README
¶
raft
This is an implementation of the Raft distributed consensus protocol. It's heavily influenced by benbjohnson's implementation. It focuses on providing a clean and usable API, and well-structured internals.
Usage
A node in a Raft network is represented by a Server structure. In a typical application, nodes will create a Server, and expose it to other nodes using a Peer interface.
Servers are only useful when they can communicate with other servers. This library includes a HTTP Transport (ingress) and HTTP Peer (egress) which combine to allow communication via REST-ish endpoints. For now, it's the simplest way to embed a Raft server in your application. See this complete example.
Several other transports are coming; see TODO, below.
Adding and removing nodes
The Raft protocol has no affordance for node discovery or "join/leave" semantics. Rather, the protocol assumes an ideal network configuration that's known a priori to nodes in the network, and describes a mechanism (called joint-consensus) to safely replicate that configuration.
My implementation of joint-consensus abides those fundamental assumptions. Nodes may be added or removed dynamically by requesting a SetConfiguration that describes a complete network topology.
TODO
Leader electiondoneLog replicationdoneBasic unit testsdoneHTTP transportdone- net/rpc transport
- Other transports?
Configuration changes (joint-consensus mode)done- Log compaction
- Robust demo application ☜ in progress
- Complex unit tests (one per scenario described in the paper)
Documentation
¶
Overview ¶
Package raft is an implementation of the Raft distributed consensus protocol.
Index ¶
Examples ¶
Constants ¶
This section is empty.
Variables ¶
var ( // IDPath is where the ID handler (GET) will be installed by the // HTTPTransport. IDPath = "/raft/id" // AppendEntriesPath is where the AppendEntries RPC handler (POST) will be // installed by the HTTPTransport. AppendEntriesPath = "/raft/appendentries" // RequestVotePath is where the requestVote RPC handler (POST) will be // installed by the HTTPTransport. RequestVotePath = "/raft/requestvote" // CommandPath is where the Command RPC handler (POST) will be installed by // the HTTPTransport. CommandPath = "/raft/command" // SetConfigurationPath is where the SetConfiguration RPC handler (POST) // will be installed by the HTTPTransport. SetConfigurationPath = "/raft/setconfiguration" )
var ( // MinimumElectionTimeoutMS can be set at package initialization. It may be // raised to achieve more reliable replication in slow networks, or lowered // to achieve faster replication in fast networks. Lowering is not // recommended. MinimumElectionTimeoutMS int32 = 250 )
Functions ¶
func HTTPTransport ¶
HTTPTransport creates an ingress bridge from the outside world to the passed server, by installing handlers for all the necessary RPCs to the passed mux.
Types ¶
type ApplyFunc ¶
ApplyFunc is a client-provided function that should apply a successfully replicated state transition, represented by cmd, to the local state machine, and return a response. commitIndex is the sequence number of the state transition, which is guaranteed to be gapless and monotonically increasing, but not necessarily duplicate-free. ApplyFuncs are not called concurrently. Therefore, clients should ensure they return quickly, i.e. << MinimumElectionTimeout.
type Peer ¶
type Peer interface {
// contains filtered or unexported methods
}
Peer is the local representation of a remote node. It's an interface that may be backed by any concrete transport: local, HTTP, net/rpc, etc. Peers must be encoding/gob encodable.
type Server ¶
type Server struct {
// contains filtered or unexported fields
}
Server is the agent that performs all of the Raft protocol logic. In a typical application, each running process that wants to be part of the distributed state machine will contain a server component.
func NewServer ¶
func NewServer(id uint64, store io.ReadWriter, a ApplyFunc) *Server
NewServer returns an initialized, un-started server. The ID must be unique in the Raft network, and greater than 0. The store will be used by the distributed log as a persistence layer. It's read-from during creation, in case a crashed server is restarted over an already-persisted log. Then, it's written-to during normal operations, when log entries are safely replicated. ApplyFunc will be called whenever a (user-domain) command has been safely replicated and committed to this server's log.
NewServer creates a server, but you'll need to couple it with a transport to make it usable. See the example(s) for usage scenarios.
Example (HTTP) ¶
package main
import (
"bytes"
"net/http"
"net/url"
"github.com/peterbourgon/raft"
)
func main() {
// A no-op ApplyFunc
a := func(uint64, []byte) []byte { return []byte{} }
// Helper function to parse URLs
mustParseURL := func(rawurl string) *url.URL {
u, err := url.Parse(rawurl)
if err != nil {
panic(err)
}
u.Path = ""
return u
}
// Helper function to construct HTTP Peers
mustNewHTTPPeer := func(u *url.URL) raft.Peer {
p, err := raft.NewHTTPPeer(u)
if err != nil {
panic(err)
}
return p
}
// Construct the server
s := raft.NewServer(1, &bytes.Buffer{}, a)
// Expose the server using a HTTP transport
raft.HTTPTransport(http.DefaultServeMux, s)
go http.ListenAndServe(":8080", nil)
// Set the initial server configuration
s.SetConfiguration(
mustNewHTTPPeer(mustParseURL("http://127.0.0.1:8080")), // this server
mustNewHTTPPeer(mustParseURL("http://10.1.1.11:8080")),
mustNewHTTPPeer(mustParseURL("http://10.1.1.12:8080")),
mustNewHTTPPeer(mustParseURL("http://10.1.1.13:8080")),
mustNewHTTPPeer(mustParseURL("http://10.1.1.14:8080")),
)
// Start the server
s.Start()
}
Output:
func (*Server) Command ¶
Command appends the passed command to the leader log. If error is nil, the command will eventually get replicated throughout the Raft network. When the command gets committed to the local server log, it's passed to the apply function, and the response from that function is provided on the passed response chan.
Example ¶
package main
import (
"bytes"
"fmt"
"github.com/peterbourgon/raft"
)
func main() {
// A no-op ApplyFunc that always returns "PONG"
ponger := func(uint64, []byte) []byte { return []byte(`PONG`) }
// Assuming you have a server started
s := raft.NewServer(1, &bytes.Buffer{}, ponger)
// Issue a command into the network
response := make(chan []byte)
if err := s.Command([]byte(`PING`), response); err != nil {
panic(err) // command not accepted
}
// After the command is replicated, we'll receive the response
fmt.Printf("%s\n", <-response)
}
Output:
func (*Server) SetConfiguration ¶
SetConfiguration sets the peers that this server will attempt to communicate with. The set peers should include a peer that represents this server. SetConfiguration must be called before starting the server. Calls to SetConfiguration after the server has been started will be replicated throughout the Raft network using the joint-consensus mechanism.
TODO we need to refactor how we parse entries: a single code path from any source (snapshot, persisted log at startup, or over the network) into the log, and as part of that flow, checking if the entry is a configuration and emitting it to the configuration structure. This implies an unfortunate coupling: whatever processes log entries must have both the configuration and the log as data sinks.
Source Files