iop/apps/node/internal/bootstrap/module_test.go
toki 0ffcb88db0 feat: provider-resource-admission-ownership alignment
- Archive provider-resource-admission-ownership milestone/SDD
- Align contract: CP-edge wire, runtime refresh, node runtime, OpenAI surface
- Update roadmap: phase state, priority queue
- Update specs: control-plane ops, OpenAI surface, edge execution, provider pool refresh
- Add node runtime supervisor bootstrapping and unit tests
- Fix control-plane edge registry handler and http_views
- Fix edge model queue admission and long context queue tests
2026-07-22 20:45:04 +09:00

944 lines
32 KiB
Go

package bootstrap_test
import (
"context"
"errors"
"fmt"
"net"
"os"
"sync"
"sync/atomic"
"testing"
"time"
toki "git.toki-labs.com/toki/proto-socket/go"
"git.toki-labs.com/toki/proto-socket/go/packets"
"go.uber.org/fx"
"go.uber.org/zap"
"google.golang.org/protobuf/proto"
"iop/apps/node/internal/bootstrap"
"iop/apps/node/internal/transport"
"iop/packages/go/config"
iop "iop/proto/gen/iop"
)
func bootstrapEdgeParserMap() toki.ParserMap {
return toki.ParserMap{
toki.TypeNameOf(&iop.RegisterRequest{}): func(b []byte) (proto.Message, error) {
m := &iop.RegisterRequest{}
return m, proto.Unmarshal(b, m)
},
toki.TypeNameOf(&iop.NodeReadyRequest{}): func(b []byte) (proto.Message, error) {
m := &iop.NodeReadyRequest{}
return m, proto.Unmarshal(b, m)
},
}
}
// bindEdgeReadyResponder answers the node's post-handler NodeReadyRequest with a
// positive ack, so connectRuntime's SignalReady completes and the session is
// treated as fully established. Mock edges that accept registration must install
// it or the node's ready handshake times out and reconnects. When readyCh is
// non-nil it fires only after the ready response has been written to the socket.
// This lets a test safely close the edge side after a fully established session,
// instead of racing the in-flight ready handshake.
func bindEdgeReadyResponder(client *toki.TcpClient, readyCh chan<- struct{}) {
client.Communicator.AddRequestListener(
toki.TypeNameOf(&iop.NodeReadyRequest{}),
func(message proto.Message, requestNonce int32) {
if _, ok := message.(*iop.NodeReadyRequest); !ok {
return
}
data, err := proto.Marshal(&iop.NodeReadyResponse{Ready: true})
if err != nil {
return
}
if err := client.QueuePacket(&packets.PacketBase{
TypeName: toki.TypeNameOf(&iop.NodeReadyResponse{}),
ResponseNonce: requestNonce,
Data: data,
}); err != nil {
return
}
if readyCh != nil {
select {
case readyCh <- struct{}{}:
default:
}
}
},
)
}
func freeAddrForBootstrap(t *testing.T) (host string, port int, addr string) {
t.Helper()
l, err := net.Listen("tcp", "127.0.0.1:0")
if err != nil {
t.Fatalf("listen: %v", err)
}
addr = l.Addr().String()
l.Close()
h, portStr, _ := net.SplitHostPort(addr)
fmt.Sscanf(portStr, "%d", &port)
return h, port, addr
}
func useTempCwd(t *testing.T) {
t.Helper()
previous, err := os.Getwd()
if err != nil {
t.Fatalf("get cwd: %v", err)
}
if err := os.Chdir(t.TempDir()); err != nil {
t.Fatalf("chdir temp: %v", err)
}
t.Cleanup(func() {
if err := os.Chdir(previous); err != nil {
t.Fatalf("restore cwd: %v", err)
}
})
}
// startAcceptingEdge starts a mock edge bound to host:port that accepts every
// RegisterRequest. The returned channels fire once per registration and
// disconnect, respectively.
func startAcceptingEdge(t *testing.T, ctx context.Context, host string, port int, nodeID string) (chan struct{}, chan struct{}) {
t.Helper()
regCh := make(chan struct{}, 16)
disconnectCh := make(chan struct{}, 16)
server := toki.NewTcpServer(host, port, func(conn net.Conn) *toki.TcpClient {
client := toki.NewTcpClient(conn, 30, 10, bootstrapEdgeParserMap())
client.AddDisconnectListener(func(_ *toki.TcpClient) {
disconnectCh <- struct{}{}
})
toki.AddRequestListenerTyped[*iop.RegisterRequest, *iop.RegisterResponse](
&client.Communicator,
func(_ *iop.RegisterRequest) (*iop.RegisterResponse, error) {
regCh <- struct{}{}
return &iop.RegisterResponse{
Accepted: true,
NodeId: nodeID,
Config: &iop.NodeConfigPayload{Runtime: &iop.NodeRuntimeConfig{Concurrency: 1}},
}, nil
},
)
bindEdgeReadyResponder(client, nil)
return client
})
if err := server.Start(ctx); err != nil {
t.Fatalf("start accepting edge: %v", err)
}
t.Cleanup(func() { server.Stop() })
return regCh, disconnectCh
}
// startRejectingEdge starts a mock edge bound to host:port that rejects every
// RegisterRequest with the given reason.
func startRejectingEdge(t *testing.T, ctx context.Context, host string, port int, reason string) {
t.Helper()
server := toki.NewTcpServer(host, port, func(conn net.Conn) *toki.TcpClient {
client := toki.NewTcpClient(conn, 30, 10, bootstrapEdgeParserMap())
toki.AddRequestListenerTyped[*iop.RegisterRequest, *iop.RegisterResponse](
&client.Communicator,
func(_ *iop.RegisterRequest) (*iop.RegisterResponse, error) {
return &iop.RegisterResponse{Accepted: false, Reason: reason}, nil
},
)
return client
})
if err := server.Start(ctx); err != nil {
t.Fatalf("start rejecting edge: %v", err)
}
t.Cleanup(func() { server.Stop() })
}
// TestSupervisorConnectsWhenEdgeStartsAfterNode verifies the actual DialEdge
// connection-refused path: Node starts first against a closed port, keeps
// retrying in the same process, and establishes exactly one session after Edge
// begins listening on that same port. SDD S16.
func TestSupervisorConnectsWhenEdgeStartsAfterNode(t *testing.T) {
useTempCwd(t)
host, port, addr := freeAddrForBootstrap(t)
const refusedAttempts = 3
var dialCount int32
retryReady := make(chan struct{}, refusedAttempts)
releaseRetry := make(chan struct{})
observedDialer := bootstrap.DialFunc(func(ctx context.Context, edgeAddr, token string, logger *zap.Logger) (*transport.RegisterResult, error) {
atomic.AddInt32(&dialCount, 1)
return transport.DialEdge(ctx, edgeAddr, token, logger)
})
fakeSleeper := func(ctx context.Context, _ time.Duration) {
retryReady <- struct{}{}
if atomic.LoadInt32(&dialCount) >= refusedAttempts {
select {
case <-releaseRetry:
case <-ctx.Done():
}
}
}
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 0, IntervalSec: 1},
Logging: config.LoggingConf{Level: "error"},
}
app := fx.New(bootstrap.Module(cfg, bootstrap.WithDialer(observedDialer), bootstrap.WithSleeper(fakeSleeper)), fx.NopLogger)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
defer func() { _ = app.Stop(context.Background()) }()
// Edge is not listening yet. Each completed sleep entry follows a real
// DialEdge connection-refused result. Hold the third retry until the listener
// is live so the next actual dial succeeds deterministically.
for attempt := 1; attempt <= refusedAttempts; attempt++ {
select {
case <-retryReady:
case <-time.After(5 * time.Second):
t.Fatalf("timeout waiting for refused DialEdge attempt %d", attempt)
}
}
if count := atomic.LoadInt32(&dialCount); count != refusedAttempts {
t.Fatalf("expected %d real refused dials before Edge startup, got %d", refusedAttempts, count)
}
serverCtx, serverCancel := context.WithTimeout(context.Background(), 10*time.Second)
t.Cleanup(serverCancel)
regCh, _ := startAcceptingEdge(t, serverCtx, host, port, "delayed-node")
close(releaseRetry)
select {
case <-regCh:
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for delayed initial registration")
}
if count := atomic.LoadInt32(&dialCount); count != refusedAttempts+1 {
t.Fatalf("expected %d dials (3 real refusals + 1 success), got %d", refusedAttempts+1, count)
}
// Exactly one session must be established.
select {
case <-regCh:
t.Fatal("unexpected second registration; supervisor should hold one session")
case <-time.After(200 * time.Millisecond):
}
}
// TestSupervisorUnlimitedAttemptsPassesTen verifies that an explicit
// max_attempts=0 (unlimited) keeps retrying past the former hard-coded 10-attempt
// cap and eventually connects. SDD S16/S17.
func TestSupervisorUnlimitedAttemptsPassesTen(t *testing.T) {
useTempCwd(t)
host, port, addr := freeAddrForBootstrap(t)
serverCtx, serverCancel := context.WithTimeout(context.Background(), 10*time.Second)
t.Cleanup(serverCancel)
regCh, _ := startAcceptingEdge(t, serverCtx, host, port, "unlimited-node")
const failFirst = 15 // exceeds the old hard-coded 10-attempt cap
var dialCount int32
fakeDialer := bootstrap.DialFunc(func(ctx context.Context, edgeAddr, token string, logger *zap.Logger) (*transport.RegisterResult, error) {
if atomic.AddInt32(&dialCount, 1) <= failFirst {
return nil, errors.New("edge unavailable")
}
return transport.DialEdge(ctx, edgeAddr, token, logger)
})
fakeSleeper := func(context.Context, time.Duration) {}
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 0, IntervalSec: 1},
Logging: config.LoggingConf{Level: "error"},
}
app := fx.New(bootstrap.Module(cfg, bootstrap.WithDialer(fakeDialer), bootstrap.WithSleeper(fakeSleeper)), fx.NopLogger)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
defer func() { _ = app.Stop(context.Background()) }()
select {
case <-regCh:
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for registration after >10 unlimited attempts")
}
count := atomic.LoadInt32(&dialCount)
if count != failFirst+1 {
t.Fatalf("expected %d dials, got %d", failFirst+1, count)
}
if count <= 10 {
t.Fatalf("unlimited policy must retry past 10 attempts, got %d", count)
}
}
// TestSupervisorFiniteExhaustionExitsOne verifies that a finite reconnect policy
// makes exactly max_attempts initial-connect attempts and then terminates the
// node with exit code 1. SDD S17.
func TestSupervisorFiniteExhaustionExitsOne(t *testing.T) {
useTempCwd(t)
_, _, addr := freeAddrForBootstrap(t) // no server listening
var dialCount int32
fakeDialer := bootstrap.DialFunc(func(_ context.Context, _, _ string, _ *zap.Logger) (*transport.RegisterResult, error) {
atomic.AddInt32(&dialCount, 1)
return nil, errors.New("edge unavailable")
})
fakeSleeper := func(context.Context, time.Duration) {}
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 3, IntervalSec: 0},
Logging: config.LoggingConf{Level: "error"},
}
app := fx.New(bootstrap.Module(cfg, bootstrap.WithDialer(fakeDialer), bootstrap.WithSleeper(fakeSleeper)), fx.NopLogger)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
defer func() { _ = app.Stop(context.Background()) }()
select {
case sig := <-app.Wait():
if sig.ExitCode != 1 {
t.Fatalf("expected exit code 1, got %d", sig.ExitCode)
}
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for finite exhaustion shutdown")
}
if count := atomic.LoadInt32(&dialCount); count != 3 {
t.Fatalf("expected exactly 3 initial-connect attempts, got %d", count)
}
}
// TestSupervisorFatalFailureDoesNotRetry verifies that an authoritative
// registration rejection is non-retryable: the supervisor exits with code 1
// after a single attempt without any retry. SDD S17.
func TestSupervisorFatalFailureDoesNotRetry(t *testing.T) {
useTempCwd(t)
host, port, addr := freeAddrForBootstrap(t)
serverCtx, serverCancel := context.WithTimeout(context.Background(), 10*time.Second)
t.Cleanup(serverCancel)
startRejectingEdge(t, serverCtx, host, port, "bad-token")
var dialCount int32
fakeDialer := bootstrap.DialFunc(func(ctx context.Context, edgeAddr, token string, logger *zap.Logger) (*transport.RegisterResult, error) {
atomic.AddInt32(&dialCount, 1)
return transport.DialEdge(ctx, edgeAddr, token, logger)
})
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 5, IntervalSec: 0},
Logging: config.LoggingConf{Level: "error"},
}
app := fx.New(bootstrap.Module(cfg, bootstrap.WithDialer(fakeDialer)), fx.NopLogger)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
defer func() { _ = app.Stop(context.Background()) }()
select {
case sig := <-app.Wait():
if sig.ExitCode != 1 {
t.Fatalf("expected exit code 1 on fatal failure, got %d", sig.ExitCode)
}
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for fatal shutdown")
}
if count := atomic.LoadInt32(&dialCount); count != 1 {
t.Fatalf("fatal failure must not retry: expected 1 dial, got %d", count)
}
}
// TestSupervisorShutdownCancelsRetry verifies that a local shutdown cancels a
// pending retry sleep, ends cleanly with no error, and runs no additional
// attempt. SDD S17.
func TestSupervisorShutdownCancelsRetry(t *testing.T) {
useTempCwd(t)
_, _, addr := freeAddrForBootstrap(t) // no server listening
var dialCount int32
fakeDialer := bootstrap.DialFunc(func(_ context.Context, _, _ string, _ *zap.Logger) (*transport.RegisterResult, error) {
atomic.AddInt32(&dialCount, 1)
return nil, errors.New("edge unavailable")
})
sleeping := make(chan struct{}, 1)
fakeSleeper := func(ctx context.Context, _ time.Duration) {
select {
case sleeping <- struct{}{}:
default:
}
<-ctx.Done() // block until the supervisor is cancelled by shutdown
}
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 0, IntervalSec: 1},
Logging: config.LoggingConf{Level: "error"},
}
app := fx.New(bootstrap.Module(cfg, bootstrap.WithDialer(fakeDialer), bootstrap.WithSleeper(fakeSleeper)), fx.NopLogger)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
// Wait until the supervisor has dialed once and parked in the retry sleep.
select {
case <-sleeping:
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for supervisor to enter retry sleep")
}
countAtSleep := atomic.LoadInt32(&dialCount)
if countAtSleep != 1 {
t.Fatalf("expected exactly 1 dial before parking in sleep, got %d", countAtSleep)
}
// Local shutdown must cancel the sleep and stop cleanly with no error.
stopCtx, stopCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer stopCancel()
if err := app.Stop(stopCtx); err != nil {
t.Fatalf("app stop must be clean, got: %v", err)
}
if count := atomic.LoadInt32(&dialCount); count != countAtSleep {
t.Fatalf("shutdown must not trigger extra dial: had %d, got %d", countAtSleep, count)
}
}
// TestSupervisorDoesNotStartMetricsWhenPortDisabled guards the Node default
// against treating port zero as an ephemeral wildcard listener. It waits for
// the reconnect supervisor to be active before asserting the injected starter
// was never called, then verifies normal shutdown still cancels the retry.
func TestSupervisorDoesNotStartMetricsWhenPortDisabled(t *testing.T) {
useTempCwd(t)
_, _, addr := freeAddrForBootstrap(t)
fakeDialer := bootstrap.DialFunc(func(_ context.Context, _, _ string, _ *zap.Logger) (*transport.RegisterResult, error) {
return nil, errors.New("edge unavailable")
})
sleeping := make(chan struct{}, 1)
fakeSleeper := func(ctx context.Context, _ time.Duration) {
select {
case sleeping <- struct{}{}:
default:
}
<-ctx.Done()
}
metricsStarted := make(chan struct{}, 1)
metricsStarter := func(int) error {
metricsStarted <- struct{}{}
return nil
}
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 0, IntervalSec: 1},
Logging: config.LoggingConf{Level: "error"},
Metrics: config.MetricsConf{Port: 0},
}
app := fx.New(
bootstrap.Module(
cfg,
bootstrap.WithDialer(fakeDialer),
bootstrap.WithSleeper(fakeSleeper),
bootstrap.WithMetricsStarter(metricsStarter),
),
fx.NopLogger,
)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
select {
case <-sleeping:
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for supervisor to enter retry sleep")
}
select {
case <-metricsStarted:
t.Fatal("metrics starter must not run when metrics port is disabled")
case <-time.After(100 * time.Millisecond):
}
stopCtx, stopCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer stopCancel()
if err := app.Stop(stopCtx); err != nil {
t.Fatalf("app stop must be clean, got: %v", err)
}
}
// TestSupervisorShutdownWinsInFlightDialFatal fixes the ordering between a
// local stop and a dialer that observes cancellation but returns a fatal result
// afterward. Cancellation must produce a clean stop, no exit 1, and no retry.
func TestSupervisorShutdownWinsInFlightDialFatal(t *testing.T) {
useTempCwd(t)
_, _, addr := freeAddrForBootstrap(t)
started := make(chan struct{}, 1)
var dialCount int32
dialer := bootstrap.DialFunc(func(ctx context.Context, _, _ string, logger *zap.Logger) (*transport.RegisterResult, error) {
atomic.AddInt32(&dialCount, 1)
started <- struct{}{}
<-ctx.Done()
return transport.DialEdge(context.Background(), "127.0.0.1:0", "tok", logger)
})
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 0, IntervalSec: 1},
Logging: config.LoggingConf{Level: "error"},
}
app := fx.New(bootstrap.Module(cfg, bootstrap.WithDialer(dialer)), fx.NopLogger)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
select {
case <-started:
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for in-flight dial")
}
stopCtx, stopCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer stopCancel()
if err := app.Stop(stopCtx); err != nil {
t.Fatalf("app stop must be clean, got: %v", err)
}
if count := atomic.LoadInt32(&dialCount); count != 1 {
t.Fatalf("shutdown must not retry fatal late result: got %d dials", count)
}
select {
case sig := <-app.Wait():
t.Fatalf("shutdown must not request failure exit, got %+v", sig)
case <-time.After(100 * time.Millisecond):
}
}
// TestSupervisorShutdownClosesInFlightDialSuccess verifies that a dialer which
// observes cancellation and then returns a successful registered session cannot
// install or leak that late owner. The one registered session is closed by the
// post-dial cancellation fence and no failure exit or retry is emitted.
func TestSupervisorShutdownClosesInFlightDialSuccess(t *testing.T) {
useTempCwd(t)
host, port, addr := freeAddrForBootstrap(t)
serverCtx, serverCancel := context.WithTimeout(context.Background(), 10*time.Second)
t.Cleanup(serverCancel)
regCh, disconnectCh := startAcceptingEdge(t, serverCtx, host, port, "late-success-node")
started := make(chan struct{}, 1)
var dialCount int32
dialer := bootstrap.DialFunc(func(ctx context.Context, edgeAddr, token string, logger *zap.Logger) (*transport.RegisterResult, error) {
atomic.AddInt32(&dialCount, 1)
started <- struct{}{}
<-ctx.Done()
lateCtx, lateCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer lateCancel()
return transport.DialEdge(lateCtx, edgeAddr, token, logger)
})
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 0, IntervalSec: 1},
Logging: config.LoggingConf{Level: "error"},
}
app := fx.New(bootstrap.Module(cfg, bootstrap.WithDialer(dialer)), fx.NopLogger)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
select {
case <-started:
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for in-flight dial")
}
stopCtx, stopCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer stopCancel()
stopDone := make(chan error, 1)
go func() { stopDone <- app.Stop(stopCtx) }()
select {
case <-regCh:
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for late successful registration")
}
select {
case err := <-stopDone:
if err != nil {
t.Fatalf("app stop must be clean, got: %v", err)
}
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for app stop")
}
select {
case <-disconnectCh:
case <-time.After(5 * time.Second):
t.Fatal("late successful session leaked after supervisor shutdown")
}
if count := atomic.LoadInt32(&dialCount); count != 1 {
t.Fatalf("shutdown must not retry late success: got %d dials", count)
}
select {
case <-regCh:
t.Fatal("unexpected second registration after shutdown")
case sig := <-app.Wait():
t.Fatalf("shutdown must not request failure exit, got %+v", sig)
case <-time.After(100 * time.Millisecond):
}
}
// TestReconnectSupervisorReestablishesSession verifies that after the edge-side
// client closes the connection, the supervisor reconnects and re-registers.
func TestReconnectSupervisorReestablishesSession(t *testing.T) {
useTempCwd(t)
host, port, addr := freeAddrForBootstrap(t)
serverCtx, serverCancel := context.WithTimeout(context.Background(), 10*time.Second)
t.Cleanup(serverCancel)
// regCh fires each time a RegisterRequest is handled; readyCh fires when a
// connection completes its ready handshake so the test can close it only after
// the session is fully established.
regCh := make(chan struct{}, 8)
readyCh := make(chan struct{}, 8)
acceptedCh := make(chan *toki.TcpClient, 8)
server := toki.NewTcpServer(host, port, func(conn net.Conn) *toki.TcpClient {
client := toki.NewTcpClient(conn, 30, 10, bootstrapEdgeParserMap())
toki.AddRequestListenerTyped[*iop.RegisterRequest, *iop.RegisterResponse](
&client.Communicator,
func(_ *iop.RegisterRequest) (*iop.RegisterResponse, error) {
regCh <- struct{}{}
return &iop.RegisterResponse{
Accepted: true,
NodeId: "reconnect-test-node",
Config: &iop.NodeConfigPayload{Runtime: &iop.NodeRuntimeConfig{Concurrency: 1}},
}, nil
},
)
bindEdgeReadyResponder(client, readyCh)
acceptedCh <- client
return client
})
if err := server.Start(serverCtx); err != nil {
t.Fatalf("start server: %v", err)
}
t.Cleanup(func() { server.Stop() })
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 5, IntervalSec: 0},
Logging: config.LoggingConf{Level: "error"},
}
// Wrap the real dialer so the test can wait for the initial session to be
// fully established before inducing a disconnect (OnStart is non-blocking).
established := make(chan struct{}, 1)
var establishOnce sync.Once
fakeDialer := bootstrap.DialFunc(func(ctx context.Context, edgeAddr, token string, logger *zap.Logger) (*transport.RegisterResult, error) {
res, err := transport.DialEdge(ctx, edgeAddr, token, logger)
if err == nil {
establishOnce.Do(func() { established <- struct{}{} })
}
return res, err
})
app := fx.New(bootstrap.Module(cfg, bootstrap.WithDialer(fakeDialer)), fx.NopLogger)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
defer func() { _ = app.Stop(context.Background()) }()
// Wait for first registration and full establishment.
select {
case <-regCh:
case <-time.After(3 * time.Second):
t.Fatal("timeout waiting for first registration")
}
select {
case <-established:
case <-time.After(3 * time.Second):
t.Fatal("timeout waiting for initial session establishment")
}
var firstClient *toki.TcpClient
select {
case firstClient = <-acceptedCh:
case <-time.After(time.Second):
t.Fatal("timeout waiting for first accepted client")
}
// Wait for the ready handshake before closing so the disconnect is induced on a
// fully-established session, not mid-SignalReady.
select {
case <-readyCh:
case <-time.After(3 * time.Second):
t.Fatal("timeout waiting for initial ready handshake")
}
// Force edge-side close → remote disconnect on node.
_ = firstClient.Close()
// Wait for second registration (supervisor reconnected and re-registered).
select {
case <-regCh:
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for supervisor reconnect registration")
}
}
// TestReconnectSupervisorExhaustsRetries verifies that when all reconnect
// attempts fail, the node cleans up resources and calls Shutdown(ExitCode(1)).
func TestReconnectSupervisorExhaustsRetries(t *testing.T) {
useTempCwd(t)
host, port, addr := freeAddrForBootstrap(t)
serverCtx, serverCancel := context.WithTimeout(context.Background(), 10*time.Second)
t.Cleanup(serverCancel)
acceptedCh := make(chan *toki.TcpClient, 4)
readyCh := make(chan struct{}, 4)
server := toki.NewTcpServer(host, port, func(conn net.Conn) *toki.TcpClient {
client := toki.NewTcpClient(conn, 30, 10, bootstrapEdgeParserMap())
toki.AddRequestListenerTyped[*iop.RegisterRequest, *iop.RegisterResponse](
&client.Communicator,
func(_ *iop.RegisterRequest) (*iop.RegisterResponse, error) {
return &iop.RegisterResponse{
Accepted: true,
NodeId: "exhaust-test-node",
Config: &iop.NodeConfigPayload{Runtime: &iop.NodeRuntimeConfig{Concurrency: 1}},
}, nil
},
)
bindEdgeReadyResponder(client, readyCh)
acceptedCh <- client
return client
})
if err := server.Start(serverCtx); err != nil {
t.Fatalf("start server: %v", err)
}
t.Cleanup(func() { server.Stop() })
// Fake dialer: first call uses real DialEdge; subsequent calls fail immediately.
// established fires once the initial session is fully established so the test
// can induce the disconnect without racing the initial registration.
established := make(chan struct{}, 1)
var dialCount int32
fakeDialer := bootstrap.DialFunc(func(ctx context.Context, edgeAddr, token string, logger *zap.Logger) (*transport.RegisterResult, error) {
if atomic.AddInt32(&dialCount, 1) == 1 {
res, err := transport.DialEdge(ctx, edgeAddr, token, logger)
if err == nil {
established <- struct{}{}
}
return res, err
}
return nil, errors.New("fake reconnect failure")
})
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 3, IntervalSec: 0},
Logging: config.LoggingConf{Level: "error"},
}
app := fx.New(
bootstrap.Module(cfg, bootstrap.WithDialer(fakeDialer)),
fx.NopLogger,
)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
defer func() { _ = app.Stop(context.Background()) }()
// Wait for the initial session to establish, then close it from the edge side.
select {
case <-established:
case <-time.After(3 * time.Second):
t.Fatal("timeout waiting for initial session establishment")
}
var firstClient *toki.TcpClient
select {
case firstClient = <-acceptedCh:
case <-time.After(3 * time.Second):
t.Fatal("timeout waiting for first connection")
}
// Close only after the ready handshake so the first session counts as an
// established connection, not a mid-SignalReady failure.
select {
case <-readyCh:
case <-time.After(3 * time.Second):
t.Fatal("timeout waiting for initial ready handshake")
}
_ = firstClient.Close()
// Supervisor exhausts retries → Shutdown(ExitCode(1)).
select {
case sig := <-app.Wait():
if sig.ExitCode != 1 {
t.Fatalf("expected exit code 1, got %d", sig.ExitCode)
}
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for supervisor exhaustion shutdown")
}
// 1 initial dial + 3 retry attempts = 4 total.
if count := atomic.LoadInt32(&dialCount); count != 4 {
t.Fatalf("expected 4 dial attempts (1 initial + 3 retries), got %d", count)
}
}
// TestReconnectSupervisorPolicyTimingAndLimit verifies that the supervisor
// respects MaxAttempts=10 and that the fake sleeper receives IntervalSec=10s
// before each reconnect attempt (SDD reconnect_waiting state).
func TestReconnectSupervisorPolicyTimingAndLimit(t *testing.T) {
useTempCwd(t)
host, port, addr := freeAddrForBootstrap(t)
serverCtx, serverCancel := context.WithTimeout(context.Background(), 10*time.Second)
t.Cleanup(serverCancel)
acceptedCh := make(chan *toki.TcpClient, 4)
readyCh := make(chan struct{}, 4)
server := toki.NewTcpServer(host, port, func(conn net.Conn) *toki.TcpClient {
client := toki.NewTcpClient(conn, 30, 10, bootstrapEdgeParserMap())
toki.AddRequestListenerTyped[*iop.RegisterRequest, *iop.RegisterResponse](
&client.Communicator,
func(_ *iop.RegisterRequest) (*iop.RegisterResponse, error) {
return &iop.RegisterResponse{
Accepted: true,
NodeId: "policy-test-node",
Config: &iop.NodeConfigPayload{Runtime: &iop.NodeRuntimeConfig{Concurrency: 1}},
}, nil
},
)
bindEdgeReadyResponder(client, readyCh)
acceptedCh <- client
return client
})
if err := server.Start(serverCtx); err != nil {
t.Fatalf("start server: %v", err)
}
t.Cleanup(func() { server.Stop() })
// Fake dialer: first call uses real DialEdge; subsequent calls fail immediately.
// established fires once the initial session is fully established so the test
// can induce the disconnect without racing the initial registration.
established := make(chan struct{}, 1)
var dialCount int32
fakeDialer := bootstrap.DialFunc(func(ctx context.Context, edgeAddr, token string, logger *zap.Logger) (*transport.RegisterResult, error) {
if atomic.AddInt32(&dialCount, 1) == 1 {
res, err := transport.DialEdge(ctx, edgeAddr, token, logger)
if err == nil {
established <- struct{}{}
}
return res, err
}
return nil, errors.New("fake reconnect failure")
})
// Fake sleeper records received durations without blocking.
var slMu sync.Mutex
var sleptDurations []time.Duration
fakeSleeper := func(_ context.Context, d time.Duration) {
slMu.Lock()
sleptDurations = append(sleptDurations, d)
slMu.Unlock()
}
cfg := &config.NodeConfig{
Transport: config.TransportConf{EdgeAddr: addr, Token: "tok"},
Reconnect: config.ReconnectConf{MaxAttempts: 10, IntervalSec: 10},
Logging: config.LoggingConf{Level: "error"},
}
app := fx.New(
bootstrap.Module(cfg, bootstrap.WithDialer(fakeDialer), bootstrap.WithSleeper(fakeSleeper)),
fx.NopLogger,
)
startCtx, startCancel := context.WithTimeout(context.Background(), 5*time.Second)
defer startCancel()
if err := app.Start(startCtx); err != nil {
t.Fatalf("app start: %v", err)
}
defer func() { _ = app.Stop(context.Background()) }()
// Wait for the initial session to establish, then close it from the edge side.
select {
case <-established:
case <-time.After(3 * time.Second):
t.Fatal("timeout waiting for initial session establishment")
}
var firstClient *toki.TcpClient
select {
case firstClient = <-acceptedCh:
case <-time.After(3 * time.Second):
t.Fatal("timeout waiting for first connection")
}
// Close only after the ready handshake so the first session counts as an
// established connection, not a mid-SignalReady failure.
select {
case <-readyCh:
case <-time.After(3 * time.Second):
t.Fatal("timeout waiting for initial ready handshake")
}
_ = firstClient.Close()
// Supervisor exhausts all 10 retries → Shutdown(ExitCode(1)).
select {
case sig := <-app.Wait():
if sig.ExitCode != 1 {
t.Fatalf("expected exit code 1, got %d", sig.ExitCode)
}
case <-time.After(5 * time.Second):
t.Fatal("timeout waiting for supervisor exhaustion shutdown")
}
// 1 initial dial + 10 retry attempts = 11 total.
if count := atomic.LoadInt32(&dialCount); count != 11 {
t.Fatalf("expected 11 dial attempts (1 initial + 10 retries), got %d", count)
}
// Each retry is preceded by a sleep of IntervalSec=10s (reconnect_waiting state).
slMu.Lock()
durations := append([]time.Duration(nil), sleptDurations...)
slMu.Unlock()
if len(durations) != 10 {
t.Fatalf("expected 10 sleep calls (one before each retry), got %d: %v", len(durations), durations)
}
for i, d := range durations {
if d != 10*time.Second {
t.Errorf("sleep[%d] = %v, want 10s", i, d)
}
}
}