From 4cf4ee0121b228dd14c6339508bea5f90ce5b20d Mon Sep 17 00:00:00 2001
From: Hugo Dutka <hugo@coder.com>
Date: Mon, 15 Jun 2026 14:13:22 +0200
Subject: [PATCH] chore(coderd/x/chatd): log all chat errors (#26371)

When a chat hits a terminal error, for example "Request failed
unexpectedly", we don't log the full underlying error anywhere. This
fixes that.
---
 coderd/x/chatd/chatd_test.go | 74 ++++++++++++++++++++++++++++++++++++
 coderd/x/chatd/generation.go | 15 ++++++++
 2 files changed, 89 insertions(+)

diff --git a/coderd/x/chatd/chatd_test.go b/coderd/x/chatd/chatd_test.go
index 3ed65b6953..9873e91161 100644
--- a/coderd/x/chatd/chatd_test.go
+++ b/coderd/x/chatd/chatd_test.go
@@ -8221,6 +8221,80 @@ func newActiveTestServer(
 	return server
 }
 
+// sinkFieldValue returns the value of the named field from a captured log
+// entry.
+func sinkFieldValue(fields slog.Map, name string) (any, bool) {
+	for _, f := range fields {
+		if f.Name == name {
+			return f.Value, true
+		}
+	}
+	return nil, false
+}
+
+// TestActiveServer_GenerationErrorLogged drives a full chat worker against a
+// provider that returns a terminal error and asserts that chatd logs the
+// unsanitized failure so an administrator can later diagnose the underlying
+// reason, even though the user-facing message is sanitized.
+func TestActiveServer_GenerationErrorLogged(t *testing.T) {
+	t.Parallel()
+
+	ctx := testutil.Context(t, testutil.WaitLong)
+	db, ps := dbtestutil.NewDB(t)
+	sink := testutil.NewFakeSink(t)
+
+	const providerErrMessage = "synthetic provider failure for logging test"
+	openAIURL := chattest.NewOpenAI(t, func(req *chattest.OpenAIRequest) chattest.OpenAIResponse {
+		if !req.Stream {
+			return chattest.OpenAINonStreamingResponse("title")
+		}
+		// A 400 is non-retryable, so the worker fails the turn immediately
+		// instead of entering retry backoff.
+		return chattest.OpenAIErrorResponse(http.StatusBadRequest, "invalid_request_error", providerErrMessage)
+	})
+	user, org, model := seedChatDependenciesWithProvider(t, db, "openai", openAIURL)
+	server := newActiveTestServer(t, db, ps, func(cfg *chatd.Config) {
+		cfg.Logger = sink.Logger()
+	})
+
+	chat := createChatThroughServer(ctx, t, db, server, org.ID, user.ID, model.ID, "hello")
+	failed := waitForChatStatus(ctx, t, db, chat.ID, database.ChatStatusError)
+	require.True(t, failed.LastError.Valid)
+
+	isGenerationFailure := func(e slog.SinkEntry) bool {
+		return e.Level == slog.LevelWarn && e.Message == "chat generation failed"
+	}
+	var entry slog.SinkEntry
+	testutil.Eventually(ctx, t, func(context.Context) bool {
+		entries := sink.Entries(isGenerationFailure)
+		if len(entries) == 0 {
+			return false
+		}
+		entry = entries[0]
+		return true
+	}, testutil.IntervalFast)
+
+	chatID, ok := sinkFieldValue(entry.Fields, "chat_id")
+	require.True(t, ok, "chat_id field present")
+	require.Equal(t, chat.ID, chatID)
+
+	provider, ok := sinkFieldValue(entry.Fields, "provider")
+	require.True(t, ok, "provider field present")
+	require.Equal(t, "openai", provider)
+
+	statusCode, ok := sinkFieldValue(entry.Fields, "status_code")
+	require.True(t, ok, "status_code field present")
+	require.Equal(t, http.StatusBadRequest, statusCode)
+
+	// The unsanitized cause must be logged so administrators can see the
+	// underlying provider reason, even though the persisted user-facing
+	// message omits it.
+	errValue, ok := sinkFieldValue(entry.Fields, "error")
+	require.True(t, ok, "error field present")
+	require.Contains(t, fmt.Sprintf("%v", errValue), providerErrMessage)
+	require.NotContains(t, chatLastErrorMessage(failed.LastError), providerErrMessage)
+}
+
 func TestProposeChatTitle_DebugRun(t *testing.T) {
 	t.Parallel()
 
diff --git a/coderd/x/chatd/generation.go b/coderd/x/chatd/generation.go
index 47b17c8e45..43b255262e 100644
--- a/coderd/x/chatd/generation.go
+++ b/coderd/x/chatd/generation.go
@@ -13,6 +13,7 @@ import (
 	"github.com/sqlc-dev/pqtype"
 	"golang.org/x/xerrors"
 
+	"cdr.dev/slog/v3"
 	"github.com/coder/coder/v2/coderd/database"
 	"github.com/coder/coder/v2/coderd/x/chatd/chatdebug"
 	"github.com/coder/coder/v2/coderd/x/chatd/chaterror"
@@ -1017,6 +1018,20 @@ func (s *taskStarter) finishGenerationError(
 	cause error,
 	attemptFence generationAttemptFence,
 ) error {
+	classified := chaterror.Classify(cause)
+	// Log the unsanitized cause before persisting so administrators can
+	// diagnose the failure even when the classified user-facing message
+	// omits the underlying reason, and even if the persist below fails.
+	s.opts.Logger.Warn(ctx, "chat generation failed",
+		slog.F("chat_id", input.ChatID),
+		slog.F("worker_id", input.WorkerID),
+		slog.F("generation_attempt", input.GenerationAttempt),
+		slog.F("error_kind", classified.Kind),
+		slog.F("provider", classified.Provider),
+		slog.F("status_code", classified.StatusCode),
+		slog.F("retryable", classified.Retryable),
+		slog.Error(cause),
+	)
 	lastError, message := generationLastError(cause)
 	var committed database.Chat
 	err := machine.Update(ctx, func(tx *chatstate.Tx, store database.Store) error {