What causes "poll request failed: %w" in micro/go-micro?

Make poll failures retryable: tolerate N consecutive poll errors before aborting, instead of returning on the first failure.

micro/go-micro · error

poll request failed: %w

Error message

poll request failed: %w

What it means

pollPrediction wraps transport errors from http.DefaultClient.Do when polling GET /api/v1/model/prediction/{id} for image-generation status. A single failed poll aborts the whole GenerateImage flow, even though prediction polling is inherently retry-friendly. The wrapped error preserves the underlying cause (timeout, reset, cancellation).

Source

Thrown at ai/atlascloud/atlascloud.go:1013

				return nil, err
			}
			if result != nil {
				return result, nil
			}
		}
	}
}

func (p *Provider) pollPrediction(ctx context.Context, url string) (*ai.ImageResponse, error) {
	httpReq, err := http.NewRequestWithContext(ctx, http.MethodGet, url, nil)
	if err != nil {
		return nil, err
	}
	httpReq.Header.Set("Authorization", "Bearer "+p.opts.APIKey)

	httpResp, err := http.DefaultClient.Do(httpReq)
	if err != nil {
		return nil, fmt.Errorf("poll request failed: %w", err)
	}
	defer httpResp.Body.Close()

	body, _ := io.ReadAll(httpResp.Body)

	var pollResp struct {
		Data struct {
			Status  string   `json:"status"`
			Outputs []string `json:"outputs"`
			Error   string   `json:"error"`
		} `json:"data"`
	}
	if err := json.Unmarshal(body, &pollResp); err != nil {
		return nil, fmt.Errorf("failed to parse poll response: %w", err)
	}

	switch pollResp.Data.Status {
	case "completed":

View on GitHub (pinned to 24529f1404)

Solutions

Make poll failures retryable: tolerate N consecutive poll errors before aborting, instead of returning on the first failure.
Check errors.Is(err, context.Canceled)/DeadlineExceeded to distinguish caller cancellation from real network faults.
Increase per-request timeout or keepalive settings if polls time out under load.
Verify the BaseURL prediction path is stable/reachable; add exponential backoff between retries.

Example fix

// before
if err != nil {
    return nil, fmt.Errorf("poll request failed: %w", err)
}
// after
if err != nil {
    if errors.Is(err, context.Canceled) {
        return nil, fmt.Errorf("poll request failed: %w", err)
    }
    if consecutiveFails++; consecutiveFails <= 5 {
        continue // tolerate transient poll failures
    }
    return nil, fmt.Errorf("poll request failed: %w", err)
}

Defensive patterns

Strategy: retry

Try / catch

img, err := provider.GenerateImage(ctx, req)
if err != nil {
    if errors.Is(err, context.Canceled) {
        return err // caller gave up; do not retry
    }
    if strings.Contains(err.Error(), "poll request failed") {
        var nerr net.Error
        if errors.As(err, &nerr) && nerr.Timeout() {
            // resume polling with the same prediction flow or retry
        }
    }
    return err
}

Prevention

Treat individual poll failures as transient; only abort after several consecutive failures.
Size the context deadline to cover the full polling window (many 2s ticks), not a single request.
Never cancel the polling context prematurely if the image is still needed.
Add backoff between polls and jitter to avoid hammering a struggling endpoint.

When it happens

Trigger: Any poll tick's http.Do fails: transient connection reset, ctx cancelled (e.g. caller gave up or a deadline set on the polling ctx expired), DNS blip, or the prediction endpoint being temporarily unavailable while the prediction is still processing.

Common situations: Long-running image predictions outliving the client timeout on individual polls, network instability during a 2-second-interval poll loop, caller cancelling the context after losing interest, or prediction ID endpoint returning intermittent 5xx at transport level.

Related errors

AI-assisted analysis of micro/go-micro@24529f1404 (2026-09-01). Data as JSON: /api/errors/74adccf4d8975613. Report an issue: GitHub.