Added websocket hub for handling websocket clients, with a wsserve command

which runs a plain TCP server for receiving websocket hub clients, with
server config, CSRF and rate limit middleware.

- internal/websocket: Hub pushing messages to WebSocket clients (ServeHTTP,
  example.com/ws upgrade with optional CSRF check) and plain TCP clients (ServeTCP,
  newline-delimited JSON envelopes); ping/pong keepalive, per-client message
  rate limit, slow consumer handling, graceful Shutdown, debug logging of
  messages sent and received
- internal/websocket/tcpserver: standalone TCP server for a hub, with accept
  back-off, a per-IP connection limit (server.websocket.connection_limit) and
  graceful shutdown
- internal/cli/wsserve: `wsserve` command running the TCP server, with
  --listen/--host/--port overrides
- internal/middleware: keyed token bucket rate limiter, real client IP from
  trusted proxies
- internal/httpx: RFC 9457 problem responses, request IDs in context
- internal/config: server section (host, port, shutdown timeout, rate limiter
  idle TTL, websocket); util.ByteSize settings (e.g. "32KiB"); lists of
  sections as single keys; empty lists stored as nil; isExplicitlySet for
  printLoadedConfig; JobSearchConfigSource renamed to JSCSourcesSource;
  TestSections checks every section's methods
- internal/util: ByteSize, parsed and formatted with go-humanize
This commit is contained in:
bwilliams committed 2026-10-14 15:55:25 +01:00
1 parent 803a8929e6
commit c9f42b62ad
42 files changed
+3619 -57

No files matched your search

+182
View File
@@ -0,0 +1,182 @@
// Package tcpserver runs a TCP server whose clients join a websocket.Hub
// directly, without an HTTP upgrade: each connection carries newline-delimited
// JSON envelopes (see websocket.Hub.ServeTCP). It can serve a hub on its own,
// or provide a hub's clients in tests.
package tcpserver
import (
"context"
"encoding/json"
"errors"
"fmt"
"log/slog"
"net"
"net/netip"
"strconv"
"time"
"git.avsdev.uk/Go/job-search/internal/config"
"git.avsdev.uk/Go/job-search/internal/middleware"
"git.avsdev.uk/Go/job-search/internal/websocket"
)
// rejectTimeout caps how long telling a rate limited client why it is being
// turned away may take.
const rejectTimeout = time.Second
// Server listens on the configured host and port and runs each connection as
// a client of its hub. New connections from each client IP are limited by
// websocket.connection_limit.
type Server struct {
conf *config.JSCServer
log *slog.Logger
hub *websocket.Hub
}
// New builds a server for hub's plain TCP clients. Run starts it; it shuts
// hub down when it stops, so hub should not be shared with another server.
// Push to hub to reach the server's clients.
func New(conf *config.JSCServer, hub *websocket.Hub, log *slog.Logger) *Server {
return &Server{
conf: conf,
log: log,
hub: hub,
}
}
// Run listens on the configured host and port and serves plain TCP clients
// until ctx is cancelled, then shuts down gracefully (see serve).
func (s *Server) Run(ctx context.Context) error {
addr := net.JoinHostPort(s.conf.Host, strconv.Itoa(s.conf.Port))
listen, err := net.Listen("tcp", addr)
if err != nil {
return fmt.Errorf("listen %s: %w", addr, err)
}
return s.serve(ctx, listen)
}
// serve accepts connections on listen until ctx is cancelled or listen fails,
// then shuts down: it stops accepting, then closes the hub's clients
// gracefully, waiting up to the shutdown timeout (for ever if it is 0 or
// less). It returns nil after a shutdown caused by ctx, or the error that
// stopped it accepting, joined with any error from the hub's shutdown.
func (s *Server) serve(ctx context.Context, listen net.Listener) error {
// Clients outlive ctx until the hub has closed them, so the shutdown can
// tell them why; cancelling ctx itself would drop them without a word.
connCtx, cancelConns := context.WithCancel(context.WithoutCancel(ctx))
defer cancelConns()
stop := context.AfterFunc(ctx, func() {
listen.Close() //nolint:errcheck,gosec // Accept returns net.ErrClosed, which ends the loop
})
defer stop()
s.log.Info("listening", "addr", listen.Addr().String())
acceptErr := s.accept(ctx, connCtx, listen)
if ctx.Err() != nil {
acceptErr = nil // cancelling ctx is how the server is stopped
}
listen.Close() //nolint:errcheck,gosec // closed already when ctx was cancelled
s.log.Info("shutting down")
shutdownCtx := context.Background()
if s.conf.Timeouts.ShutdownTimeout > 0 {
var cancel context.CancelFunc
shutdownCtx, cancel = context.WithTimeout(
shutdownCtx, s.conf.Timeouts.ShutdownTimeout,
)
defer cancel()
}
errs := []error{acceptErr}
if err := s.hub.Shutdown(shutdownCtx); err != nil {
errs = append(errs, fmt.Errorf("websocket shutdown: %w", err))
}
return errors.Join(errs...)
}
// accept runs each connection on listen as a hub client, with connCtx, until
// ctx is cancelled (it returns ctx's error) or listen is closed some other way
// (it returns the error). Other accept errors are retried after a back off.
// Connections over the connection limit are turned away (see reject).
func (s *Server) accept(
ctx, connCtx context.Context, listen net.Listener,
) error {
limiter := s.connectionLimiter(ctx)
var delay time.Duration
for {
newConn, err := listen.Accept()
if err != nil {
if ctx.Err() != nil {
return ctx.Err()
}
if errors.Is(err, net.ErrClosed) {
return fmt.Errorf("accept: %w", err)
}
// Back off and retry e.g. when out of file descriptors.
delay = min(max(2*delay, 5*time.Millisecond), time.Second)
s.log.Warn("tcp accept failed", "error", err, "retry_in", delay)
select {
case <-time.After(delay):
continue
case <-ctx.Done():
return ctx.Err()
}
}
delay = 0
if limiter != nil {
ip := remoteIP(newConn)
if allowed, _ := limiter.Allow(ip); !allowed {
s.log.Warn("tcp connection rate limited", "ip", ip)
go reject(newConn, "rate limited")
continue
}
}
go s.hub.ServeTCP(connCtx, newConn)
}
}
// connectionLimiter returns a limiter of new connections per client IP for
// websocket.connection_limit, or nil if it is off. Its idle entries are
// evicted after rate_limits.idle_ttl until ctx is cancelled.
func (s *Server) connectionLimiter(ctx context.Context) middleware.Limiter {
rule := s.conf.WebSocket.ConnectionLimit
if !rule.Enabled || rule.PerSecond <= 0 {
return nil
}
return middleware.NewKeyedLimiter(
ctx, float64(rule.PerSecond), max(rule.Burst, 1), s.conf.RateLimits.IdleTTL,
)
}
// reject tells a client why it is being turned away, with an error envelope
// as the hub would send, then closes its connection.
func reject(conn net.Conn, reason string) {
defer conn.Close() //nolint:errcheck // turned away; nothing to do if closing fails
message, err := json.Marshal(websocket.Envelope{
Type: websocket.TypeError,
Error: reason,
})
if err != nil {
return
}
conn.SetWriteDeadline(time.Now().Add(rejectTimeout)) //nolint:errcheck,gosec // best effort
conn.Write(append(message, '\n')) //nolint:errcheck,gosec // best effort
}
// remoteIP returns the address of the connection's peer, or "" if it has no
// IP address.
func remoteIP(conn net.Conn) string {
addrPort, err := netip.ParseAddrPort(conn.RemoteAddr().String())
if err != nil {
return ""
}
return addrPort.Addr().Unmap().String()
}
+263
View File
@@ -0,0 +1,263 @@
package tcpserver
import (
"bufio"
"context"
"encoding/json"
"errors"
"io"
"log/slog"
"net"
"strings"
"syscall"
"testing"
"time"
"git.avsdev.uk/Go/job-search/internal/config"
"git.avsdev.uk/Go/job-search/internal/websocket"
)
var discardLog = slog.New(slog.NewTextHandler(io.Discard, nil))
// testConf is a server configuration for tests, changed by mutate if it is
// not nil.
func testConf(mutate func(*config.JSCServer)) *config.JSCServer {
c := &config.JSCServer{
Host: "127.0.0.1",
Port: 0,
WebSocket: config.JSCServerWebSocket{WriteTimeout: time.Second},
Timeouts: config.JSCServerTimeouts{ShutdownTimeout: time.Second},
}
if mutate != nil {
mutate(c)
}
return c
}
// newServer returns a server for conf with its own hub.
func newServer(conf *config.JSCServer) (*Server, *websocket.Hub) {
h := websocket.NewHub(conf.WebSocket, discardLog)
return New(conf, h, discardLog), h
}
// startServer serves on a local port, returning the server's hub, its
// address, a function to cancel it and a channel with serve's result.
func startServer(
t *testing.T, conf *config.JSCServer,
) (*websocket.Hub, string, context.CancelFunc, chan error) {
t.Helper()
s, h := newServer(conf)
ln, err := net.Listen("tcp", "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
ctx, cancel := context.WithCancel(context.Background())
t.Cleanup(cancel)
done := make(chan error, 1)
go func() { done <- s.serve(ctx, ln) }()
return h, ln.Addr().String(), cancel, done
}
// tcpClient is a plain TCP test client reading newline-delimited envelopes.
type tcpClient struct {
t *testing.T
conn net.Conn
r *bufio.Reader
}
func dial(t *testing.T, addr string) *tcpClient {
t.Helper()
conn, err := net.Dial("tcp", addr)
if err != nil {
t.Fatal(err)
}
t.Cleanup(func() { conn.Close() })
return &tcpClient{t: t, conn: conn, r: bufio.NewReader(conn)}
}
// read returns the next envelope, failing the test after a second.
func (c *tcpClient) read() websocket.Envelope {
c.t.Helper()
c.conn.SetReadDeadline(time.Now().Add(time.Second))
line, err := c.r.ReadString('\n')
if err != nil {
c.t.Fatalf("read: %v (got %q)", err, line)
}
var env websocket.Envelope
if err := json.Unmarshal([]byte(line), &env); err != nil {
c.t.Fatalf("bad envelope %q: %v", line, err)
}
return env
}
// waitClosed checks the server closes the connection, skipping any lines
// sent first. A reset counts as closed.
func (c *tcpClient) waitClosed() {
c.t.Helper()
c.conn.SetReadDeadline(time.Now().Add(time.Second))
if _, err := io.ReadAll(c.r); err != nil && !errors.Is(err, syscall.ECONNRESET) {
c.t.Fatalf("connection not closed: %v", err)
}
}
// Clients get pushed messages; cancelling the server's context tells them it
// is shutting down, closes them, stops accepting and returns nil.
func TestServerShutdown(t *testing.T) {
h, addr, cancel, done := startServer(t, testConf(nil))
c := dial(t, addr)
if e := c.read(); e.Type != websocket.TypeWelcome {
t.Fatalf("want welcome, got %+v", e)
}
for h.Count() != 1 {
time.Sleep(time.Millisecond)
}
h.Push("item.created", map[string]string{"id": "x"})
if e := c.read(); e.Type != "item.created" {
t.Fatalf("want item.created, got %+v", e)
}
cancel()
if e := c.read(); e.Type != websocket.TypeError || e.Error != "server shutting down" {
t.Fatalf("want shutdown error, got %+v", e)
}
c.waitClosed()
select {
case err := <-done:
if err != nil {
t.Fatalf("serve = %v", err)
}
case <-time.After(2 * time.Second):
t.Fatal("serve did not return")
}
if conn, err := net.Dial("tcp", addr); err == nil {
conn.Close()
t.Error("listener still accepting")
}
}
// Over websocket.connection_limit, a client is told it is rate limited and
// closed; other clients are unaffected.
func TestServerConnectionLimit(t *testing.T) {
conf := testConf(func(c *config.JSCServer) {
c.WebSocket.ConnectionLimit = config.JSCRateLimitRule{
Enabled: true, PerSecond: 1, Burst: 1,
}
c.RateLimits.IdleTTL = time.Minute
})
_, addr, cancel, _ := startServer(t, conf)
defer cancel()
first := dial(t, addr)
if e := first.read(); e.Type != websocket.TypeWelcome {
t.Fatalf("first: want welcome, got %+v", e)
}
second := dial(t, addr)
if e := second.read(); e.Type != websocket.TypeError || e.Error != "rate limited" {
t.Fatalf("second: want rate limited, got %+v", e)
}
second.waitClosed()
first.conn.Write([]byte(`{"type":"ping","id":"1"}` + "\n"))
if e := first.read(); e.Type != websocket.TypePong {
t.Fatalf("first: want pong, got %+v", e)
}
}
// With the connection limit off (or no rate), there is no limiter.
func TestServerConnectionLimitOff(t *testing.T) {
for _, rule := range []config.JSCRateLimitRule{
{Enabled: false, PerSecond: 1, Burst: 1},
{Enabled: true, PerSecond: 0, Burst: 1},
} {
s, _ := newServer(testConf(func(c *config.JSCServer) {
c.WebSocket.ConnectionLimit = rule
}))
if l := s.connectionLimiter(context.Background()); l != nil {
t.Errorf("rule %+v: got a limiter", rule)
}
}
}
func TestServerRun(t *testing.T) {
s, _ := newServer(testConf(nil))
ctx, cancel := context.WithCancel(context.Background())
time.AfterFunc(20*time.Millisecond, cancel)
if err := s.Run(ctx); err != nil {
t.Fatalf("Run = %v", err)
}
}
func TestServerRunListenError(t *testing.T) {
s, _ := newServer(testConf(func(c *config.JSCServer) { c.Port = -1 }))
if err := s.Run(context.Background()); err == nil || !strings.Contains(err.Error(), "listen") {
t.Fatalf("Run = %v; want a listen error", err)
}
}
// flakyListener fails Accept with a temporary error a number of times, then
// as if closed.
type flakyListener struct {
net.Listener
fails int
}
func (l *flakyListener) Accept() (net.Conn, error) {
if l.fails > 0 {
l.fails--
return nil, errors.New("too many open files")
}
return nil, net.ErrClosed
}
// Accept errors other than a closed listener are retried after a back off.
func TestServerAcceptRetries(t *testing.T) {
s, _ := newServer(testConf(nil))
ln, err := net.Listen("tcp", "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer ln.Close()
fl := &flakyListener{Listener: ln, fails: 3}
err = s.accept(context.Background(), context.Background(), fl)
if !errors.Is(err, net.ErrClosed) || fl.fails != 0 {
t.Fatalf("accept = %v with %d failures left; want net.ErrClosed", err, fl.fails)
}
}
// Cancelling ctx stops accept while it is backing off.
func TestServerAcceptCancelledInBackOff(t *testing.T) {
s, _ := newServer(testConf(nil))
ln, err := net.Listen("tcp", "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer ln.Close()
ctx, cancel := context.WithCancel(context.Background())
fl := &flakyListener{Listener: ln, fails: 1 << 30}
time.AfterFunc(20*time.Millisecond, cancel)
if err := s.accept(ctx, ctx, fl); !errors.Is(err, context.Canceled) {
t.Fatalf("accept = %v; want context.Canceled", err)
}
}
// A listener closed from elsewhere is an error, after a graceful shutdown.
func TestServerListenerClosed(t *testing.T) {
s, _ := newServer(testConf(nil))
ln, err := net.Listen("tcp", "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
ln.Close()
if err := s.serve(context.Background(), ln); !errors.Is(err, net.ErrClosed) {
t.Fatalf("serve = %v; want net.ErrClosed", err)
}
}
func TestRemoteIP(t *testing.T) {
a, b := net.Pipe()
defer a.Close()
defer b.Close()
if ip := remoteIP(a); ip != "" {
t.Errorf("remoteIP(pipe) = %q; want empty", ip)
}
}