server: capture failed (non-200) LLM requests (#862)
Store a request/response capture for non-200 responses so failed requests can be inspected in the activity log's Capture dialog, matching the existing behavior for successful requests. - extract storeCapture/decodeResponseBody helpers to share capture logic between the success and non-200 paths - record non-200 bodies (decompressed) so error details are viewable - the activity UI already gates the View button on has_capture, so it now appears for failed requests with no UI changes - add tests for capturing failed requests and the disabled-captures case closes #766
This commit is contained in:
Benson Wong
GitHub
@@ -90,6 +90,172 @@ func TestMetricsMonitor_RecordMetadata(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
func TestMetricsMonitor_RecordFailedRequestCapture(t *testing.T) {
|
||||
mm := newMetricsMonitor(logmon.NewWriter(io.Discard), 10, 5)
|
||||
r := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
|
||||
reqHeaders := map[string]string{"content-type": "application/json"}
|
||||
|
||||
w := httptest.NewRecorder()
|
||||
copier := newBodyCopier(w)
|
||||
copier.Header().Set("Content-Type", "application/json")
|
||||
copier.WriteHeader(http.StatusBadGateway)
|
||||
copier.Write([]byte(`{"error":{"message":"model unavailable"}}`))
|
||||
|
||||
reqBody := []byte(`{"model":"m","messages":[]}`)
|
||||
mm.record("m", r, copier, captureAll, reqBody, reqHeaders)
|
||||
|
||||
entries := mm.getMetrics()
|
||||
if len(entries) != 1 {
|
||||
t.Fatalf("want 1 entry, got %d", len(entries))
|
||||
}
|
||||
entry := entries[0]
|
||||
if entry.RespStatusCode != http.StatusBadGateway {
|
||||
t.Errorf("status = %d, want %d", entry.RespStatusCode, http.StatusBadGateway)
|
||||
}
|
||||
if entry.ErrorMsg != "model unavailable" {
|
||||
t.Errorf("error_msg = %q, want extracted message", entry.ErrorMsg)
|
||||
}
|
||||
if !entry.HasCapture {
|
||||
t.Fatal("failed request should capture the request so it can be inspected")
|
||||
}
|
||||
|
||||
got := mm.getCaptureByID(entry.ID)
|
||||
if got == nil {
|
||||
t.Fatal("capture not found")
|
||||
}
|
||||
if string(got.ReqBody) != `{"model":"m","messages":[]}` {
|
||||
t.Errorf("req body = %q", got.ReqBody)
|
||||
}
|
||||
if len(got.RespBody) != 0 {
|
||||
t.Errorf("resp body stored for failed request (len=%d); want none", len(got.RespBody))
|
||||
}
|
||||
if got.RespHeaders["Content-Type"] != "application/json" {
|
||||
t.Errorf("resp Content-Type = %q", got.RespHeaders["Content-Type"])
|
||||
}
|
||||
}
|
||||
|
||||
func TestMetricsMonitor_RecordFailedRequestStatusFallback(t *testing.T) {
|
||||
// Non-JSON error body: ErrorMsg falls back to the HTTP status text.
|
||||
mm := newMetricsMonitor(logmon.NewWriter(io.Discard), 10, 5)
|
||||
r := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
|
||||
|
||||
w := httptest.NewRecorder()
|
||||
copier := newBodyCopier(w)
|
||||
copier.WriteHeader(http.StatusBadGateway)
|
||||
copier.Write([]byte("<html>upstream down</html>"))
|
||||
|
||||
mm.record("m", r, copier, captureAll, nil, nil)
|
||||
|
||||
entries := mm.getMetrics()
|
||||
if len(entries) != 1 {
|
||||
t.Fatalf("want 1 entry, got %d", len(entries))
|
||||
}
|
||||
if entries[0].ErrorMsg != "502 Bad Gateway" {
|
||||
t.Errorf("error_msg = %q, want status text", entries[0].ErrorMsg)
|
||||
}
|
||||
}
|
||||
|
||||
func TestMetricsMonitor_RecordFailedRequestCaptureDisabled(t *testing.T) {
|
||||
mm := newMetricsMonitor(logmon.NewWriter(io.Discard), 10, 0) // captures disabled
|
||||
r := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
|
||||
|
||||
w := httptest.NewRecorder()
|
||||
copier := newBodyCopier(w)
|
||||
copier.WriteHeader(http.StatusInternalServerError)
|
||||
copier.Write([]byte(`{"error":"boom"}`))
|
||||
|
||||
mm.record("m", r, copier, captureAll, []byte("req"), nil)
|
||||
|
||||
entries := mm.getMetrics()
|
||||
if len(entries) != 1 {
|
||||
t.Fatalf("want 1 entry, got %d", len(entries))
|
||||
}
|
||||
if entries[0].HasCapture {
|
||||
t.Fatal("captures disabled, HasCapture should be false")
|
||||
}
|
||||
// ErrorMsg is independent of whether captures are enabled.
|
||||
if entries[0].ErrorMsg != "boom" {
|
||||
t.Errorf("error_msg = %q, want boom", entries[0].ErrorMsg)
|
||||
}
|
||||
if mm.getCaptureByID(entries[0].ID) != nil {
|
||||
t.Fatal("no capture should be stored when disabled")
|
||||
}
|
||||
}
|
||||
|
||||
func TestMetricsMonitor_RecordDecompressionFailureSetsErrorMsg(t *testing.T) {
|
||||
mm := newMetricsMonitor(logmon.NewWriter(io.Discard), 10, 5)
|
||||
r := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
|
||||
|
||||
w := httptest.NewRecorder()
|
||||
copier := newBodyCopier(w)
|
||||
copier.Header().Set("Content-Encoding", "gzip")
|
||||
copier.WriteHeader(http.StatusOK)
|
||||
copier.Write([]byte("not-really-gzip"))
|
||||
|
||||
mm.record("m", r, copier, captureAll, []byte("req"), nil)
|
||||
|
||||
entries := mm.getMetrics()
|
||||
if len(entries) != 1 {
|
||||
t.Fatalf("want 1 entry, got %d", len(entries))
|
||||
}
|
||||
if entries[0].ErrorMsg == "" {
|
||||
t.Fatal("expected ErrorMsg for decompression failure")
|
||||
}
|
||||
// Raw bytes must not be stored when the body could not be decoded.
|
||||
if entries[0].HasCapture {
|
||||
t.Fatal("decompression failure should not store a capture")
|
||||
}
|
||||
}
|
||||
|
||||
func TestMetricsMonitor_DecodeResponseBody(t *testing.T) {
|
||||
mm := newMetricsMonitor(logmon.NewWriter(io.Discard), 10, 5)
|
||||
|
||||
// No Content-Encoding: body returned unchanged.
|
||||
w := httptest.NewRecorder()
|
||||
copier := newBodyCopier(w)
|
||||
copier.Write([]byte("plain"))
|
||||
got, err := mm.decodeResponseBody(copier, "/p")
|
||||
if err != nil || string(got) != "plain" {
|
||||
t.Fatalf("plain body = %q, err = %v", got, err)
|
||||
}
|
||||
|
||||
// Bogus gzip payload: returns an error and no body (no raw bytes kept).
|
||||
w2 := httptest.NewRecorder()
|
||||
copier2 := newBodyCopier(w2)
|
||||
copier2.Header().Set("Content-Encoding", "gzip")
|
||||
copier2.Write([]byte("not-really-gzip"))
|
||||
got, err = mm.decodeResponseBody(copier2, "/p")
|
||||
if err == nil {
|
||||
t.Fatal("expected decompression error")
|
||||
}
|
||||
if got != nil {
|
||||
t.Errorf("expected nil body on failure, got %q", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestServer_ExtractErrorMessage(t *testing.T) {
|
||||
cases := []struct {
|
||||
name string
|
||||
body string
|
||||
want string
|
||||
}{
|
||||
{"openai object", `{"error":{"message":"rate limited"}}`, "rate limited"},
|
||||
{"string error", `{"error":"bad request"}`, "bad request"},
|
||||
{"message field", `{"message":"nope"}`, "nope"},
|
||||
{"detail field", `{"detail":"oops"}`, "oops"},
|
||||
{"object error ignored", `{"error":{"code":42}}`, ""},
|
||||
{"no error", `{"usage":{}}`, ""},
|
||||
{"invalid json", `not-json`, ""},
|
||||
}
|
||||
for _, tc := range cases {
|
||||
t.Run(tc.name, func(t *testing.T) {
|
||||
if got := extractErrorMessage([]byte(tc.body)); got != tc.want {
|
||||
t.Errorf("extractErrorMessage = %q, want %q", got, tc.want)
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
func TestServer_ParseMetrics_Infill(t *testing.T) {
|
||||
// /infill responses are arrays; timings live in the last element.
|
||||
body := `[{"content":"a"},{"content":"b","timings":{"prompt_n":5,"predicted_n":9,"prompt_ms":10,"predicted_ms":20}}]`
|
||||
|
||||
Reference in New Issue
Block a user