What causes "err.Error() (Fprintf write failure)" in vitessio/vitess?

Retry the debug request with a stable connection

err.Error() (Fprintf write failure)

Error message

err.Error() (Fprintf write failure)

What it means

After executing the stats template, WriteScatterStats appends a summary line via fmt.Fprintf to the response. If that write fails (client gone, connection reset), the raw error is passed to http.Error with a 500. Because headers/body are already sent, the 500 cannot change the status code; the error text is effectively only visible in logs or as garbage in the stream.

Source

Thrown at go/vt/vtgate/executor_scatter_stats.go:138

		http.Error(w, err.Error(), 500)
		return
	}

	t := template.New("template")
	t, err = t.Parse(statsHTML)
	if err != nil {
		http.Error(w, err.Error(), 500)
		return
	}

	err = t.Execute(w, results)
	if err != nil {
		http.Error(w, err.Error(), 500)
	}

	_, err = fmt.Fprintf(w, "Percentage of time spent on scatter queries: %2.2f%%", results.PercentTimeScatter)
	if err != nil {
		http.Error(w, err.Error(), 500)
	}
}

const statsHTML = `
<thead>
	<tr>
		<th>Query</th>
		<th># of executions</th>
		<th>Avg time/query</th>
		<th>% time of reads</th>
		<th>% time of scatters</th>
		<th>% of reads</th>
		<th>% of scatters</th>
	</tr>
</thead>
{{range .Items}}
<tr class="medium">
	<td>{{.Query}}</td>

View on GitHub (pinned to 01a25a7d17)

Solutions

Retry the debug request with a stable connection
Check vtgate logs for the write error; treat client-disconnect errors as benign noise
Increase any client/proxy read timeouts that cut the response short

Defensive patterns

Strategy: fallback

Validate before calling

resp, err := http.Get("http://vtgate:15001/scatterStats")
if err != nil {
	// connection failed before write; use fallback stats source
}

Try / catch

resp, err := http.Get(url)
if err != nil {
	return err
}
defer resp.Body.Close()
body, err := io.ReadAll(resp.Body)
if err != nil {
	log.Warn("partial scatterStats read", slog.Any("error", err))
}

Prevention

Use adequate client timeouts so the response finishes streaming
Reconnect-and-retry scrapers instead of dropping mid-response
Treat trailing-write errors on already-committed responses as benign
Pull PercentTimeScatter from metrics endpoints when available

When it happens

Trigger: Calling the /scatterStats debug endpoint when the HTTP connection is broken or closed before the final Fprintf write completes.

Common situations: Client (curl/browser/monitoring scraper) disconnects during streaming of the stats page; response already committed so http.Error is a no-op on status.

Related errors

AI-assisted analysis of vitessio/vitess@01a25a7d17 (2026-09-01). Data as JSON: /api/errors/e8a37d846565e4bd. Report an issue: GitHub.