File size: 8,616 Bytes
3d7a2d5 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 | package agent
import (
"context"
"runtime"
"testing"
"time"
"charm.land/fantasy"
"charm.land/x/vcr"
"github.com/charmbracelet/crush/internal/message"
"github.com/stretchr/testify/require"
)
// fakeLanguageModel is a [fantasy.LanguageModel] stub that records the
// context its methods were called with and can be configured with a custom
// stream body.
type fakeLanguageModel struct {
generateCtx context.Context
streamCtx context.Context
stream func(yield func(fantasy.StreamPart) bool)
}
func (f *fakeLanguageModel) Generate(ctx context.Context, _ fantasy.Call) (*fantasy.Response, error) {
f.generateCtx = ctx
return &fantasy.Response{}, nil
}
func (f *fakeLanguageModel) Stream(ctx context.Context, _ fantasy.Call) (fantasy.StreamResponse, error) {
f.streamCtx = ctx
if f.stream == nil {
return func(yield func(fantasy.StreamPart) bool) {
yield(fantasy.StreamPart{})
}, nil
}
return f.stream, nil
}
func (f *fakeLanguageModel) GenerateObject(context.Context, fantasy.ObjectCall) (*fantasy.ObjectResponse, error) {
return &fantasy.ObjectResponse{}, nil
}
func (f *fakeLanguageModel) StreamObject(context.Context, fantasy.ObjectCall) (fantasy.ObjectStreamResponse, error) {
return nil, nil
}
func (f *fakeLanguageModel) Provider() string { return "fake" }
func (f *fakeLanguageModel) Model() string { return "fake-model" }
func TestNewRequestTimeoutModel_Disabled(t *testing.T) {
t.Parallel()
inner := &fakeLanguageModel{}
require.Same(t, inner, newRequestTimeoutModel(inner, 0))
require.Same(t, inner, newRequestTimeoutModel(inner, -time.Second))
}
func TestRequestTimeoutModel_GenerateDeadline(t *testing.T) {
t.Parallel()
inner := &fakeLanguageModel{}
m := newRequestTimeoutModel(inner, 5*time.Minute)
_, err := m.Generate(t.Context(), fantasy.Call{})
require.NoError(t, err)
_, ok := inner.generateCtx.Deadline()
require.True(t, ok, "Generate should run under a deadline")
}
func TestRequestTimeoutModel_StreamDeadlineOutlivesCall(t *testing.T) {
t.Parallel()
inner := &fakeLanguageModel{}
m := newRequestTimeoutModel(inner, 5*time.Minute)
stream, err := m.Stream(t.Context(), fantasy.Call{})
require.NoError(t, err)
require.NoError(t, inner.streamCtx.Err(), "the idle timer must not fire while the stream is being consumed")
for range stream {
}
require.ErrorIs(t, inner.streamCtx.Err(), context.Canceled, "the stream context should be released after the stream ends")
}
func TestRequestTimeoutModel_StreamAbortsWhenIdle(t *testing.T) {
t.Parallel()
inner := &fakeLanguageModel{}
// A stream that outlives the idle window: it waits for the context to
// be done and reports what it observed.
streamObserved := make(chan error, 1)
inner.stream = func(yield func(fantasy.StreamPart) bool) {
<-inner.streamCtx.Done()
streamObserved <- inner.streamCtx.Err()
}
m := newRequestTimeoutModel(inner, 10*time.Millisecond)
stream, err := m.Stream(t.Context(), fantasy.Call{})
require.NoError(t, err)
done := make(chan struct{})
go func() {
defer close(done)
for range stream {
}
}()
select {
case err := <-streamObserved:
require.ErrorIs(t, err, context.Canceled)
case <-time.After(5 * time.Second):
t.Fatal("stream was not aborted by the idle timeout")
}
<-done
}
func TestRequestTimeoutModel_ActiveStreamSurvives(t *testing.T) {
t.Parallel()
inner := &fakeLanguageModel{}
// A stream that keeps sending data: total runtime exceeds the timeout,
// but every gap is shorter than the idle window, so it must finish.
inner.stream = func(yield func(fantasy.StreamPart) bool) {
for range 6 {
time.Sleep(10 * time.Millisecond)
if !yield(fantasy.StreamPart{}) {
return
}
}
}
m := newRequestTimeoutModel(inner, 25*time.Millisecond)
stream, err := m.Stream(t.Context(), fantasy.Call{})
require.NoError(t, err)
parts := 0
for part := range stream {
require.NoError(t, part.Error)
parts++
}
require.Equal(t, 6, parts)
}
// blockingModel blocks until the context is done and then returns the
// context error, the way a hung provider request would.
type blockingModel struct {
fakeLanguageModel
}
func (b *blockingModel) Generate(ctx context.Context, _ fantasy.Call) (*fantasy.Response, error) {
<-ctx.Done()
return nil, ctx.Err()
}
func TestRequestTimeoutModel_GenerateReportsTimeout(t *testing.T) {
t.Parallel()
m := newRequestTimeoutModel(&blockingModel{}, 10*time.Millisecond)
_, err := m.Generate(t.Context(), fantasy.Call{})
require.Error(t, err)
var timeoutErr *requestTimeoutError
require.ErrorAs(t, err, &timeoutErr)
require.Equal(t, 10*time.Millisecond, timeoutErr.timeout)
require.ErrorIs(t, err, context.DeadlineExceeded, "the deadline must stay detectable through the chain")
require.Contains(t, err.Error(), "timed out after 10ms")
}
func TestRequestTimeoutModel_StreamReportsTimeout(t *testing.T) {
t.Parallel()
inner := &fakeLanguageModel{}
// A provider stream that fails with the context error once the deadline
// fires, mirroring how SDKs surface mid-stream aborts.
inner.stream = func(yield func(fantasy.StreamPart) bool) {
<-inner.streamCtx.Done()
yield(fantasy.StreamPart{Error: inner.streamCtx.Err()})
}
m := newRequestTimeoutModel(inner, 10*time.Millisecond)
stream, err := m.Stream(t.Context(), fantasy.Call{})
require.NoError(t, err)
var got error
for part := range stream {
if part.Error != nil {
got = part.Error
}
}
var timeoutErr *requestTimeoutError
require.ErrorAs(t, got, &timeoutErr)
require.ErrorIs(t, got, context.DeadlineExceeded)
}
func TestRequestTimeoutModel_ParentCancelPassesThrough(t *testing.T) {
t.Parallel()
m := newRequestTimeoutModel(&blockingModel{}, 5*time.Minute)
ctx, cancel := context.WithCancel(t.Context())
defer cancel()
go func() {
time.Sleep(10 * time.Millisecond)
cancel()
}()
_, err := m.Generate(ctx, fantasy.Call{})
require.ErrorIs(t, err, context.Canceled)
var timeoutErr *requestTimeoutError
require.NotErrorAs(t, err, &timeoutErr, "user cancellation must not be reported as a timeout")
}
func TestRequestTimeoutErrorMessages(t *testing.T) {
t.Parallel()
err := &requestTimeoutError{timeout: time.Second}
require.Equal(t, "LLM request timed out after 1s", err.Error())
require.Contains(t, err.userMessage(), "1s")
require.Contains(t, err.userMessage(), "request-timeout")
err.cause = context.DeadlineExceeded
require.Equal(t, "LLM request timed out after 1s: context deadline exceeded", err.Error())
idle := &requestTimeoutError{timeout: 2 * time.Second, idle: true}
require.Equal(t, "LLM stream received no data for 2s", idle.Error())
require.Contains(t, idle.userMessage(), "stopped sending data for 2s")
require.Contains(t, idle.userMessage(), "request-timeout")
}
// timeoutOnlyModel streams a single error part shaped exactly like the one
// requestTimeoutModel produces when its deadline fires.
type timeoutOnlyModel struct {
fakeLanguageModel
}
func (m *timeoutOnlyModel) Stream(context.Context, fantasy.Call) (fantasy.StreamResponse, error) {
timeoutErr := &requestTimeoutError{timeout: time.Second, idle: true, cause: context.DeadlineExceeded}
return func(yield func(fantasy.StreamPart) bool) {
yield(fantasy.StreamPart{Type: fantasy.StreamPartTypeError, Error: timeoutErr})
}, nil
}
// TestRequestTimeoutRunFinishMessage pins what the user sees when a request
// exhausts its timeout: a "Request timed out" finish that names the elapsed
// budget and how to change it, instead of a bare provider error.
func TestRequestTimeoutRunFinishMessage(t *testing.T) {
if runtime.GOOS == "windows" {
t.Skip("skipping on windows for now")
}
env := testEnv(t)
model := &timeoutOnlyModel{}
agent, err := coderAgent(vcr.NewRecorder(t), env, model, model)
require.NoError(t, err)
session, err := env.sessions.Create(t.Context(), "timeout session")
require.NoError(t, err)
_, err = agent.Run(t.Context(), SessionAgentCall{
Prompt: "Hello",
SessionID: session.ID,
MaxOutputTokens: 10000,
})
require.Error(t, err)
msgs, err := env.messages.List(t.Context(), session.ID)
require.NoError(t, err)
var finish *message.Finish
for _, msg := range msgs {
if msg.Role != message.Assistant {
continue
}
if part := msg.FinishPart(); part != nil {
finish = part
}
}
require.NotNil(t, finish, "the assistant message should carry a finish part")
require.Equal(t, message.FinishReasonError, finish.Reason)
require.Equal(t, "Request timed out", finish.Message)
require.Contains(t, finish.Details, "stopped sending data for 1s")
require.Contains(t, finish.Details, "request-timeout")
}
|