What causes "state=done but no jsonUrl" in Tencent/WeKnora?

Retry Read once or twice — transient provider inconsistency often resolves on a subsequent task submission.

state=done but no jsonUrl

Error message

state=done but no jsonUrl

What it means

pollJob in the PaddleOCR VL cloud converter polls the remote parsing task and, when the API reports state="done", expects a result JSONURL in the response. This error is thrown when the provider marks the task done but omits ResultURL.JSONURL, so the converter has no result to download. It is a defensive guard against a malformed/incomplete provider success response rather than a local bug.

Source

Thrown at internal/infrastructure/docparser/paddleocr_vl_cloud_converter.go:238

		}

		var pollResp paddleOCRVLCloudPollResponse
		if err := json.Unmarshal(respBody, &pollResp); err != nil {
			logger.Errorf(context.Background(), "[PaddleOCR-VL Cloud] poll #%d decode error: %v", pollCount, err)
			sleepCtx(ctx, paddleOCRVLCloudPollInterval)
			continue
		}

		state := strings.ToLower(pollResp.Data.State)
		if pollCount == 1 || pollCount%6 == 0 || state == "done" || state == "failed" {
			logger.Infof(context.Background(), "[PaddleOCR-VL Cloud] poll #%d: state=%s pages=%d/%d",
				pollCount, state, pollResp.Data.ExtractProgress.ExtractedPages, pollResp.Data.ExtractProgress.TotalPages)
		}

		switch state {
		case "done":
			if pollResp.Data.ResultURL.JSONURL == "" {
				return "", fmt.Errorf("state=done but no jsonUrl")
			}
			return pollResp.Data.ResultURL.JSONURL, nil
		case "failed":
			return "", fmt.Errorf("task failed: %s", pollResp.Data.ErrorMsg)
		}

		sleepCtx(ctx, paddleOCRVLCloudPollInterval)
	}

	return "", fmt.Errorf("task timed out after %d polls", pollCount)
}

// --- result parsing ---

type paddleOCRVLCloudResultLine struct {
	Result struct {
		LayoutParsingResults []struct {
			Markdown struct {

View on GitHub (pinned to 988cbb0330)

Solutions

Retry Read once or twice — transient provider inconsistency often resolves on a subsequent task submission.
Log the full raw poll response body to confirm the JSONURL field path and whether the API contract changed.
Check the provider's API docs/changelog for a renamed result field (e.g. resultUrl vs resultURL) and update the response struct JSON tags.
Verify the account/plan still has quota and that result artifacts are not being purged by the provider.
If persistent, treat the task as failed and resubmit the document, or fall back to a local parser.

Example fix

// before
resultURL, err := reader.Read(ctx, file)
if err != nil {
    return err // state=done but no jsonUrl
}

// after
resultURL, err := reader.Read(ctx, file)
if err != nil {
    if strings.Contains(err.Error(), "no jsonUrl") {
        resultURL, err = reader.Read(ctx, file) // retry transient provider glitch
    }
    if err != nil {
        return fmt.Errorf("paddleocr cloud parse: %w", err)
    }
}

Defensive patterns

Strategy: retry

Validate before calling

// before calling Read, ensure the provider result contract is current
// (field-level validation happens server-side; here we guard the caller)
if reader == nil {
    return errors.New("paddleocr cloud reader not configured")
}

Try / catch

url, err := reader.Read(ctx, file)
if err != nil && strings.Contains(err.Error(), "no jsonUrl") {
    url, err = reader.Read(ctx, file) // one retry for provider glitch
}
if err != nil { return err }

Prevention

Pin and monitor the provider API version; watch changelogs for result-field renames.
Log raw poll responses at debug level to catch contract drift early.
Alert on this error rate — spikes indicate provider-side incidents, not your data.
Keep a local OCR parser as a fallback for critical documents.

When it happens

Trigger: pollJob receives a poll response whose data.state is "done" while pollResp.Data.ResultURL.JSONURL is the empty string; called from Read after submitting a document for cloud parsing.

Common situations: Provider-side anomalies: the OCR task finished but result generation/upload failed silently; an API version change renamed or moved the result-URL field; a proxy or gateway stripped the JSON body; expired/limited account where the platform drops result artifacts but still flips state to done.

Related errors

AI-assisted analysis of Tencent/WeKnora@988cbb0330 (2026-09-02). Data as JSON: /api/errors/e8023b6f3d8b9d9c. Report an issue: GitHub.