
lineLen := len(line)
if lineLen < 2 {
return ""
}
if line[0] == '[' && line[lineLen-1] == ']' {
return line[1 : lineLen-1]
}
return ""
}
func checkLine(line string) (string, string, error) {
key := ""
value := ""
sp := strings.SplitN(line, "=", 2)
if len(sp) != 2 {
return key, value, errors.New("WRONG: " + line)
}
key = strings.TrimSpace(sp[0])
value = strings.TrimSpace(sp[1])
return key, value, nil
}
func checkComment(line string) bool {
line = strings.TrimSpace(line)
for p := range commentPrefix {
if strings.HasPrefix(line, commentPrefix[p]) {
return true
}
}
return false
}

View File

@ -5,3 +5,6 @@ cc = dd, 2 ejkl ijfadjfl
# 12jfiahdoif
dd = c \
oadi
[test]
a = c c d

View File

@ -0,0 +1,241 @@
package etcd
import (
"crypto/tls"
"errors"
"io/ioutil"
"net"
"net/http"
"path"
"strings"
"time"
)
const (
HTTP = iota
HTTPS
)
type Cluster struct {
Leader string
Machines []string
}
type Config struct {
CertFile string
KeyFile string
Scheme string
Timeout time.Duration
}
type Client struct {
cluster Cluster
config Config
httpClient *http.Client
}
// Setup a basic conf and cluster
func NewClient() *Client {
// default leader and machines
cluster := Cluster{
Leader: "0.0.0.0:4001",
Machines: make([]string, 1),
}
cluster.Machines[0] = "0.0.0.0:4001"
config := Config{
// default use http
Scheme: "http",
// default timeout is one second
Timeout: time.Second,
}
tr := &http.Transport{
Dial: dialTimeout,
TLSClientConfig: &tls.Config{
InsecureSkipVerify: true,
},
}
return &Client{
cluster: cluster,
config: config,
httpClient: &http.Client{Transport: tr},
}
}
func (c *Client) SetCertAndKey(cert string, key string) (bool, error) {
if cert != "" && key != "" {
tlsCert, err := tls.LoadX509KeyPair(cert, key)
if err != nil {
return false, err
}
tr := &http.Transport{
TLSClientConfig: &tls.Config{
Certificates: []tls.Certificate{tlsCert},
InsecureSkipVerify: true,
},
Dial: dialTimeout,
}
c.httpClient = &http.Client{Transport: tr}
return true, nil
}
return false, errors.New("Require both cert and key path")
}
func (c *Client) SetScheme(scheme int) (bool, error) {
if scheme == HTTP {
c.config.Scheme = "http"
return true, nil
}
if scheme == HTTPS {
c.config.Scheme = "https"
return true, nil
}
return false, errors.New("Unknown Scheme")
}
// Try to sync from the given machine
func (c *Client) SetCluster(machines []string) bool {
success := c.internalSyncCluster(machines)
return success
}
// sycn cluster information using the existing machine list
func (c *Client) SyncCluster() bool {
success := c.internalSyncCluster(c.cluster.Machines)
return success
}
// sync cluster information by providing machine list
func (c *Client) internalSyncCluster(machines []string) bool {
for _, machine := range machines {
httpPath := c.createHttpPath(machine, "machines")
resp, err := c.httpClient.Get(httpPath)
if err != nil {
// try another machine in the cluster
continue
} else {
b, err := ioutil.ReadAll(resp.Body)
if err != nil {
// try another machine in the cluster
continue
}
// update Machines List
c.cluster.Machines = strings.Split(string(b), ",")
logger.Debug("sync.machines ", c.cluster.Machines)
return true
}
}
return false
}
// serverName should contain both hostName and port
func (c *Client) createHttpPath(serverName string, _path string) string {
httpPath := path.Join(serverName, _path)
httpPath = c.config.Scheme + "://" + httpPath
return httpPath
}
// Dial with timeout.
func dialTimeout(network, addr string) (net.Conn, error) {
return net.DialTimeout(network, addr, time.Second)
}
func (c *Client) getHttpPath(s ...string) string {
httpPath := path.Join(c.cluster.Leader, version)
for _, seg := range s {
httpPath = path.Join(httpPath, seg)
}
httpPath = c.config.Scheme + "://" + httpPath
return httpPath
}
func (c *Client) updateLeader(httpPath string) {
// httpPath http://127.0.0.1:4001/v1...
leader := strings.Split(httpPath, "://")[1]
// we want to have 127.0.0.1:4001
leader = strings.Split(leader, "/")[0]
logger.Debugf("update.leader[%s,%s]", c.cluster.Leader, leader)
c.cluster.Leader = leader
}
// Wrap GET, POST and internal error handling
func (c *Client) sendRequest(method string, _path string, body string) (*http.Response, error) {
var resp *http.Response
var err error
var req *http.Request
retry := 0
// if we connect to a follower, we will retry until we found a leader
for {
httpPath := c.getHttpPath(_path)
logger.Debug("send.request.to ", httpPath)
if body == "" {
req, _ = http.NewRequest(method, httpPath, nil)
} else {
req, _ = http.NewRequest(method, httpPath, strings.NewReader(body))
req.Header.Set("Content-Type", "application/x-www-form-urlencoded; param=value")
}
resp, err = c.httpClient.Do(req)
logger.Debug("recv.response.from ", httpPath)
// network error, change a machine!
if err != nil {
retry++
if retry > 2*len(c.cluster.Machines) {
return nil, errors.New("Cannot reach servers")
}
num := retry % len(c.cluster.Machines)
logger.Debug("update.leader[", c.cluster.Leader, ",", c.cluster.Machines[num], "]")
c.cluster.Leader = c.cluster.Machines[num]
time.Sleep(time.Millisecond * 200)
continue
}
if resp != nil {
if resp.StatusCode == http.StatusTemporaryRedirect {
httpPath := resp.Header.Get("Location")
resp.Body.Close()
if httpPath == "" {
return nil, errors.New("Cannot get redirection location")
}
c.updateLeader(httpPath)
logger.Debug("send.redirect")
// try to connect the leader
continue
} else if resp.StatusCode == http.StatusInternalServerError {
retry++
if retry > 2*len(c.cluster.Machines) {
return nil, errors.New("Cannot reach servers")
}
resp.Body.Close()
continue
} else {
logger.Debug("send.return.response ", httpPath)
break
}
}
logger.Debug("error.from ", httpPath, " ", err.Error())
return nil, err
}
return resp, nil
}

View File

@ -0,0 +1,38 @@
package etcd
import (
"fmt"
"testing"
)
// To pass this test, we need to create a cluster of 3 machines
// The server should be listening on 127.0.0.1:4001, 4002, 4003
func TestSync(t *testing.T) {
fmt.Println("Make sure there are three nodes at 0.0.0.0:4001-4003")
c := NewClient()
success := c.SyncCluster()
if !success {
t.Fatal("cannot sync machines")
}
badMachines := []string{"abc", "edef"}
success = c.SetCluster(badMachines)
if success {
t.Fatal("should not sync on bad machines")
}
goodMachines := []string{"127.0.0.1:4002"}
success = c.SetCluster(goodMachines)
if !success {
t.Fatal("cannot sync machines")
} else {
fmt.Println(c.cluster.Machines)
}
}

View File

@ -0,0 +1,19 @@
package etcd
import (
"github.com/ccding/go-logging/logging"
)
var logger, _ = logging.SimpleLogger("go-etcd")
func init() {
logger.SetLevel(logging.FATAL)
}
func OpenDebug() {
logger.SetLevel(logging.NOTSET)
}
func CloseDebug() {
logger.SetLevel(logging.FATAL)
}

View File

@ -0,0 +1,41 @@
package etcd
import (
"encoding/json"
"github.com/coreos/etcd/store"
"io/ioutil"
"net/http"
"path"
)
func (c *Client) Delete(key string) (*store.Response, error) {
resp, err := c.sendRequest("DELETE", path.Join("keys", key), "")
if err != nil {
return nil, err
}
b, err := ioutil.ReadAll(resp.Body)
resp.Body.Close()
if err != nil {
return nil, err
}
if resp.StatusCode != http.StatusOK {
return nil, handleError(b)
}
var result store.Response
err = json.Unmarshal(b, &result)
if err != nil {
return nil, err
}
return &result, nil
}

View File

@ -0,0 +1,22 @@
package etcd
import (
"testing"
)
func TestDelete(t *testing.T) {
c := NewClient()
c.Set("foo", "bar", 100)
result, err := c.Delete("foo")
if err != nil {
t.Fatal(err)
}
if result.PrevValue != "bar" || result.Value != "" {
t.Fatalf("Delete failed with %s %s", result.PrevValue,
result.Value)
}
}

View File

@ -0,0 +1,24 @@
package etcd
import (
"encoding/json"
"fmt"
)
type EtcdError struct {
ErrorCode int `json:"errorCode"`
Message string `json:"message"`
Cause string `json:"cause,omitempty"`
}
func (e EtcdError) Error() string {
return fmt.Sprintf("%d: %s (%s)", e.ErrorCode, e.Message, e.Cause)
}
func handleError(b []byte) error {
var err EtcdError
json.Unmarshal(b, &err)
return err
}

View File

@ -0,0 +1,83 @@
package etcd
import (
"encoding/json"
"github.com/coreos/etcd/store"
"io/ioutil"
"net/http"
"path"
)
func (c *Client) Get(key string) ([]*store.Response, error) {
logger.Debugf("get %s [%s]", key, c.cluster.Leader)
resp, err := c.sendRequest("GET", path.Join("keys", key), "")
if err != nil {
return nil, err
}
b, err := ioutil.ReadAll(resp.Body)
resp.Body.Close()
if err != nil {
return nil, err
}
if resp.StatusCode != http.StatusOK {
return nil, handleError(b)
}
return convertGetResponse(b)
}
// GetTo gets the value of the key from a given machine address.
// If the given machine is not available it returns an error.
// Mainly use for testing purpose
func (c *Client) GetFrom(key string, addr string) ([]*store.Response, error) {
httpPath := c.createHttpPath(addr, path.Join(version, "keys", key))
resp, err := c.httpClient.Get(httpPath)
if err != nil {
return nil, err
}
b, err := ioutil.ReadAll(resp.Body)
resp.Body.Close()
if err != nil {
return nil, err
}
if resp.StatusCode != http.StatusOK {
return nil, handleError(b)
}
return convertGetResponse(b)
}
// Convert byte stream to response.
func convertGetResponse(b []byte) ([]*store.Response, error) {
var results []*store.Response
var result *store.Response
err := json.Unmarshal(b, &result)
if err != nil {
err = json.Unmarshal(b, &results)
if err != nil {
return nil, err
}
} else {
results = make([]*store.Response, 1)
results[0] = result
}
return results, nil
}

View File

@ -0,0 +1,46 @@
package etcd
import (
"testing"
"time"
)
func TestGet(t *testing.T) {
c := NewClient()
c.Set("foo", "bar", 100)
// wait for commit
time.Sleep(100 * time.Millisecond)
results, err := c.Get("foo")
if err != nil || results[0].Key != "/foo" || results[0].Value != "bar" {
if err != nil {
t.Fatal(err)
}
t.Fatalf("Get failed with %s %s %v", results[0].Key, results[0].Value, results[0].TTL)
}
results, err = c.Get("goo")
if err == nil {
t.Fatalf("should not be able to get non-exist key")
}
results, err = c.GetFrom("foo", "0.0.0.0:4001")
if err != nil || results[0].Key != "/foo" || results[0].Value != "bar" {
if err != nil {
t.Fatal(err)
}
t.Fatalf("Get failed with %s %s %v", results[0].Key, results[0].Value, results[0].TTL)
}
results, err = c.GetFrom("foo", "0.0.0.0:4009")
if err == nil {
t.Fatal("should not get from port 4009")
}
}

View File

@ -0,0 +1,23 @@
package etcd
import (
"testing"
"time"
)
func TestList(t *testing.T) {
c := NewClient()
c.Set("foo_list/foo", "bar", 100)
c.Set("foo_list/fooo", "barbar", 100)
c.Set("foo_list/foooo/foo", "barbarbar", 100)
// wait for commit
time.Sleep(time.Second)
_, err := c.Get("foo_list")
if err != nil {
t.Fatal(err)
}
}

View File

@ -0,0 +1,90 @@
package etcd
import (
"encoding/json"
"fmt"
"github.com/coreos/etcd/store"
"io/ioutil"
"net/http"
"net/url"
"path"
)
func (c *Client) Set(key string, value string, ttl uint64) (*store.Response, error) {
logger.Debugf("set %s, %s, ttl: %d, [%s]", key, value, ttl, c.cluster.Leader)
v := url.Values{}
v.Set("value", value)
if ttl > 0 {
v.Set("ttl", fmt.Sprintf("%v", ttl))
}
resp, err := c.sendRequest("POST", path.Join("keys", key), v.Encode())
if err != nil {
return nil, err
}
b, err := ioutil.ReadAll(resp.Body)
resp.Body.Close()
if err != nil {
return nil, err
}
if resp.StatusCode != http.StatusOK {
return nil, handleError(b)
}
return convertSetResponse(b)
}
// SetTo sets the value of the key to a given machine address.
// If the given machine is not available or is not leader it returns an error
// Mainly use for testing purpose.
func (c *Client) SetTo(key string, value string, ttl uint64, addr string) (*store.Response, error) {
v := url.Values{}
v.Set("value", value)
if ttl > 0 {
v.Set("ttl", fmt.Sprintf("%v", ttl))
}
httpPath := c.createHttpPath(addr, path.Join(version, "keys", key))
resp, err := c.httpClient.PostForm(httpPath, v)
if err != nil {
return nil, err
}
b, err := ioutil.ReadAll(resp.Body)
resp.Body.Close()
if err != nil {
return nil, err
}
if resp.StatusCode != http.StatusOK {
return nil, handleError(b)
}
return convertSetResponse(b)
}
// Convert byte stream to response.
func convertSetResponse(b []byte) (*store.Response, error) {
var result store.Response
err := json.Unmarshal(b, &result)
if err != nil {
return nil, err
}
return &result, nil
}

View File

@ -0,0 +1,42 @@
package etcd
import (
"testing"
"time"
)
func TestSet(t *testing.T) {
c := NewClient()
result, err := c.Set("foo", "bar", 100)
if err != nil || result.Key != "/foo" || result.Value != "bar" || result.TTL != 99 {
if err != nil {
t.Fatal(err)
}
t.Fatalf("Set 1 failed with %s %s %v", result.Key, result.Value, result.TTL)
}
time.Sleep(time.Second)
result, err = c.Set("foo", "bar", 100)
if err != nil || result.Key != "/foo" || result.Value != "bar" || result.PrevValue != "bar" || result.TTL != 99 {
if err != nil {
t.Fatal(err)
}
t.Fatalf("Set 2 failed with %s %s %v", result.Key, result.Value, result.TTL)
}
result, err = c.SetTo("toFoo", "bar", 100, "0.0.0.0:4001")
if err != nil || result.Key != "/toFoo" || result.Value != "bar" || result.TTL != 99 {
if err != nil {
t.Fatal(err)
}
t.Fatalf("SetTo failed with %s %s %v", result.Key, result.Value, result.TTL)
}
}

View File

@ -0,0 +1,57 @@
package etcd
import (
"encoding/json"
"fmt"
"github.com/coreos/etcd/store"
"io/ioutil"
"net/http"
"net/url"
"path"
)
func (c *Client) TestAndSet(key string, prevValue string, value string, ttl uint64) (*store.Response, bool, error) {
logger.Debugf("set %s, %s[%s], ttl: %d, [%s]", key, value, prevValue, ttl, c.cluster.Leader)
v := url.Values{}
v.Set("value", value)
v.Set("prevValue", prevValue)
if ttl > 0 {
v.Set("ttl", fmt.Sprintf("%v", ttl))
}
resp, err := c.sendRequest("POST", path.Join("keys", key), v.Encode())
if err != nil {
return nil, false, err
}
b, err := ioutil.ReadAll(resp.Body)
resp.Body.Close()
if err != nil {
return nil, false, err
}
if resp.StatusCode != http.StatusOK {
return nil, false, handleError(b)
}
var result store.Response
err = json.Unmarshal(b, &result)
if err != nil {
return nil, false, err
}
if result.PrevValue == prevValue && result.Value == value {
return &result, true, nil
}
return &result, false, nil
}

View File

@ -0,0 +1,39 @@
package etcd
import (
"testing"
"time"
)
func TestTestAndSet(t *testing.T) {
c := NewClient()
c.Set("foo_testAndSet", "bar", 100)
time.Sleep(time.Second)
results := make(chan bool, 3)
for i := 0; i < 3; i++ {
testAndSet("foo_testAndSet", "bar", "barbar", results, c)
}
count := 0
for i := 0; i < 3; i++ {
result := <-results
if result {
count++
}
}
if count != 1 {
t.Fatalf("test and set fails %v", count)
}
}
func testAndSet(key string, prevValue string, value string, ch chan bool, c *Client) {
_, success, _ := c.TestAndSet(key, prevValue, value, 0)
ch <- success
}

View File

@ -0,0 +1,3 @@
package etcd
var version = "v1"

View File

@ -0,0 +1,117 @@
package etcd
import (
"encoding/json"
"errors"
"fmt"
"github.com/coreos/etcd/store"
"io/ioutil"
"net/http"
"net/url"
"path"
)
type respAndErr struct {
resp *http.Response
err error
}
// Watch any change under the given prefix.
// When a sinceIndex is given, watch will try to scan from that index to the last index
// and will return any changes under the given prefix during the history
// If a receiver channel is given, it will be a long-term watch. Watch will block at the
// channel. And after someone receive the channel, it will go on to watch that prefix.
// If a stop channel is given, client can close long-term watch using the stop channel
func (c *Client) Watch(prefix string, sinceIndex uint64, receiver chan *store.Response, stop chan bool) (*store.Response, error) {
logger.Debugf("watch %s [%s]", prefix, c.cluster.Leader)
if receiver == nil {
return c.watchOnce(prefix, sinceIndex, stop)
} else {
for {
resp, err := c.watchOnce(prefix, sinceIndex, stop)
if resp != nil {
sinceIndex = resp.Index + 1
receiver <- resp
} else {
return nil, err
}
}
}
return nil, nil
}
// helper func
// return when there is change under the given prefix
func (c *Client) watchOnce(key string, sinceIndex uint64, stop chan bool) (*store.Response, error) {
var resp *http.Response
var err error
if sinceIndex == 0 {
// Get request if no index is given
resp, err = c.sendRequest("GET", path.Join("watch", key), "")
if err != nil {
return nil, err
}
} else {
// Post
v := url.Values{}
v.Set("index", fmt.Sprintf("%v", sinceIndex))
ch := make(chan respAndErr)
if stop != nil {
go func() {
resp, err = c.sendRequest("POST", path.Join("watch", key), v.Encode())
ch <- respAndErr{resp, err}
}()
// select at stop or continue to receive
select {
case res := <-ch:
resp, err = res.resp, res.err
case <-stop:
resp, err = nil, errors.New("User stoped watch")
}
} else {
resp, err = c.sendRequest("POST", path.Join("watch", key), v.Encode())
}
if err != nil {
return nil, err
}
}
b, err := ioutil.ReadAll(resp.Body)
resp.Body.Close()
if err != nil {
return nil, err
}
if resp.StatusCode != http.StatusOK {
return nil, handleError(b)
}
var result store.Response
err = json.Unmarshal(b, &result)
if err != nil {
return nil, err
}
return &result, nil
}

View File

@ -0,0 +1,62 @@
package etcd
import (
"fmt"
"github.com/coreos/etcd/store"
"testing"
"time"
)
func TestWatch(t *testing.T) {
c := NewClient()
go setHelper("bar", c)
result, err := c.Watch("watch_foo", 0, nil, nil)
if err != nil || result.Key != "/watch_foo/foo" || result.Value != "bar" {
if err != nil {
t.Fatal(err)
}
t.Fatalf("Watch failed with %s %s %v %v", result.Key, result.Value, result.TTL, result.Index)
}
result, err = c.Watch("watch_foo", result.Index, nil, nil)
if err != nil || result.Key != "/watch_foo/foo" || result.Value != "bar" {
if err != nil {
t.Fatal(err)
}
t.Fatalf("Watch with Index failed with %s %s %v %v", result.Key, result.Value, result.TTL, result.Index)
}
ch := make(chan *store.Response, 10)
stop := make(chan bool, 1)
go setLoop("bar", c)
go reciver(ch, stop)
c.Watch("watch_foo", 0, ch, stop)
}
func setHelper(value string, c *Client) {
time.Sleep(time.Second)
c.Set("watch_foo/foo", value, 100)
}
func setLoop(value string, c *Client) {
time.Sleep(time.Second)
for i := 0; i < 10; i++ {
newValue := fmt.Sprintf("%s_%v", value, i)
c.Set("watch_foo/foo", newValue, 100)
time.Sleep(time.Second / 10)
}
}
func reciver(c chan *store.Response, stop chan bool) {
for i := 0; i < 10; i++ {
<-c
}
stop <- true
}

View File

@ -1,7 +1,7 @@
all: test
coverage:
gocov test github.com/benbjohnson/go-raft | gocov-html > coverage.html
gocov test github.com/coreos/go-raft | gocov-html > coverage.html
open coverage.html
dependencies:

View File

@ -1,4 +1,5 @@
[![Stories in Ready](http://badge.waffle.io/benbjohnson/go-raft.png)](http://waffle.io/benbjohnson/go-raft)
[![Build Status](https://travis-ci.org/benbjohnson/go-raft.png?branch=master)](https://travis-ci.org/benbjohnson/go-raft)
go-raft
=======

View File

@ -2,7 +2,7 @@ package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
"io/ioutil"
)

View File

@ -2,7 +2,7 @@ package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
"io/ioutil"
)

View File

@ -5,7 +5,7 @@ import (
"code.google.com/p/goprotobuf/proto"
"errors"
"fmt"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
"os"
"sync"
@ -141,9 +141,6 @@ func (l *Log) currentTerm() uint64 {
// Opens the log file and reads existing entries. The log can remain open and
// continue to append entries to the end of the log.
func (l *Log) open(path string) error {
l.mutex.Lock()
defer l.mutex.Unlock()
// Read all the entries from the log if one exists.
var readBytes int64
@ -168,7 +165,6 @@ func (l *Log) open(path string) error {
// Read the file and decode entries.
for {
// Instantiate log entry and decode into it.
entry, _ := newLogEntry(l, 0, 0, nil)
entry.Position, _ = l.file.Seek(0, os.SEEK_CUR)
@ -192,6 +188,9 @@ func (l *Log) open(path string) error {
readBytes += int64(n)
}
l.results = make([]*logResult, len(l.entries))
l.compact(l.startIndex, l.startTerm)
debugln("open.log.recovery number of log ", len(l.entries))
return nil
}
@ -282,9 +281,9 @@ func (l *Log) getEntryResult(entry *LogEntry, clear bool) (interface{}, error) {
if entry == nil {
panic("raft: Log entry required for error retrieval")
}
debugln("getEntryResult.result index: ", entry.Index-l.startIndex-1)
// If a result exists for the entry then return it with its error.
if entry.Index > l.startIndex && entry.Index <= uint64(len(l.results)) {
if entry.Index > l.startIndex && entry.Index <= l.startIndex+uint64(len(l.results)) {
if result := l.results[entry.Index-l.startIndex-1]; result != nil {
// keep the records before remove it
@ -310,8 +309,7 @@ func (l *Log) getEntryResult(entry *LogEntry, clear bool) (interface{}, error) {
func (l *Log) commitInfo() (index uint64, term uint64) {
l.mutex.RLock()
defer l.mutex.RUnlock()
// If we don't have any entries then just return zeros.
// If we don't have any committed entries then just return zeros.
if l.commitIndex == 0 {
return 0, 0
}
@ -322,6 +320,7 @@ func (l *Log) commitInfo() (index uint64, term uint64) {
}
// Return the last index & term from the last committed entry.
debugln("commitInfo.get.[", l.commitIndex, "/", l.startIndex, "]")
entry := l.entries[l.commitIndex-1-l.startIndex]
return entry.Index, entry.Term
}
@ -395,6 +394,7 @@ func (l *Log) setCommitIndex(index uint64) error {
// Apply the changes to the state machine and store the error code.
returnValue, err := l.ApplyFunc(command)
debugln("setCommitIndex.set.result index: ", entryIndex)
l.results[entryIndex] = &logResult{returnValue: returnValue, err: err}
}
return nil
@ -555,22 +555,27 @@ func (l *Log) writeEntry(entry *LogEntry, w io.Writer) (int64, error) {
// Log compaction
//--------------------------------------
// compaction the log before index
// compact the log before index (including index)
func (l *Log) compact(index uint64, term uint64) error {
var entries []*LogEntry
var results []*logResult
l.mutex.Lock()
defer l.mutex.Unlock()
if index == 0 {
return nil
}
// nothing to compaction
// the index may be greater than the current index if
// we just recovery from on snapshot
if index >= l.internalCurrentIndex() {
entries = make([]*LogEntry, 0)
results = make([]*logResult, 0)
} else {
// get all log entries after index
entries = l.entries[index-l.startIndex:]
results = l.results[index-l.startIndex:]
}
// create a new log file and add all the entries
@ -604,6 +609,7 @@ func (l *Log) compact(index uint64, term uint64) error {
// compaction the in memory log
l.entries = entries
l.results = results
l.startIndex = index
l.startTerm = term
return nil

View File

@ -5,7 +5,7 @@ import (
"code.google.com/p/goprotobuf/proto"
"encoding/json"
"fmt"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
)

View File

@ -2,7 +2,7 @@ package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
"io/ioutil"
)

View File

@ -2,7 +2,7 @@ package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
"io/ioutil"
)

View File

@ -242,7 +242,7 @@ func (s *Server) LastCommandName() string {
func (s *Server) GetState() string {
s.mutex.RLock()
defer s.mutex.RUnlock()
return fmt.Sprintf("Name: %s, State: %s, Term: %v, Index: %v ", s.name, s.state, s.currentTerm, s.log.commitIndex)
return fmt.Sprintf("Name: %s, State: %s, Term: %v, CommitedIndex: %v ", s.name, s.state, s.currentTerm, s.log.commitIndex)
}
// Check if the server is promotable
@ -361,6 +361,8 @@ func (s *Server) Start() error {
s.debugln("start from previous saved state")
}
debugln(s.GetState())
go s.loop()
return nil
@ -385,6 +387,8 @@ func (s *Server) readConf() error {
return err
}
peerNames := make([]string, 0)
for {
var peerName string
_, err = fmt.Fscanf(s.confFile, "%s\n", &peerName)
@ -392,16 +396,20 @@ func (s *Server) readConf() error {
if err != nil {
if err == io.EOF {
s.debugln("server.peer.conf: finish")
return nil
break
}
return err
}
s.debugln("server.peer.conf.read: ", peerName)
peer := newPeer(s, peerName, s.heartbeatTimeout)
peerNames = append(peerNames, peerName)
}
s.peers[peer.name] = peer
s.confFile.Truncate(0)
s.confFile.Seek(0, os.SEEK_SET)
for _, peerName := range peerNames {
s.AddPeer(peerName)
}
return nil
@ -961,10 +969,13 @@ func (s *Server) AddPeer(name string) error {
// Only add the peer if it doesn't have the same name.
if s.name != name {
_, err := fmt.Fprintln(s.confFile, name)
s.debugln("server.peer.conf.write: ", name)
if err != nil {
return err
// when loading snapshot s.confFile should be nil
if s.confFile != nil {
_, err := fmt.Fprintln(s.confFile, name)
s.debugln("server.peer.conf.write: ", name)
if err != nil {
return err
}
}
peer := newPeer(s, name, s.heartbeatTimeout)
if s.State() == Leader {
@ -1019,7 +1030,6 @@ func (s *Server) Snapshot() {
for {
// TODO: change this... to something reasonable
time.Sleep(1 * time.Second)
s.takeSnapshot()
}
}
@ -1033,7 +1043,7 @@ func (s *Server) takeSnapshot() error {
lastIndex, lastTerm := s.log.commitInfo()
if lastIndex == 0 || lastTerm == 0 {
if lastIndex == 0 {
return errors.New("No logs")
}

View File

@ -2,7 +2,7 @@ package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
"io/ioutil"
)

View File

@ -2,7 +2,7 @@ package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
"io/ioutil"
)

View File

@ -2,7 +2,7 @@ package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
"io/ioutil"
)

View File

@ -2,7 +2,7 @@ package raft
import (
"code.google.com/p/goprotobuf/proto"
"github.com/benbjohnson/go-raft/protobuf"
"github.com/coreos/go-raft/protobuf"
"io"
"io/ioutil"
)

1
third_party/update vendored
View File

@ -3,7 +3,6 @@
packages="
github.com/coreos/go-raft
github.com/coreos/go-etcd
github.com/benbjohnson/go-raft
github.com/ccding/go-logging
github.com/ccding/go-config-reader
bitbucket.org/kardianos/osext

View File

@ -1,5 +1,3 @@
package main
var version = "v1"
var releaseVersion = "etcd pre-0.1"
const version = "v1"