opensource/coder
3
0
Fork 0
mirror of https://github.com/coder/coder.git synced 2026-09-24 15:04:27 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix(coderd/x/chatd): handle truncated provider streams (#25074)

Browse Source 
coder/fantasy now fails closed when Anthropic or OpenAI Responses
streams close before their provider terminal events instead of yielding
a successful finish.

This bumps the fantasy replacement to coder/fantasy#33 and teaches chat
error classification to treat those failures as retryable timeout errors
with explicit stream-closed messages.

<img width="875" height="311" alt="image"
src="https://github.com/user-attachments/assets/69c6f7b5-c885-46d2-a88b-b7a2b111bd55"
/>
This commit is contained in:
Ethan
2026-05-08 15:52:42 +10:00
committed by GitHub
parent de9cdca77e
commit b6dbc5614c
9 changed files with 215 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
enterprise/coderd/x/chatd/chatd_test.go
+30 -8
View File
@@ -1444,6 +1444,7 @@ func TestSubscribeRelayDrainWithinGraceLeavesBufferRetained(t *testing.T) {
db, ps := dbtestutil.NewDB(t)
workerID := uuid.New()
subscriberID := uuid.New()
ctx := testutil.Context(t, testutil.WaitLong)
openAIURL := chattest.NewOpenAI(t, func(req *chattest.OpenAIRequest) chattest.OpenAIResponse {
if !req.Stream {
@@ -1458,16 +1459,19 @@ func TestSubscribeRelayDrainWithinGraceLeavesBufferRetained(t *testing.T) {
// Freeze the worker's clock so streamJanitorLoop cannot race the
// buffer-retained assertion on slow CI.
workerClock := quartz.NewMock(t)
trapAcquire := workerClock.Trap().NewTicker("chatd", "acquire")
defer trapAcquire.Close()
worker := osschatd.New(osschatd.Config{
Logger: workerLogger,
Database: db,
ReplicaID: workerID,
Pubsub: ps,
PendingChatAcquireInterval: time.Hour,
PendingChatAcquireInterval: time.Millisecond,
InFlightChatStaleAfter: testutil.WaitSuperLong,
Clock: workerClock,
})
worker.Start()
trapAcquire.MustWait(ctx).MustRelease(ctx)
t.Cleanup(func() {
require.NoError(t, worker.Close())
})
@@ -1501,23 +1505,41 @@ func TestSubscribeRelayDrainWithinGraceLeavesBufferRetained(t *testing.T) {
return snapshot, relayEvents, cancel, nil
}, subscriberClock)
ctx := testutil.Context(t, testutil.WaitLong)
user, org, model := seedChatDependencies(t, db)
setOpenAIProviderBaseURL(ctx, t, db, openAIURL)
chat := seedWaitingChat(t, db, org.ID, user, model, "relay-drain-characterization")
// Seed the pending turn directly instead of using SendMessage.
// SendMessage publishes a pending control notification that is
// irrelevant to this relay-retention case. Under CI that
// notification can arrive after processChat arms its control
// subscription and interrupt the worker before it emits parts.
dbgen.ChatMessage(t, db, database.ChatMessage{
ChatID: chat.ID,
CreatedBy: uuid.NullUUID{UUID: user.ID, Valid: true},
ModelConfigID: uuid.NullUUID{UUID: model.ID, Valid: true},
Role: database.ChatMessageRoleUser,
Content: pqtype.NullRawMessage{
RawMessage: json.RawMessage(`[{"type":"text","text":"hello"}]`),
Valid: true,
},
})
_, err := db.UpdateChatStatus(ctx, database.UpdateChatStatusParams{
ID: chat.ID,
Status: database.ChatStatusPending,
})
require.NoError(t, err)
// Attach before processing so the relay opens as soon as
// status=running arrives.
_, events, subCancel, ok := subscriber.Subscribe(ctx, chat.ID, nil, 0)
require.True(t, ok)
_, err := worker.SendMessage(ctx, osschatd.SendMessageOptions{
ChatID: chat.ID,
CreatedBy: user.ID,
Content: []codersdk.ChatMessagePart{codersdk.ChatMessageText("hello")},
})
require.NoError(t, err)
// Wake the worker with the acquire ticker. This keeps the
// setup free of pending control notifications while still
// exercising the normal processing loop.
workerClock.Advance(time.Millisecond).MustWait(ctx)
// Drain events until processing has clearly completed: we need
// the assistant message and at least one message_part so we know