tendermint/rpc/lib/client/ws_client.go
Anton Kaliaev f6539737de
new pubsub package
comment out failing consensus tests for now

rewrite rpc httpclient to use new pubsub package

import pubsub as tmpubsub, query as tmquery

make event IDs constants
EventKey -> EventTypeKey

rename EventsPubsub to PubSub

mempool does not use pubsub

rename eventsSub to pubsub

new subscribe API

fix channel size issues and consensus tests bugs

refactor rpc client

add missing discardFromChan method

add mutex

rename pubsub to eventBus

remove IsRunning from WSRPCConnection interface (not needed)

add a comment in broadcastNewRoundStepsAndVotes

rename registerEventCallbacks to broadcastNewRoundStepsAndVotes

See https://dave.cheney.net/2014/03/19/channel-axioms

stop eventBuses after reactor tests

remove unnecessary Unsubscribe

return subscribe helper function

move discardFromChan to where it is used

subscribe now returns an err

this gives us ability to refuse to subscribe if pubsub is at its max
capacity.

use context for control overflow

cache queries

handle err when subscribing in replay_test

rename testClientID to testSubscriber

extract var

set channel buffer capacity to 1 in replay_file

fix byzantine_test

unsubscribe from single event, not all events

refactor httpclient to return events to appropriate channels

return failing testReplayCrashBeforeWriteVote test

fix TestValidatorSetChanges

refactor code a bit

fix testReplayCrashBeforeWriteVote

add comment

fix TestValidatorSetChanges

fixes from Bucky's review

update comment [ci skip]

test TxEventBuffer

update changelog

fix TestValidatorSetChanges (2nd attempt)

only do wg.Done when no errors

benchmark event bus

create pubsub server inside NewEventBus

only expose config params (later if needed)

set buffer capacity to 0 so we are not testing cache

new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ}

This should allow to subscribe to all transactions! or a specific one
using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'"

use TimeoutCommit instead of afterPublishEventNewBlockTimeout

TimeoutCommit is the time a node waits after committing a block, before
it goes into the next height. So it will finish everything from the last
block, but then wait a bit. The idea is this gives it time to hear more
votes from other validators, to strengthen the commit it includes in the
next block. But it also gives it time to hear about new transactions.

waitForBlockWithUpdatedVals

rewrite WAL crash tests

Task:
test that we can recover from any WAL crash.

Solution:
the old tests were relying on event hub being run in the same thread (we
were injecting the private validator's last signature).

when considering a rewrite, we considered two possible solutions: write
a "fuzzy" testing system where WAL is crashing upon receiving a new
message, or inject failures and trigger them in tests using something
like https://github.com/coreos/gofail.

remove sleep

no cs.Lock around wal.Save

test different cases (empty block, non-empty block, ...)

comments

add comments

test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks

fixes as per Bucky's last review

reset subscriptions on UnsubscribeAll

use a simple counter to track message for which we panicked

also, set a smaller part size for all test cases
2017-10-30 00:32:22 -05:00

472 lines
12 KiB
Go

package rpcclient
import (
"context"
"encoding/json"
"fmt"
"math/rand"
"net"
"net/http"
"sync"
"time"
"github.com/gorilla/websocket"
"github.com/pkg/errors"
metrics "github.com/rcrowley/go-metrics"
types "github.com/tendermint/tendermint/rpc/lib/types"
cmn "github.com/tendermint/tmlibs/common"
)
const (
defaultMaxReconnectAttempts = 25
defaultWriteWait = 0
defaultReadWait = 0
defaultPingPeriod = 0
)
// WSClient is a WebSocket client. The methods of WSClient are safe for use by
// multiple goroutines.
type WSClient struct {
cmn.BaseService
conn *websocket.Conn
Address string // IP:PORT or /path/to/socket
Endpoint string // /websocket/url/endpoint
Dialer func(string, string) (net.Conn, error)
// Time between sending a ping and receiving a pong. See
// https://godoc.org/github.com/rcrowley/go-metrics#Timer.
PingPongLatencyTimer metrics.Timer
// Single user facing channel to read RPCResponses from, closed only when the client is being stopped.
ResponsesCh chan types.RPCResponse
// Callback, which will be called each time after successful reconnect.
onReconnect func()
// internal channels
send chan types.RPCRequest // user requests
backlog chan types.RPCRequest // stores a single user request received during a conn failure
reconnectAfter chan error // reconnect requests
readRoutineQuit chan struct{} // a way for readRoutine to close writeRoutine
wg sync.WaitGroup
mtx sync.RWMutex
sentLastPingAt time.Time
reconnecting bool
// Maximum reconnect attempts (0 or greater; default: 25).
maxReconnectAttempts int
// Time allowed to write a message to the server. 0 means block until operation succeeds.
writeWait time.Duration
// Time allowed to read the next message from the server. 0 means block until operation succeeds.
readWait time.Duration
// Send pings to server with this period. Must be less than readWait. If 0, no pings will be sent.
pingPeriod time.Duration
}
// NewWSClient returns a new client. See the commentary on the func(*WSClient)
// functions for a detailed description of how to configure ping period and
// pong wait time. The endpoint argument must begin with a `/`.
func NewWSClient(remoteAddr, endpoint string, options ...func(*WSClient)) *WSClient {
addr, dialer := makeHTTPDialer(remoteAddr)
c := &WSClient{
Address: addr,
Dialer: dialer,
Endpoint: endpoint,
PingPongLatencyTimer: metrics.NewTimer(),
maxReconnectAttempts: defaultMaxReconnectAttempts,
readWait: defaultReadWait,
writeWait: defaultWriteWait,
pingPeriod: defaultPingPeriod,
}
c.BaseService = *cmn.NewBaseService(nil, "WSClient", c)
for _, option := range options {
option(c)
}
return c
}
// MaxReconnectAttempts sets the maximum number of reconnect attempts before returning an error.
// It should only be used in the constructor and is not Goroutine-safe.
func MaxReconnectAttempts(max int) func(*WSClient) {
return func(c *WSClient) {
c.maxReconnectAttempts = max
}
}
// ReadWait sets the amount of time to wait before a websocket read times out.
// It should only be used in the constructor and is not Goroutine-safe.
func ReadWait(readWait time.Duration) func(*WSClient) {
return func(c *WSClient) {
c.readWait = readWait
}
}
// WriteWait sets the amount of time to wait before a websocket write times out.
// It should only be used in the constructor and is not Goroutine-safe.
func WriteWait(writeWait time.Duration) func(*WSClient) {
return func(c *WSClient) {
c.writeWait = writeWait
}
}
// PingPeriod sets the duration for sending websocket pings.
// It should only be used in the constructor - not Goroutine-safe.
func PingPeriod(pingPeriod time.Duration) func(*WSClient) {
return func(c *WSClient) {
c.pingPeriod = pingPeriod
}
}
// OnReconnect sets the callback, which will be called every time after
// successful reconnect.
func OnReconnect(cb func()) func(*WSClient) {
return func(c *WSClient) {
c.onReconnect = cb
}
}
// String returns WS client full address.
func (c *WSClient) String() string {
return fmt.Sprintf("%s (%s)", c.Address, c.Endpoint)
}
// OnStart implements cmn.Service by dialing a server and creating read and
// write routines.
func (c *WSClient) OnStart() error {
err := c.dial()
if err != nil {
return err
}
c.ResponsesCh = make(chan types.RPCResponse)
c.send = make(chan types.RPCRequest)
// 1 additional error may come from the read/write
// goroutine depending on which failed first.
c.reconnectAfter = make(chan error, 1)
// capacity for 1 request. a user won't be able to send more because the send
// channel is unbuffered.
c.backlog = make(chan types.RPCRequest, 1)
c.startReadWriteRoutines()
go c.reconnectRoutine()
return nil
}
// OnStop implements cmn.Service.
func (c *WSClient) OnStop() {}
// Stop overrides cmn.Service#Stop. There is no other way to wait until Quit
// channel is closed.
func (c *WSClient) Stop() bool {
success := c.BaseService.Stop()
// only close user-facing channels when we can't write to them
c.wg.Wait()
close(c.ResponsesCh)
return success
}
// IsReconnecting returns true if the client is reconnecting right now.
func (c *WSClient) IsReconnecting() bool {
c.mtx.RLock()
defer c.mtx.RUnlock()
return c.reconnecting
}
// IsActive returns true if the client is running and not reconnecting.
func (c *WSClient) IsActive() bool {
return c.IsRunning() && !c.IsReconnecting()
}
// Send the given RPC request to the server. Results will be available on
// ResponsesCh, errors, if any, on ErrorsCh. Will block until send succeeds or
// ctx.Done is closed.
func (c *WSClient) Send(ctx context.Context, request types.RPCRequest) error {
select {
case c.send <- request:
c.Logger.Info("sent a request", "req", request)
return nil
case <-ctx.Done():
return ctx.Err()
}
}
// Call the given method. See Send description.
func (c *WSClient) Call(ctx context.Context, method string, params map[string]interface{}) error {
request, err := types.MapToRequest("ws-client", method, params)
if err != nil {
return err
}
return c.Send(ctx, request)
}
// CallWithArrayParams the given method with params in a form of array. See
// Send description.
func (c *WSClient) CallWithArrayParams(ctx context.Context, method string, params []interface{}) error {
request, err := types.ArrayToRequest("ws-client", method, params)
if err != nil {
return err
}
return c.Send(ctx, request)
}
///////////////////////////////////////////////////////////////////////////////
// Private methods
func (c *WSClient) dial() error {
dialer := &websocket.Dialer{
NetDial: c.Dialer,
Proxy: http.ProxyFromEnvironment,
}
rHeader := http.Header{}
conn, _, err := dialer.Dial("ws://"+c.Address+c.Endpoint, rHeader)
if err != nil {
return err
}
c.conn = conn
return nil
}
// reconnect tries to redial up to maxReconnectAttempts with exponential
// backoff.
func (c *WSClient) reconnect() error {
attempt := 0
c.mtx.Lock()
c.reconnecting = true
c.mtx.Unlock()
defer func() {
c.mtx.Lock()
c.reconnecting = false
c.mtx.Unlock()
}()
// 1s == (1e9 ns) == (1 Billion ns)
billionNs := float64(time.Second.Nanoseconds())
for {
jitterSeconds := time.Duration(rand.Float64() * billionNs)
backoffDuration := jitterSeconds + ((1 << uint(attempt)) * time.Second)
c.Logger.Info("reconnecting", "attempt", attempt+1, "backoff_duration", backoffDuration)
time.Sleep(backoffDuration)
err := c.dial()
if err != nil {
c.Logger.Error("failed to redial", "err", err)
} else {
c.Logger.Info("reconnected")
if c.onReconnect != nil {
go c.onReconnect()
}
return nil
}
attempt++
if attempt > c.maxReconnectAttempts {
return errors.Wrap(err, "reached maximum reconnect attempts")
}
}
}
func (c *WSClient) startReadWriteRoutines() {
c.wg.Add(2)
c.readRoutineQuit = make(chan struct{})
go c.readRoutine()
go c.writeRoutine()
}
func (c *WSClient) processBacklog() error {
select {
case request := <-c.backlog:
if c.writeWait > 0 {
c.conn.SetWriteDeadline(time.Now().Add(c.writeWait))
}
err := c.conn.WriteJSON(request)
if err != nil {
c.Logger.Error("failed to resend request", "err", err)
c.reconnectAfter <- err
// requeue request
c.backlog <- request
return err
}
c.Logger.Info("resend a request", "req", request)
default:
}
return nil
}
func (c *WSClient) reconnectRoutine() {
for {
select {
case originalError := <-c.reconnectAfter:
// wait until writeRoutine and readRoutine finish
c.wg.Wait()
err := c.reconnect()
if err != nil {
c.Logger.Error("failed to reconnect", "err", err, "original_err", originalError)
c.Stop()
return
} else {
// drain reconnectAfter
LOOP:
for {
select {
case <-c.reconnectAfter:
default:
break LOOP
}
}
err = c.processBacklog()
if err == nil {
c.startReadWriteRoutines()
}
}
case <-c.Quit:
return
}
}
}
// The client ensures that there is at most one writer to a connection by
// executing all writes from this goroutine.
func (c *WSClient) writeRoutine() {
var ticker *time.Ticker
if c.pingPeriod > 0 {
// ticker with a predefined period
ticker = time.NewTicker(c.pingPeriod)
} else {
// ticker that never fires
ticker = &time.Ticker{C: make(<-chan time.Time)}
}
defer func() {
ticker.Stop()
c.conn.Close()
c.wg.Done()
}()
for {
select {
case request := <-c.send:
if c.writeWait > 0 {
c.conn.SetWriteDeadline(time.Now().Add(c.writeWait))
}
err := c.conn.WriteJSON(request)
if err != nil {
c.Logger.Error("failed to send request", "err", err)
c.reconnectAfter <- err
// add request to the backlog, so we don't lose it
c.backlog <- request
return
}
case <-ticker.C:
if c.writeWait > 0 {
c.conn.SetWriteDeadline(time.Now().Add(c.writeWait))
}
err := c.conn.WriteMessage(websocket.PingMessage, []byte{})
if err != nil {
c.Logger.Error("failed to write ping", "err", err)
c.reconnectAfter <- err
return
}
c.mtx.Lock()
c.sentLastPingAt = time.Now()
c.mtx.Unlock()
c.Logger.Debug("sent ping")
case <-c.readRoutineQuit:
return
case <-c.Quit:
c.conn.WriteMessage(websocket.CloseMessage, websocket.FormatCloseMessage(websocket.CloseNormalClosure, ""))
return
}
}
}
// The client ensures that there is at most one reader to a connection by
// executing all reads from this goroutine.
func (c *WSClient) readRoutine() {
defer func() {
c.conn.Close()
c.wg.Done()
}()
c.conn.SetPongHandler(func(string) error {
// gather latency stats
c.mtx.RLock()
t := c.sentLastPingAt
c.mtx.RUnlock()
c.PingPongLatencyTimer.UpdateSince(t)
c.Logger.Debug("got pong")
return nil
})
for {
// reset deadline for every message type (control or data)
if c.readWait > 0 {
c.conn.SetReadDeadline(time.Now().Add(c.readWait))
}
_, data, err := c.conn.ReadMessage()
if err != nil {
if !websocket.IsUnexpectedCloseError(err, websocket.CloseNormalClosure) {
return
}
c.Logger.Error("failed to read response", "err", err)
close(c.readRoutineQuit)
c.reconnectAfter <- err
return
}
var response types.RPCResponse
err = json.Unmarshal(data, &response)
if err != nil {
c.Logger.Error("failed to parse response", "err", err, "data", string(data))
continue
}
c.Logger.Info("got response", "resp", response.Result)
// Combine a non-blocking read on BaseService.Quit with a non-blocking write on ResponsesCh to avoid blocking
// c.wg.Wait() in c.Stop(). Note we rely on Quit being closed so that it sends unlimited Quit signals to stop
// both readRoutine and writeRoutine
select {
case <-c.Quit:
case c.ResponsesCh <- response:
}
}
}
///////////////////////////////////////////////////////////////////////////////
// Predefined methods
// Subscribe to a query. Note the server must have a "subscribe" route
// defined.
func (c *WSClient) Subscribe(ctx context.Context, query string) error {
params := map[string]interface{}{"query": query}
return c.Call(ctx, "subscribe", params)
}
// Unsubscribe from a query. Note the server must have a "unsubscribe" route
// defined.
func (c *WSClient) Unsubscribe(ctx context.Context, query string) error {
params := map[string]interface{}{"query": query}
return c.Call(ctx, "unsubscribe", params)
}
// UnsubscribeAll from all. Note the server must have a "unsubscribe_all" route
// defined.
func (c *WSClient) UnsubscribeAll(ctx context.Context) error {
params := map[string]interface{}{}
return c.Call(ctx, "unsubscribe_all", params)
}