iop/apps/edge/internal/openai/usage_metrics_test.go

1112 lines
48 KiB
Go

package openai
import (
"context"
"fmt"
"net/http"
"net/http/httptest"
"strings"
"testing"
"time"
"github.com/prometheus/client_golang/prometheus/testutil"
"iop/packages/go/config"
iop "iop/proto/gen/iop"
)
// principalTokenCfg builds an EdgeOpenAIConf whose bearer token maps to a fixed
// principal so usage-metric label values are populated.
func principalTokenCfg(rawToken, adapter string) config.EdgeOpenAIConf {
return config.EdgeOpenAIConf{
Adapter: adapter,
PrincipalTokens: []config.OpenAIPrincipalTokenConf{
{TokenRef: "iop-tok-alice", TokenHashSHA256: sha256Hex(rawToken), PrincipalRef: "user:alice", PrincipalAlias: "alice"},
},
}
}
func requestTokenValue(t *testing.T, l usageLabels, tokenType string) float64 {
t.Helper()
return testutil.ToFloat64(openAIUsageTokensTotal.WithLabelValues(
l.edgeID, l.principalRef, l.principalAlias, l.tokenRef,
l.modelGroup, l.endpoint, l.responseMode, tokenType,
))
}
// TestOpenAIUsageMetricsLabelsExcludeSecretsAndHighCardinality asserts the label
// allowlists never carry raw tokens, prompt/response text, or per-request
// high-cardinality identifiers (SDD S08).
func TestOpenAIUsageMetricsLabelsExcludeSecretsAndHighCardinality(t *testing.T) {
// Substrings that would indicate a per-request identifier, a raw secret, or
// raw payload text. response_mode / model_group are low-cardinality enums and
// are intentionally not caught here.
forbidden := []string{
"request_id", "session_id", "run_id", "token_hash", "bearer",
"authorization", "api_key", "prompt", "secret", "raw",
}
// token_ref is an allowlisted stable alias, not a raw token; guard against a
// bare "token" label that could carry the secret itself.
for _, set := range [][]string{usageRequestLabelNames, usageTokenLabelNames, usageReasoningLabelNames} {
for _, name := range set {
if name == "token" {
t.Fatalf("label %q may carry a raw token value; use token_ref", name)
}
for _, bad := range forbidden {
if strings.Contains(name, bad) {
t.Fatalf("label %q contains forbidden fragment %q", name, bad)
}
}
}
}
// The allowlist must still carry the intended identity/attribution labels.
for _, want := range []string{"edge_id", "principal_ref", "principal_alias", "token_ref", "model_group", "endpoint", "response_mode"} {
if !containsString(usageRequestLabelNames, want) {
t.Fatalf("request labels missing %q", want)
}
}
if !containsString(usageRequestLabelNames, "usage_source") || !containsString(usageRequestLabelNames, "status") {
t.Fatalf("request labels must include status and usage_source: %v", usageRequestLabelNames)
}
if !containsString(usageTokenLabelNames, "token_type") {
t.Fatalf("token labels must include token_type: %v", usageTokenLabelNames)
}
}
// TestOpenAIUsageMetricsCountTokenTypes drives a normalized chat completion whose
// run reports input/output/reasoning/cached usage and asserts each token type is
// counted under a provider_reported request (SDD S06).
func TestOpenAIUsageMetricsCountTokenTypes(t *testing.T) {
const rawToken = "sk-count-raw-token"
const edgeID = "edge-count-token-types"
const model = "count-model"
fake := &fakeRunService{events: make(chan *iop.RunEvent, 2)}
fake.events <- &iop.RunEvent{Type: "delta", Delta: "hi"}
fake.events <- &iop.RunEvent{Type: "complete", Usage: &iop.Usage{
InputTokens: 10, OutputTokens: 5, ReasoningTokens: 3, CachedInputTokens: 2,
}}
srv := NewServer(principalTokenCfg(rawToken, "ollama"), fake, nil)
srv.SetEdgeID(edgeID)
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: "normalized",
}
before := map[string]float64{
tokenTypeInput: requestTokenValue(t, labels, tokenTypeInput),
tokenTypeOutput: requestTokenValue(t, labels, tokenTypeOutput),
tokenTypeReasoning: requestTokenValue(t, labels, tokenTypeReasoning),
tokenTypeCachedInput: requestTokenValue(t, labels, tokenTypeCachedInput),
}
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, "normalized", usageStatusSuccess, usageSourceProviderReported,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"count-model",
"messages":[{"role":"user","content":"hi"}]
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
for tokenType, want := range map[string]float64{
tokenTypeInput: 10, tokenTypeOutput: 5, tokenTypeReasoning: 3, tokenTypeCachedInput: 2,
} {
got := requestTokenValue(t, labels, tokenType) - before[tokenType]
if got != want {
t.Fatalf("token_type %s: got delta %v, want %v", tokenType, got, want)
}
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, "normalized", usageStatusSuccess, usageSourceProviderReported,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total success/provider_reported: got delta %v, want 1", reqAfter-reqBefore)
}
}
// TestOpenAIReasoningObservedDoesNotEstimateTokens verifies that reasoning text
// observed without a provider-reported reasoning token count increments only the
// auxiliary reasoning counters and never the reasoning token counter (SDD S07).
func TestOpenAIReasoningObservedDoesNotEstimateTokens(t *testing.T) {
const rawToken = "sk-reasoning-raw-token"
const edgeID = "edge-reasoning-observed"
const model = "reasoning-model"
const reasoning = "abcde" // 5 chars
fake := &fakeRunService{events: make(chan *iop.RunEvent, 3)}
fake.events <- &iop.RunEvent{Type: "reasoning_delta", Delta: reasoning}
fake.events <- &iop.RunEvent{Type: "delta", Delta: "answer"}
fake.events <- &iop.RunEvent{Type: "complete", Usage: &iop.Usage{InputTokens: 8, OutputTokens: 4}}
srv := NewServer(principalTokenCfg(rawToken, "ollama"), fake, nil)
srv.SetEdgeID(edgeID)
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: "normalized",
}
reasoningBefore := requestTokenValue(t, labels, tokenTypeReasoning)
observedBefore := testutil.ToFloat64(openAIReasoningObservedTotal.WithLabelValues(labels.reasoningValues()...))
charsBefore := testutil.ToFloat64(openAIReasoningCharsTotal.WithLabelValues(labels.reasoningValues()...))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"reasoning-model",
"messages":[{"role":"user","content":"hi"}]
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
if got := requestTokenValue(t, labels, tokenTypeReasoning) - reasoningBefore; got != 0 {
t.Fatalf("reasoning token counter must not move when provider did not report reasoning tokens: got delta %v", got)
}
if got := testutil.ToFloat64(openAIReasoningObservedTotal.WithLabelValues(labels.reasoningValues()...)) - observedBefore; got != 1 {
t.Fatalf("reasoning_observed_total: got delta %v, want 1", got)
}
if got := testutil.ToFloat64(openAIReasoningCharsTotal.WithLabelValues(labels.reasoningValues()...)) - charsBefore; got != float64(len(reasoning)) {
t.Fatalf("reasoning_chars_total: got delta %v, want %d", got, len(reasoning))
}
}
// TestProviderTunnelPassthroughPreservesBodyAndObservesUsage verifies that a pure
// passthrough provider response is relayed byte-for-byte while its usage is
// observed into metrics (SDD S05/S06).
func TestProviderTunnelPassthroughPreservesBodyAndObservesUsage(t *testing.T) {
const rawToken = "sk-passthrough-raw-token"
const edgeID = "edge-passthrough-usage"
const model = "ornith:35b"
providerBody := `{"choices":[{"message":{"role":"assistant","content":"hi there"}}],"usage":{"prompt_tokens":12,"completion_tokens":7,"prompt_tokens_details":{"cached_tokens":3},"completion_tokens_details":{"reasoning_tokens":4}}}`
fake := &fakeRunService{tunnelFrames: staticProviderTunnelFrames(providerBody)}
srv := NewServer(config.EdgeOpenAIConf{
PrincipalTokens: []config.OpenAIPrincipalTokenConf{
{TokenRef: "iop-tok-alice", TokenHashSHA256: sha256Hex(rawToken), PrincipalRef: "user:alice", PrincipalAlias: "alice"},
},
}, fake, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-vllm": "Ornith-1.0-35B"}}})
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: responseModePassthrough,
}
before := map[string]float64{
tokenTypeInput: requestTokenValue(t, labels, tokenTypeInput),
tokenTypeOutput: requestTokenValue(t, labels, tokenTypeOutput),
tokenTypeReasoning: requestTokenValue(t, labels, tokenTypeReasoning),
tokenTypeCachedInput: requestTokenValue(t, labels, tokenTypeCachedInput),
}
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"ornith:35b",
"messages":[{"role":"user","content":"hello"}]
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
if w.Body.String() != providerBody {
t.Fatalf("passthrough body must be preserved byte-for-byte.\n got: %s\nwant: %s", w.Body.String(), providerBody)
}
if len(fake.tunnelReqsSnapshot()) != 1 {
t.Fatalf("expected exactly one provider tunnel dispatch, got %d", len(fake.tunnelReqsSnapshot()))
}
for tokenType, want := range map[string]float64{
tokenTypeInput: 12, tokenTypeOutput: 7, tokenTypeReasoning: 4, tokenTypeCachedInput: 3,
} {
got := requestTokenValue(t, labels, tokenType) - before[tokenType]
if got != want {
t.Fatalf("passthrough token_type %s: got delta %v, want %v", tokenType, got, want)
}
}
}
// TestOpenAIScopeExcludesA2AAndNonOpenAISurfaces documents that usage metering is
// scoped to the OpenAI-compatible input surface only (SDD S04). The endpoint
// label allowlist carries the OpenAI routes and no A2A/CLI endpoint value, and
// this plan does not modify the A2A surface.
func TestOpenAIScopeExcludesA2AAndNonOpenAISurfaces(t *testing.T) {
endpoints := map[string]struct{}{
usageEndpointChatCompletions: {},
usageEndpointResponses: {},
}
for _, forbidden := range []string{"a2a", "cli", "ollama", "message/send", "tasks"} {
if _, ok := endpoints[forbidden]; ok {
t.Fatalf("usage endpoint label must not include non-OpenAI surface %q", forbidden)
}
}
if !containsString(usageTokenLabelNames, "endpoint") {
t.Fatalf("token metrics must carry an endpoint label to keep OpenAI scope explicit")
}
}
// TestProviderTunnelSidebandStreamingEmitsUsageMetrics verifies that a
// passthrough+sideband streaming response increments the request counter
// with status=success and usage_source=provider_reported (REVIEW_USAGE_METRIC-1).
func TestProviderTunnelSidebandStreamingEmitsUsageMetrics(t *testing.T) {
const rawToken = "sk-sideband-stream-token"
const edgeID = "edge-sideband-stream-metrics"
const model = "pool-model"
providerBody := `{"choices":[{"delta":{"content":"hi"}}]}
data: [DONE]
`
frames := make(chan *iop.ProviderTunnelFrame, 5)
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_RESPONSE_START,
StatusCode: 200,
Headers: map[string]string{"Content-Type": "text/event-stream"},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY, Body: []byte(providerBody)}
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_USAGE,
Usage: &iop.Usage{InputTokens: 5, OutputTokens: 3},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_END, End: true}
close(frames)
srv := NewServer(principalTokenCfg(rawToken, "ollama"), &fakeRunService{tunnelFrames: frames}, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"stream":true,
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total success/provider_reported: got delta %v, want 1", reqAfter-reqBefore)
}
}
// TestProviderTunnelSidebandNonStreamingEmitsUsageMetrics verifies that a
// passthrough+sideband non-streaming response increments the request counter
// with status=success and usage_source=provider_reported (REVIEW_USAGE_METRIC-1).
func TestProviderTunnelSidebandNonStreamingEmitsUsageMetrics(t *testing.T) {
const rawToken = "sk-sideband-nonstream-token"
const edgeID = "edge-sideband-nonstream-metrics"
const model = "pool-model"
providerBody := `{"id":"cmpl-1","choices":[{"message":{"role":"assistant","content":"hi"}}]}`
frames := make(chan *iop.ProviderTunnelFrame, 5)
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_RESPONSE_START,
StatusCode: 200,
Headers: map[string]string{"Content-Type": "application/json"},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY, Body: []byte(providerBody)}
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_USAGE,
Usage: &iop.Usage{InputTokens: 7, OutputTokens: 4},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_END, End: true}
close(frames)
srv := NewServer(principalTokenCfg(rawToken, "ollama"), &fakeRunService{tunnelFrames: frames}, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: responseModePassthroughSideband,
}
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total success/provider_reported: got delta %v, want 1", reqAfter-reqBefore)
}
// Verify token counters also incremented.
for tokenType, want := range map[string]float64{
tokenTypeInput: 7,
tokenTypeOutput: 4,
} {
got := requestTokenValue(t, labels, tokenType)
if got != want {
t.Fatalf("token_type %s: got %v, want %v", tokenType, got, want)
}
}
}
// TestProviderTunnelSidebandStreamCallerCancelEmitsCancelMetric verifies that a
// passthrough+sideband streaming caller disconnect (pre-cancelled context) emits
// the cancel metric (REVIEW_USAGE_METRIC-1).
func TestProviderTunnelSidebandStreamCallerCancelEmitsCancelMetric(t *testing.T) {
const rawToken = "sk-sideband-cancel-token"
const edgeID = "edge-sideband-cancel"
const model = "pool-model"
// Create a cancelled context so the sideband stream detects caller disconnect.
ctx, cancel := context.WithCancel(context.Background())
cancel()
// Frame channel never closed to simulate an in-flight tunnel.
frames := make(chan *iop.ProviderTunnelFrame)
fake := &fakeRunService{tunnelFrames: frames, tunnelWaitTimeout: 3 * time.Second}
srv := NewServer(principalTokenCfg(rawToken, "ollama"), fake, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusCancel, usageSourceUnavailable,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"stream":true,
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
req = req.WithContext(ctx)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusCancel, usageSourceUnavailable,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total cancel: got delta %v, want 1", reqAfter-reqBefore)
}
}
// TestChatCompletionsDispatchFailureEmitsErrorMetric verifies that a tunnel
// dispatch failure increments the request counter with status=error
// (REVIEW_USAGE_METRIC-2).
func TestChatCompletionsDispatchFailureEmitsErrorMetric(t *testing.T) {
const rawToken = "sk-dispatch-fail-token"
const edgeID = "edge-dispatch-fail"
const model = "pool-model"
fake := &fakeRunService{tunnelErr: fmt.Errorf("node unavailable")}
srv := NewServer(principalTokenCfg(rawToken, "ollama"), fake, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthrough, usageStatusError, usageSourceUnavailable,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"messages":[{"role":"user","content":"hello"}]
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusBadGateway {
t.Fatalf("expected 502, got %d body=%s", w.Code, w.Body.String())
}
if len(fake.tunnelReqsSnapshot()) == 0 {
t.Fatal("expected tunnel dispatch attempt, got none")
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthrough, usageStatusError, usageSourceUnavailable,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total error: got delta %v, want 1", reqAfter-reqBefore)
}
}
// TestToolValidationRetryDispatchFailureEmitsErrorMetric verifies that a
// non-stream Chat tool-validation retry SubmitRun failure emits the error
// request metric exactly once and returns tool_validation_retry_error (REVIEW_USAGE_METRIC_RETRY-1).
func TestToolValidationRetryDispatchFailureEmitsErrorMetric(t *testing.T) {
const rawToken = "sk-retry-dispatch-fail-token"
const edgeID = "edge-retry-dispatch-fail"
const model = "retry-model"
fake := &fakeRunService{
eventRuns: []chan *iop.RunEvent{
bufferedRunEvents(&iop.RunEvent{
Type: "complete",
Metadata: map[string]string{
"finish_reason": "tool_calls",
runtimeMetadataOpenAIToolCalls: `[{"id":"call_bad","type":"function","function":{"name":"cmd","arguments":"{\"unexpected\":true}"}}]`,
},
}),
},
runIDs: []string{"run-bad"},
submitErrAfter: 2,
}
srv := NewServer(principalTokenCfg(rawToken, "ollama"), fake, nil)
srv.SetEdgeID(edgeID)
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, "normalized", usageStatusError, usageSourceUnavailable,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"retry-model",
"messages":[{"role":"user","content":"hello"}],
"tools":[{"type":"function","function":{"name":"cmd","parameters":{"type":"object","properties":{"commands":{"type":"array","items":{"type":"string"}}},"required":["commands"]}}}]}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if !strings.Contains(w.Body.String(), `"type":"tool_validation_retry_error"`) {
t.Fatalf("expected tool_validation_retry_error, got: %s", w.Body.String())
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, "normalized", usageStatusError, usageSourceUnavailable,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total error: got delta %v, want 1", reqAfter-reqBefore)
}
}
// TestBufferedStreamToolValidationRetryDispatchFailureEmitsErrorMetric verifies
// that a buffered stream Chat tool-validation retry SubmitRun failure emits
// the error request metric exactly once (REVIEW_USAGE_METRIC_RETRY-1).
func TestBufferedStreamToolValidationRetryDispatchFailureEmitsErrorMetric(t *testing.T) {
const rawToken = "sk-buf-retry-dispatch-fail-token"
const edgeID = "edge-buf-retry-dispatch"
const model = "buf-retry-model"
fake := &fakeRunService{
eventRuns: []chan *iop.RunEvent{
bufferedRunEvents(&iop.RunEvent{
Type: "complete",
Metadata: map[string]string{
"finish_reason": "tool_calls",
runtimeMetadataOpenAIToolCalls: `[{"id":"call_bad","type":"function","function":{"name":"cmd","arguments":"{\"unexpected\":true}"}}]`,
},
}),
},
runIDs: []string{"run-bad"},
submitErrAfter: 2,
}
srv := NewServer(principalTokenCfg(rawToken, ""), fake, nil)
srv.SetEdgeID(edgeID)
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, "normalized", usageStatusError, usageSourceUnavailable,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"buf-retry-model",
"stream":true,
"messages":[{"role":"user","content":"hello"}],
"tools":[{"type":"function","function":{"name":"cmd","parameters":{"type":"object","properties":{"commands":{"type":"array","items":{"type":"string"}}},"required":["commands"]}}}]}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
body := w.Body.String()
if !strings.Contains(body, `tool_validation_retry_error`) {
t.Fatalf("expected tool_validation_retry_error in stream, got: %s", body)
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, "normalized", usageStatusError, usageSourceUnavailable,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total error: got delta %v, want 1", reqAfter-reqBefore)
}
}
// TestProviderTunnelSidebandStreamBodyWriteFailureEmitsCancelMetric verifies
// that a passthrough+sideband streaming body write failure to the caller
// propagates cancel upstream and emits status=cancel (REVIEW_USAGE_METRIC_RETRY-2).
func TestProviderTunnelSidebandStreamBodyWriteFailureEmitsCancelMetric(t *testing.T) {
const rawToken = "sk-sideband-bodyfail-cancel"
const edgeID = "edge-sideband-bodyfail-cancel"
const model = "pool-bodyfail"
// Use a tunnel that emits a single body frame, then the response writer
// will fail on the second body frame write.
frames := make(chan *iop.ProviderTunnelFrame, 3)
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_RESPONSE_START,
StatusCode: http.StatusOK,
}
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY,
Body: []byte(`{"ok":true}`),
}
// Second body frame triggers write failure.
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY,
Body: []byte(`more`),
}
// Use a writer that fails on the second write.
fake := &fakeRunService{tunnelFrames: frames, tunnelWaitTimeout: 3 * time.Second}
srv := NewServer(principalTokenCfg(rawToken, "ollama"), fake, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served"}}})
cancelBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusCancel, usageSourceUnavailable,
))
errorBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusError, usageSourceUnavailable,
))
// Wrap the response writer so the second Write call fails.
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-bodyfail",
"stream":true,
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := &failingWriteRecorder{
ResponseRecorder: httptest.NewRecorder(),
failAfter: 2,
}
srv.routes().ServeHTTP(w, req)
cancelAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusCancel, usageSourceUnavailable,
))
errorAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusError, usageSourceUnavailable,
))
if cancelAfter-cancelBefore != 1 {
t.Fatalf("requests_total cancel: got delta %v, want 1", cancelAfter-cancelBefore)
}
if errorAfter-errorBefore != 0 {
t.Fatalf("requests_total error: got delta %v, want 0", errorAfter-errorBefore)
}
}
// TestProviderTunnelSidebandStreamingEmitsUsageMetricsBodyOnly verifies that a
// passthrough+sideband streaming response that contains usage ONLY in the body
// (no separate USAGE frame) still emits metrics with usage_source=provider_reported.
// This is a regression test for REVIEW_USAGE_METRIC-1: body-parsed usage must not
// be ignored when no proto USAGE frame is present.
func TestProviderTunnelSidebandStreamingEmitsUsageMetricsBodyOnly(t *testing.T) {
const rawToken = "sk-body-only-stream-token"
const edgeID = "edge-body-only-stream-metrics"
const model = "pool-model"
// BODY is SSE-formatted (matching actual provider tunnel stream): delta chunk
// followed by a usage chunk on the next line.
providerBody := `data: {"choices":[{"delta":{"content":"hi"}}]}
data: {"usage":{"prompt_tokens":10,"completion_tokens":5,"prompt_tokens_details":{"cached_tokens":2},"completion_tokens_details":{"reasoning_tokens":3},"total_tokens":20}}
data: [DONE]
`
frames := make(chan *iop.ProviderTunnelFrame, 5)
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_RESPONSE_START,
StatusCode: 200,
Headers: map[string]string{"Content-Type": "text/event-stream"},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY, Body: []byte(providerBody)}
// NOTE: No USAGE frame — usage is only in the body JSON.
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_END, End: true}
close(frames)
srv := NewServer(principalTokenCfg(rawToken, "ollama"), &fakeRunService{tunnelFrames: frames}, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: responseModePassthroughSideband,
}
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"stream":true,
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total success/provider_reported: got delta %v, want 1", reqAfter-reqBefore)
}
// Verify token counters were populated from body-only usage including
// reasoning and cached_input (REVIEW_REVIEW_REVIEW_USAGE_METRIC_RETRY).
for tokenType, want := range map[string]float64{
tokenTypeInput: 10,
tokenTypeOutput: 5,
tokenTypeReasoning: 3,
tokenTypeCachedInput: 2,
} {
got := requestTokenValue(t, labels, tokenType)
if got != want {
t.Fatalf("token_type %s: got %v, want %v", tokenType, got, want)
}
}
}
// TestProviderTunnelSidebandNonStreamingEmitsUsageMetricsBodyOnly verifies that a
// passthrough+sideband non-streaming response that contains usage ONLY in the body
// (no separate USAGE frame) still emits metrics with usage_source=provider_reported.
func TestProviderTunnelSidebandNonStreamingEmitsUsageMetricsBodyOnly(t *testing.T) {
const rawToken = "sk-body-only-nonstream-token"
const edgeID = "edge-body-only-nonstream-metrics"
const model = "pool-model"
// BODY contains OpenAI-compatible usage JSON including detail objects;
// no separate USAGE frame.
providerBody := `{"id":"cmpl-1","choices":[{"message":{"role":"assistant","content":"hi"}}],"usage":{"prompt_tokens":12,"completion_tokens":8,"prompt_tokens_details":{"cached_tokens":3},"completion_tokens_details":{"reasoning_tokens":4},"total_tokens":27}}`
frames := make(chan *iop.ProviderTunnelFrame, 5)
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_RESPONSE_START,
StatusCode: 200,
Headers: map[string]string{"Content-Type": "application/json"},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY, Body: []byte(providerBody)}
// NOTE: No USAGE frame — usage is only in the body JSON.
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_END, End: true}
close(frames)
srv := NewServer(principalTokenCfg(rawToken, "ollama"), &fakeRunService{tunnelFrames: frames}, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: responseModePassthroughSideband,
}
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total success/provider_reported: got delta %v, want 1", reqAfter-reqBefore)
}
// Verify token counters were populated from body-only usage including
// reasoning and cached_input (REVIEW_REVIEW_REVIEW_USAGE_METRIC_RETRY).
for tokenType, want := range map[string]float64{
tokenTypeInput: 12,
tokenTypeOutput: 8,
tokenTypeReasoning: 4,
tokenTypeCachedInput: 3,
} {
got := requestTokenValue(t, labels, tokenType)
if got != want {
t.Fatalf("token_type %s: got %v, want %v", tokenType, got, want)
}
}
}
// TestProviderTunnelSidebandStreamProtoOnlyReasoningCachedInput verifies that when
// only a proto USAGE frame is present (no body usage), the metric path preserves
// reasoning and cached_input from the proto (REVIEW_REVIEW_REVIEW_USAGE_METRIC_RETRY).
func TestProviderTunnelSidebandStreamProtoOnlyReasoningCachedInput(t *testing.T) {
const rawToken = "sk-proto-only-stream-token"
const edgeID = "edge-proto-only-stream"
const model = "pool-model"
// BODY contains no usage object.
providerBody := `data: {"choices":[{"delta":{"content":"hi"}}]}
data: [DONE]
`
frames := make(chan *iop.ProviderTunnelFrame, 5)
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_RESPONSE_START,
StatusCode: 200,
Headers: map[string]string{"Content-Type": "text/event-stream"},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY, Body: []byte(providerBody)}
// Proto USAGE frame carries full token breakdown including reasoning/cached_input.
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_USAGE,
Usage: &iop.Usage{InputTokens: 10, OutputTokens: 5, ReasoningTokens: 3, CachedInputTokens: 2},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_END, End: true}
close(frames)
srv := NewServer(principalTokenCfg(rawToken, "ollama"), &fakeRunService{tunnelFrames: frames}, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: responseModePassthroughSideband,
}
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"stream":true,
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total success/provider_reported: got delta %v, want 1", reqAfter-reqBefore)
}
// All four token types must be present from proto-only.
for tokenType, want := range map[string]float64{
tokenTypeInput: 10,
tokenTypeOutput: 5,
tokenTypeReasoning: 3,
tokenTypeCachedInput: 2,
} {
got := requestTokenValue(t, labels, tokenType)
if got != want {
t.Fatalf("token_type %s: got %v, want %v", tokenType, got, want)
}
}
}
// TestProviderTunnelSidebandResponseProtoOnlyReasoningCachedInput verifies that
// for non-streaming sideband, proto-only USAGE frame preserves reasoning and
// cached_input in metrics (REVIEW_REVIEW_REVIEW_USAGE_METRIC_RETRY).
func TestProviderTunnelSidebandResponseProtoOnlyReasoningCachedInput(t *testing.T) {
const rawToken = "sk-proto-only-nonstream-token"
const edgeID = "edge-proto-only-nonstream"
const model = "pool-model"
providerBody := `{"id":"cmpl-1","choices":[{"message":{"role":"assistant","content":"hi"}}]}`
frames := make(chan *iop.ProviderTunnelFrame, 5)
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_RESPONSE_START,
StatusCode: 200,
Headers: map[string]string{"Content-Type": "application/json"},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY, Body: []byte(providerBody)}
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_USAGE,
Usage: &iop.Usage{InputTokens: 7, OutputTokens: 4, ReasoningTokens: 2, CachedInputTokens: 1},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_END, End: true}
close(frames)
srv := NewServer(principalTokenCfg(rawToken, "ollama"), &fakeRunService{tunnelFrames: frames}, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: responseModePassthroughSideband,
}
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total success/provider_reported: got delta %v, want 1", reqAfter-reqBefore)
}
for tokenType, want := range map[string]float64{
tokenTypeInput: 7,
tokenTypeOutput: 4,
tokenTypeReasoning: 2,
tokenTypeCachedInput: 1,
} {
got := requestTokenValue(t, labels, tokenType)
if got != want {
t.Fatalf("token_type %s: got %v, want %v", tokenType, got, want)
}
}
}
// TestProviderTunnelSidebandStreamBodyAndProtoNoDoubleCount verifies that when
// both body usage and proto USAGE frame are present, the merge rule prevents
// double-counting input/output while preserving proto-only reasoning/cached_input
// (REVIEW_REVIEW_REVIEW_USAGE_METRIC_RETRY).
func TestProviderTunnelSidebandStreamBodyAndProtoNoDoubleCount(t *testing.T) {
const rawToken = "sk-body-and-proto-stream-token"
const edgeID = "edge-body-and-proto-stream"
const model = "pool-model"
// Body reports input/output but no reasoning/cached_input details.
providerBody := `data: {"choices":[{"delta":{"content":"hi"}}]}
data: {"usage":{"prompt_tokens":10,"completion_tokens":5,"total_tokens":15}}
data: [DONE]
`
frames := make(chan *iop.ProviderTunnelFrame, 5)
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_RESPONSE_START,
StatusCode: 200,
Headers: map[string]string{"Content-Type": "text/event-stream"},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY, Body: []byte(providerBody)}
// Proto frame carries same input/output plus additional reasoning/cached_input.
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_USAGE,
Usage: &iop.Usage{InputTokens: 10, OutputTokens: 5, ReasoningTokens: 3, CachedInputTokens: 2},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_END, End: true}
close(frames)
srv := NewServer(principalTokenCfg(rawToken, "ollama"), &fakeRunService{tunnelFrames: frames}, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: responseModePassthroughSideband,
}
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"stream":true,
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total success/provider_reported: got delta %v, want 1", reqAfter-reqBefore)
}
// Input/output should NOT be double-counted (merge rule: body wins).
// Reasoning/cached_input should come from proto (body has 0).
for tokenType, want := range map[string]float64{
tokenTypeInput: 10, // from body, NOT 10+10=20
tokenTypeOutput: 5, // from body, NOT 5+5=10
tokenTypeReasoning: 3, // from proto
tokenTypeCachedInput: 2, // from proto
} {
got := requestTokenValue(t, labels, tokenType)
if got != want {
t.Fatalf("token_type %s: got %v, want %v", tokenType, got, want)
}
}
}
// TestProviderTunnelSidebandResponseBodyAndProtoNoDoubleCount verifies the same
// merge rule for non-streaming sideband responses (REVIEW_REVIEW_REVIEW_USAGE_METRIC_RETRY).
func TestProviderTunnelSidebandResponseBodyAndProtoNoDoubleCount(t *testing.T) {
const rawToken = "sk-body-and-proto-nonstream-token"
const edgeID = "edge-body-and-proto-nonstream"
const model = "pool-model"
// Body reports input/output only.
providerBody := `{"id":"cmpl-1","choices":[{"message":{"role":"assistant","content":"hi"}}],"usage":{"prompt_tokens":12,"completion_tokens":8,"total_tokens":20}}`
frames := make(chan *iop.ProviderTunnelFrame, 5)
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_RESPONSE_START,
StatusCode: 200,
Headers: map[string]string{"Content-Type": "application/json"},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_BODY, Body: []byte(providerBody)}
// Proto frame carries same input/output plus reasoning/cached_input.
frames <- &iop.ProviderTunnelFrame{
Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_USAGE,
Usage: &iop.Usage{InputTokens: 12, OutputTokens: 8, ReasoningTokens: 5, CachedInputTokens: 4},
}
frames <- &iop.ProviderTunnelFrame{Kind: iop.ProviderTunnelFrameKind_PROVIDER_TUNNEL_FRAME_KIND_END, End: true}
close(frames)
srv := NewServer(principalTokenCfg(rawToken, "ollama"), &fakeRunService{tunnelFrames: frames}, nil)
srv.SetEdgeID(edgeID)
srv.SetModelCatalog([]config.ModelCatalogEntry{{ID: model, Providers: map[string]string{"prov-1": "served-model"}}})
labels := usageLabels{
edgeID: edgeID, principalRef: "user:alice", principalAlias: "alice", tokenRef: "iop-tok-alice",
modelGroup: model, endpoint: usageEndpointChatCompletions, responseMode: responseModePassthroughSideband,
}
reqBefore := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(`{
"model":"pool-model",
"messages":[{"role":"user","content":"hello"}],
"metadata":{"iop_response_mode":"passthrough+sideband"}
}`))
req.Header.Set("Authorization", "Bearer "+rawToken)
w := httptest.NewRecorder()
srv.routes().ServeHTTP(w, req)
if w.Code != http.StatusOK {
t.Fatalf("status: got %d body=%s", w.Code, w.Body.String())
}
reqAfter := testutil.ToFloat64(openAIRequestsTotal.WithLabelValues(
edgeID, "user:alice", "alice", "iop-tok-alice", model,
usageEndpointChatCompletions, responseModePassthroughSideband, usageStatusSuccess, usageSourceProviderReported,
))
if reqAfter-reqBefore != 1 {
t.Fatalf("requests_total success/provider_reported: got delta %v, want 1", reqAfter-reqBefore)
}
// Input/output from body only; reasoning/cached_input from proto.
for tokenType, want := range map[string]float64{
tokenTypeInput: 12, // from body, NOT 12+12=24
tokenTypeOutput: 8, // from body, NOT 8+8=16
tokenTypeReasoning: 5, // from proto
tokenTypeCachedInput: 4, // from proto
} {
got := requestTokenValue(t, labels, tokenType)
if got != want {
t.Fatalf("token_type %s: got %v, want %v", tokenType, got, want)
}
}
}
// failingWriteRecorder wraps httptest.ResponseRecorder and fails after N writes.
type failingWriteRecorder struct {
*httptest.ResponseRecorder
failAfter int
writes int
}
func (fw *failingWriteRecorder) Write(p []byte) (int, error) {
fw.writes++
if fw.writes > fw.failAfter {
return 0, fmt.Errorf("simulated write failure")
}
return fw.ResponseRecorder.Write(p)
}