metrics: stop counting client-cancelled requests as server errors (#1255)

## Summary

- Remap recorded status to `499` (NGINX-style "client closed") when
`ctx.Context().Err()` is `context.Canceled` after the handler returns
- Skip the `http_errors_total` increment for `499`s so the availability
metric reflects server-visible errors only

## Why

Under scraper-driven load, clients with short timeouts paginate
`/v0.1/servers`, give up while PG is busy, and close the TCP connection.
The handler's DB iteration returns `context canceled`, the handler
converts that to `huma.Error500InternalServerError`, and tries to write
a response to the now-closed socket. NGINX records `499` (client closed)
and never delivers anything to the client — but the registry's
middleware records `status_code=500` and bumps
`mcp_registry_http_errors_total`.

That's what's been firing the `Availability dropped below 95%` alert
during the daily 17:00 UTC bursts. Recent prod numbers from one pod
(~19h uptime):

```
mcp_registry_http_errors_total{path="/v0.1/servers", status_code="500"} 7396
mcp_registry_http_errors_total{path="/v0.1/servers/{serverName}/versions", status_code="500"} 283
mcp_registry_http_errors_total{path="/v0/servers", status_code="500"} 352
```

…against zero 5xx in NGINX ingress logs over the same window. The alert
was correct *given the data it had*; the data was misclassifying client
cancellations as server errors.

## What this does

In `internal/api/router/router.go` (the metric middleware): after
`next(ctx)`, check `ctx.Context().Err()`. If it's `context.Canceled`,
override `statusCode` to `499` and skip the error counter. Requests
counter and duration histogram still record so the cancellation rate
stays visible.

`context.DeadlineExceeded` is intentionally **not** remapped — that
would indicate a server-side timeout if we ever add per-request
deadlines, and should still count as a server error.

## What this does *not* do

- Doesn't reduce the underlying load (ServiceNow + notion catalog
scrapers driving the daily peak — separate work, see the `updated_since`
outreach we're doing).
- Doesn't change handler behaviour — handlers still call
`huma.Error500…` on `context.Canceled`. We could short-circuit earlier
(return without writing) to save the work, but that's a bigger refactor
and the metric fix unblocks the alert immediately.

## Companion change (no code, just a Grafana annotation tweak)

The `Availability dropped below 95%` alert (UID `bexrc60etvhmoa`)
currently has the annotation:

> No. of 5xx are increased from acceptable limit i.e. 5% of total
throughput

The query is on `mcp_registry_http_errors_total`, not 5xx specifically.
After this change the count is meaningful again, but the wording is
still misleading. Suggest updating to:

> Internal error rate exceeded 5% of total throughput

(no PromQL change required.)

## Test plan

- [x] `go test ./internal/api/router/...` — new test passes
- [x] `golangci-lint run ./internal/...` — clean
- [ ] After merge, watch a daily 17:00 UTC burst —
`mcp_registry_http_errors_total{status_code="500",
path="/v0.1/servers"}` should stay flat while
`mcp_registry_http_requests_total{status_code="499",
path="/v0.1/servers"}` grows. Availability alert should not fire on
cancellation-only bursts.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: rdimitrov

internal/api/router/router.go

@@ -2,7 +2,9 @@
 package router
 
 import (
+	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"net/http"
 	"strings"
@@ -19,6 +21,12 @@ import (
 	"github.com/modelcontextprotocol/registry/internal/telemetry"
 )
 
+// statusClientClosed mirrors NGINX's non-standard 499 — used for requests
+// where the client disconnected before we finished. Distinguishing these from
+// real server errors keeps the availability metric meaningful under bursts of
+// scraper traffic that time out and reconnect.
+const statusClientClosed = 499
+
 // Middleware configuration options
 type middlewareConfig struct {
 	skipPaths map[string]bool
@@ -67,6 +75,22 @@ func MetricTelemetryMiddleware(metrics *telemetry.Metrics, options ...Middleware
 		duration := time.Since(start).Seconds()
 		statusCode := ctx.Status()
 
+		// If the client disconnected before the handler finished, the handler
+		// likely converted the resulting context.Canceled into a huma 5xx and
+		// tried to write a response to a closed socket. NGINX records that
+		// case as a 499 (client closed). Without this remap we count it as a
+		// server error: a single ServiceNow-style burst that times out a
+		// few thousand list-servers requests inflates http_errors_total even
+		// though no client ever saw a 5xx, and the availability alert fires
+		// on what is effectively just slow responses.
+		//
+		// Only context.Canceled is remapped — context.DeadlineExceeded would
+		// indicate a server-side timeout we set ourselves and should still
+		// count as a server error if/when we add per-request deadlines.
+		if reqErr := ctx.Context().Err(); reqErr != nil && errors.Is(reqErr, context.Canceled) {
+			statusCode = statusClientClosed
+		}
+
 		// Combine common and custom attributes
 		attrs := []attribute.KeyValue{
 			attribute.String("method", method),
@@ -77,7 +101,9 @@ func MetricTelemetryMiddleware(metrics *telemetry.Metrics, options ...Middleware
 		// Record metrics
 		metrics.Requests.Add(ctx.Context(), 1, metric.WithAttributes(attrs...))
 
-		if statusCode >= 400 {
+		// Skip the error counter for client-closed requests so the availability
+		// metric reflects server-visible errors only.
+		if statusCode >= 400 && statusCode != statusClientClosed {
 			metrics.ErrorCount.Add(ctx.Context(), 1, metric.WithAttributes(attrs...))
 		}
 

internal/api/router/router_test.go

@@ -0,0 +1,119 @@
+package router_test
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/danielgtaylor/huma/v2"
+	"github.com/danielgtaylor/huma/v2/adapters/humago"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/modelcontextprotocol/registry/internal/api/router"
+	"github.com/modelcontextprotocol/registry/internal/telemetry"
+)
+
+// TestMetricMiddleware_ClientCancelledNotCountedAsError verifies that a request
+// the client gave up on (request-context Canceled) is recorded as 499 in
+// http_requests_total and is NOT incremented in http_errors_total — even when
+// the handler returned huma.Error500InternalServerError because its DB call
+// surfaced context.Canceled.
+//
+// Regression test for the case where a ServiceNow-style scraper burst
+// generated thousands of internal 500s for cancelled requests, tripping the
+// "Availability dropped below 95%" alert despite zero 5xx reaching clients.
+func TestMetricMiddleware_ClientCancelledNotCountedAsError(t *testing.T) {
+	shutdown, metrics, err := telemetry.InitMetrics("test")
+	require.NoError(t, err)
+	defer func() { _ = shutdown(context.Background()) }()
+
+	mux := http.NewServeMux()
+	api := humago.New(mux, huma.DefaultConfig("Test API", "1.0.0"))
+	api.UseMiddleware(router.MetricTelemetryMiddleware(metrics))
+
+	// Handler that simulates: DB returns context.Canceled because the client
+	// disconnected, handler converts to huma 500.
+	huma.Register(api, huma.Operation{
+		OperationID: "cancelled-list",
+		Method:      http.MethodGet,
+		Path:        "/v0/servers",
+	}, func(_ context.Context, _ *struct{}) (*struct{}, error) {
+		return nil, huma.Error500InternalServerError("Failed to get registry list",
+			errors.New("error iterating rows: context canceled"))
+	})
+
+	mux.Handle("/metrics", metrics.PrometheusHandler())
+
+	cancelledCtx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	req := httptest.NewRequest(http.MethodGet, "/v0/servers", nil).WithContext(cancelledCtx)
+	rec := httptest.NewRecorder()
+	mux.ServeHTTP(rec, req)
+
+	scrapeReq := httptest.NewRequest(http.MethodGet, "/metrics", nil)
+	scrapeRec := httptest.NewRecorder()
+	mux.ServeHTTP(scrapeRec, scrapeReq)
+	assert.Equal(t, http.StatusOK, scrapeRec.Code)
+
+	body := scrapeRec.Body.String()
+
+	// The request is recorded — but as 499, not 500.
+	assert.True(t,
+		containsMetric(body, `mcp_registry_http_requests_total`, `status_code="499"`, `path="/v0/servers"`),
+		"expected requests_total to record status_code=499 for the cancelled request; got:\n%s",
+		filterMetricLines(body, "mcp_registry_http_requests_total"),
+	)
+	assert.False(t,
+		containsMetric(body, `mcp_registry_http_requests_total`, `status_code="500"`, `path="/v0/servers"`),
+		"did not expect requests_total to record status_code=500 for a client-cancelled request; got:\n%s",
+		filterMetricLines(body, "mcp_registry_http_requests_total"),
+	)
+
+	// The error counter is NOT bumped for 499s — that is the point of this fix.
+	assert.False(t,
+		containsMetric(body, `mcp_registry_http_errors_total`, `status_code="499"`),
+		"errors_total must not be incremented for client-cancelled requests; got:\n%s",
+		filterMetricLines(body, "mcp_registry_http_errors_total"),
+	)
+	assert.False(t,
+		containsMetric(body, `mcp_registry_http_errors_total`, `status_code="500"`, `path="/v0/servers"`),
+		"errors_total must not record a 500 for a request the client already gave up on; got:\n%s",
+		filterMetricLines(body, "mcp_registry_http_errors_total"),
+	)
+}
+
+// containsMetric returns true iff some line in body starts with `metric{...}`
+// and that label set contains every requested label fragment.
+func containsMetric(body, metric string, labels ...string) bool {
+	for _, line := range strings.Split(body, "\n") {
+		if !strings.HasPrefix(line, metric+"{") {
+			continue
+		}
+		ok := true
+		for _, l := range labels {
+			if !strings.Contains(line, l) {
+				ok = false
+				break
+			}
+		}
+		if ok {
+			return true
+		}
+	}
+	return false
+}
+
+func filterMetricLines(body, metric string) string {
+	var lines []string
+	for _, line := range strings.Split(body, "\n") {
+		if strings.HasPrefix(line, metric) {
+			lines = append(lines, line)
+		}
+	}
+	return strings.Join(lines, "\n")
+}