CaptchaAI Callback URL Error Handling: Retry and Dead-Letter Patterns

Callbacks (pingbacks) eliminate polling, but they introduce a new failure mode: what happens when your server is down, returns an error, or times out when CaptchaAI tries to deliver the result? This tutorial covers patterns for handling callback failures without losing CAPTCHA solutions.

What Can Go Wrong

Failure Mode	Symptom	Result
Server down	CaptchaAI gets connection refused	Solution not delivered
Server returns 5xx	CaptchaAI receives error response	May not retry (depends on implementation)
Network timeout	CaptchaAI connection hangs	Solution potentially lost
Handler crashes	Request accepted but result not stored	Solution silently dropped

The solution: never rely solely on callbacks. Always have a fallback.

Pattern 1: Callback + Fallback Polling

The most reliable approach — accept callbacks when they arrive, but poll for any tasks that don't receive callbacks within a timeout.

Python

import os
import time
import threading
import requests
from flask import Flask, request

app = Flask(__name__)
API_KEY = os.environ["CAPTCHAAI_API_KEY"]

# Track task state
pending_tasks = {}  # task_id -> {"submitted_at": timestamp, "status": "pending"}
results = {}
lock = threading.Lock()


def submit_captcha(sitekey, pageurl, callback_url):
    """Submit with callback, but track for fallback polling."""
    resp = requests.post("https://ocr.captchaai.com/in.php", data={
        "key": API_KEY,
        "method": "userrecaptcha",
        "googlekey": sitekey,
        "pageurl": pageurl,
        "pingback": callback_url,
        "json": 1
    })
    data = resp.json()

    if data.get("status") == 1:
        task_id = data["request"]
        with lock:
            pending_tasks[task_id] = {
                "submitted_at": time.time(),
                "status": "pending"
            }
        return task_id
    return None


@app.route("/callback")
def captcha_callback():
    """Primary result delivery — CaptchaAI sends results here."""
    task_id = request.args.get("id")
    solution = request.args.get("code")

    with lock:
        results[task_id] = solution
        pending_tasks.pop(task_id, None)

    return "OK", 200


def fallback_poller():
    """Poll for any tasks that missed their callback."""
    while True:
        time.sleep(30)  # Check every 30 seconds

        with lock:
            stale_tasks = [
                tid for tid, info in pending_tasks.items()
                if time.time() - info["submitted_at"] > 120  # 2 min callback timeout
                and info["status"] == "pending"
            ]

        for task_id in stale_tasks:
            resp = requests.get("https://ocr.captchaai.com/res.php", params={
                "key": API_KEY,
                "action": "get",
                "id": task_id,
                "json": 1
            })
            data = resp.json()

            if data.get("status") == 1:
                with lock:
                    results[task_id] = data["request"]
                    pending_tasks.pop(task_id, None)
                print(f"Fallback poll recovered: {task_id}")
            elif data.get("request") != "CAPCHA_NOT_READY":
                # Permanent error — remove from pending
                with lock:
                    pending_tasks.pop(task_id, None)
                print(f"Task failed: {task_id} — {data.get('request')}")


# Start fallback poller in background
poller_thread = threading.Thread(target=fallback_poller, daemon=True)
poller_thread.start()

JavaScript

const express = require("express");
const axios = require("axios");

const app = express();
const API_KEY = process.env.CAPTCHAAI_API_KEY;

const pendingTasks = new Map(); // taskId -> { submittedAt, status }
const results = new Map();

async function submitCaptcha(sitekey, pageurl, callbackUrl) {
  const resp = await axios.post("https://ocr.captchaai.com/in.php", null, {
    params: {
      key: API_KEY,
      method: "userrecaptcha",
      googlekey: sitekey,
      pageurl: pageurl,
      pingback: callbackUrl,
      json: 1,
    },
  });

  if (resp.data.status === 1) {
    const taskId = resp.data.request;
    pendingTasks.set(taskId, {
      submittedAt: Date.now(),
      status: "pending",
    });
    return taskId;
  }
  return null;
}

// Primary callback endpoint
app.get("/callback", (req, res) => {
  const taskId = req.query.id;
  const solution = req.query.code;

  results.set(taskId, solution);
  pendingTasks.delete(taskId);

  res.sendStatus(200);
});

// Fallback poller
setInterval(async () => {
  const now = Date.now();
  const staleTasks = [];

  for (const [taskId, info] of pendingTasks) {
    if (now - info.submittedAt > 120000 && info.status === "pending") {
      staleTasks.push(taskId);
    }
  }

  for (const taskId of staleTasks) {
    try {
      const resp = await axios.get("https://ocr.captchaai.com/res.php", {
        params: { key: API_KEY, action: "get", id: taskId, json: 1 },
      });

      if (resp.data.status === 1) {
        results.set(taskId, resp.data.request);
        pendingTasks.delete(taskId);
        console.log(`Fallback recovered: ${taskId}`);
      } else if (resp.data.request !== "CAPCHA_NOT_READY") {
        pendingTasks.delete(taskId);
        console.log(`Task failed: ${taskId} — ${resp.data.request}`);
      }
    } catch (err) {
      console.error(`Poll error for ${taskId}: ${err.message}`);
    }
  }
}, 30000);

app.listen(3000);

Pattern 2: Dead-Letter Queue

When your callback handler processes a result but encounters an error (database down, validation failure), move the problem to a dead-letter queue instead of losing the data.

Python

import json
import os
import time
from pathlib import Path

DEAD_LETTER_DIR = Path("dead_letter")
DEAD_LETTER_DIR.mkdir(exist_ok=True)


@app.route("/callback")
def captcha_callback_with_dlq():
    task_id = request.args.get("id")
    solution = request.args.get("code")

    try:
        # Attempt normal processing
        store_result(task_id, solution)
        return "OK", 200
    except Exception as e:
        # Processing failed — save to dead-letter queue
        dead_letter = {
            "task_id": task_id,
            "solution": solution,
            "error": str(e),
            "received_at": time.time()
        }
        dlq_path = DEAD_LETTER_DIR / f"{task_id}.json"
        dlq_path.write_text(json.dumps(dead_letter))

        print(f"DLQ: {task_id} — {e}")
        return "OK", 200  # Still return 200 to CaptchaAI


def reprocess_dead_letters():
    """Retry processing dead-letter items."""
    for dlq_file in DEAD_LETTER_DIR.glob("*.json"):
        item = json.loads(dlq_file.read_text())

        try:
            store_result(item["task_id"], item["solution"])
            dlq_file.unlink()  # Remove after successful processing
            print(f"DLQ reprocessed: {item['task_id']}")
        except Exception:
            pass  # Leave in DLQ for next retry

JavaScript

const fs = require("fs");
const path = require("path");

const DLQ_DIR = path.join(__dirname, "dead_letter");
if (!fs.existsSync(DLQ_DIR)) fs.mkdirSync(DLQ_DIR);

app.get("/callback-dlq", (req, res) => {
  const taskId = req.query.id;
  const solution = req.query.code;

  try {
    storeResult(taskId, solution);
    res.sendStatus(200);
  } catch (err) {
    // Save to dead-letter queue
    const deadLetter = {
      task_id: taskId,
      solution: solution,
      error: err.message,
      received_at: Date.now(),
    };

    fs.writeFileSync(
      path.join(DLQ_DIR, `${taskId}.json`),
      JSON.stringify(deadLetter)
    );

    console.log(`DLQ: ${taskId} — ${err.message}`);
    res.sendStatus(200); // Still acknowledge to CaptchaAI
  }
});

function reprocessDeadLetters() {
  const files = fs.readdirSync(DLQ_DIR).filter((f) => f.endsWith(".json"));

  for (const file of files) {
    const filePath = path.join(DLQ_DIR, file);
    const item = JSON.parse(fs.readFileSync(filePath, "utf8"));

    try {
      storeResult(item.task_id, item.solution);
      fs.unlinkSync(filePath);
      console.log(`DLQ reprocessed: ${item.task_id}`);
    } catch (err) {
      // Leave in DLQ
    }
  }
}

// Retry DLQ every 5 minutes
setInterval(reprocessDeadLetters, 300000);

Pattern 3: Idempotent Callback Handler

Callbacks might be delivered more than once. Make your handler idempotent:

@app.route("/callback")
def idempotent_callback():
    task_id = request.args.get("id")
    solution = request.args.get("code")

    with lock:
        # Only process if not already handled
        if task_id in results:
            return "OK", 200  # Already processed — skip silently

        results[task_id] = solution
        pending_tasks.pop(task_id, None)

    return "OK", 200

Decision Matrix: Which Pattern to Use

Scenario	Best Pattern
Low volume, occasional downtime	Callback + Fallback Polling
High volume, database outages possible	Dead-Letter Queue
Multiple consumers might process same result	Idempotent Handler
Production system with SLAs	All three combined

Troubleshooting

Issue	Cause	Fix
Fallback poller finds tasks that were already delivered	Race between callback and poller	Add idempotency check — skip if already in results
DLQ growing without being processed	Reprocessor not running or failing	Check reprocessor logs; ensure underlying issue (DB) is fixed
Callback returns 200 but result is lost	Handler crashes after response is sent	Process before responding, or use DLQ pattern
Too many fallback poll requests	Too many stale tasks	Increase callback timeout threshold; check server uptime

FAQ

Should I always return 200 to CaptchaAI callbacks?

Yes. Returning an error code (4xx/5xx) doesn't help — CaptchaAI may not retry callbacks. Always accept the delivery (200 OK) and handle failures internally with DLQ or fallback polling.

How long should I wait before fallback polling?

Wait at least 120 seconds after submission. Most CAPTCHAs solve within 10–60 seconds, plus network latency for the callback delivery. Two minutes gives ample time for the callback to arrive.

Can I disable callbacks and just poll?

Yes — simply don't include the pingback parameter. But callbacks reduce API calls significantly at scale (2 calls per task instead of 10+ poll requests).