What causes "Document parsing result page is missing markdown.text." in PaddlePaddle/PaddleOCR?

Retry the status call with a small backoff — read-only and idempotent

PaddlePaddle/PaddleOCR · warning · ResultParseError

Document parsing result page is missing markdown.text.

Error message

Document parsing result page is missing markdown.text.

What it means

Raised as NetworkError from get_job_status in paddleocr/_api_client/_http.py:168 when the GET for a job's status fails at the connection layer (DNS, TLS, reset, refused). The job keeps running server-side; only the observation request failed. Transient by nature — retrying the same status call is safe and idempotent.

Source

Thrown at api_sdk/typescript/src/client.ts:312

          raw: item,
        };
      });
    });
    return { jobId, pages, dataInfo };
  }

  private parseDocParsingResult(jobId: string, jsonlData: unknown[]): DocParsingResult {
    const dataInfo: Record<string, unknown> = {};
    const pages = jsonlData.flatMap((lineObj) => {
      if (!isRecord(lineObj) || !isRecord(lineObj.result) || !Array.isArray(lineObj.result.layoutParsingResults)) {
        throw new ResultParseError("Document parsing result item is missing result.layoutParsingResults.");
      }
      if (isRecord(lineObj.result.dataInfo)) {
        Object.assign(dataInfo, lineObj.result.dataInfo);
      }
      return lineObj.result.layoutParsingResults.map((item) => {
        if (!isRecord(item) || !isRecord(item.markdown) || typeof item.markdown.text !== "string") {
          throw new ResultParseError("Document parsing result page is missing markdown.text.");
        }
        return {
          markdownText: item.markdown.text,
          markdownImages: isRecord(item.markdown.images) ? stringMap(item.markdown.images) : {},
          outputImages: isRecord(item.outputImages) ? stringMap(item.outputImages) : {},
          prunedResult: item.prunedResult,
          inputImageUrl: typeof item.inputImage === "string" ? item.inputImage : undefined,
          exports: isRecord(item.exports) ? item.exports : {},
          markdown: item.markdown,
          raw: item,
        };
      });
    });
    return { jobId, pages, dataInfo };
  }

  private resolveJob(job: Job | string, expectedTask: Job["task"]): Job {
    if (typeof job === "string") {

View on GitHub (pinned to 2661c7c0ef)

Solutions

Retry the status call with a small backoff — read-only and idempotent
If keepalive resets recur between long poll intervals, retry transparently or use a fresh session
Verify general connectivity if failures persist
Keep the jobId persisted so polling can resume after process restarts

Example fix

# backoff retry for idempotent status reads
import time
from paddleocr._api_client.errors import NetworkError
for delay in (1, 2, 4, 8):
    try:
        status = client.get_job_status(job_id)
        break
    except NetworkError:
        time.sleep(delay)
else:
    raise

Defensive patterns

Strategy: retry

Try / catch

from paddleocr._api_client.errors import NetworkError
for delay in (1, 2, 4, 8):
    try:
        status = client.get_job_status(job_id)
        return
    except NetworkError:
        time.sleep(delay)
raise

Prevention

Retry idempotent status reads with backoff instead of failing the whole job flow
Persist jobId so polling resumes after connectivity recovers
Expect keepalive resets on long poll sessions and handle them

When it happens

Trigger: Polling get_job_status during a network blip, VPN reconnect, DNS hiccup, or when a proxy drops idle keepalive connections between polls.

Common situations: Long-running jobs polled for minutes over unreliable networks; laptop sleep/resume mid-poll; container DNS flakiness; proxies closing idle sessions.

Related errors

AI-assisted analysis of PaddlePaddle/PaddleOCR@2661c7c0ef (2026-08-14). Data as JSON: /api/errors/b5459863596f810b. Report an issue: GitHub.