frontend-api-integration-patterns – skill Thiết kế & UI cho agent AI

Frontend API Integration Patterns

Overview

This skill provides production-ready patterns for integrating frontend applications with backend APIs.

Most frontend issues are not caused by APIs being difficult to call, but by incorrect handling of asynchronous behavior—leading to race conditions, stale data, duplicated requests, and poor user experience.

This skill focuses on correctness, resilience, and user experience, not just making API calls work.

---

When to Use This Skill

• Connecting frontend apps (React, React Native, Vue, etc.) to backend APIs
• Integrating ML/AI endpoints (/predict, /recommend)
• Handling asynchronous data in UI
• Fixing stale data, flickering UI, or duplicate requests
• Designing scalable frontend API layers

---

Core Patterns

1. API Layer (Separation of Concerns)

Centralize API logic and normalize errors.

export class ApiError extends Error {
  constructor(message, status, payload = null) {
    super(message);
    this.name = "ApiError";
    this.status = status;
    this.payload = payload;
  }
}

export const apiClient = async (url, options = {}) => {
  const res = await fetch(url, {
    headers: { "Content-Type": "application/json" },
    ...options,
  });

  if (!res.ok) {
    let payload = null;
    try {
      payload = await res.json();
    } catch (_) {}

    throw new ApiError(
      payload?.message || "Request failed",
      res.status,
      payload
    );
  }

  // handle empty responses safely (e.g. 204 No Content)
  if (res.status === 204) return null;

  const text = await res.text();
  return text ? JSON.parse(text) : null;
};

---

2. Race-Safe State Management

Prevent stale responses from overwriting fresh data.

useEffect(() => {
  let cancelled = false;

  const load = async () => {
    try {
      setLoading(true);
      setError(null);

      const result = await getUser();

      if (!cancelled) setData(result);
    } catch (err) {
      if (!cancelled) setError(err.message);
    } finally {
      if (!cancelled) setLoading(false);
    }
  };

  load();

  return () => {
    cancelled = true;
  };
}, []);

> Use a cancellation flag for non-fetch async logic. For network requests, prefer AbortController.

---

3. Request Cancellation (AbortController)

Cancel in-flight requests to avoid memory leaks and stale updates.

useEffect(() => {
  const controller = new AbortController();

  const load = async () => {
    try {
      const data = await getUser({ signal: controller.signal });
      setData(data);
    } catch (err) {
      if (err.name === "AbortError") return;
      setError(err.message);
    }
  };

  load();
  return () => controller.abort();
}, [userId]);

---

4. Retry with Exponential Backoff

Retry only explicitly replay-safe operations and classified transient failures. The default performs one attempt. Do not classify arbitrary programming or parsing exceptions as network errors. Honor provider Retry-After and a total deadline in the real adapter; do not enable this simple backoff for rate-limit responses. The POST prediction example intentionally does not enable retries unless the backend contract independently guarantees replay safety.

const sleep = (ms) => new Promise((r) => setTimeout(r, ms));

const fetchWithBackoff = async (fn, retries = 3, delay = 300, replaySafe = false) => {
  try {
    return await fn();
  } catch (err) {
    const isAbort = err.name === "AbortError";
    const isHttpError = typeof err.status === "number";
    const isNetworkFailure = err.code === "NETWORK_ERROR";
    const isRetryable = replaySafe && !isAbort &&
      (isNetworkFailure || (isHttpError && err.status >= 500));

    if (retries <= 0 || !isRetryable) throw err;

    const nextDelay = delay * 2 + Math.random() * 100;
    await sleep(nextDelay);

    return fetchWithBackoff(fn, retries - 1, nextDelay, replaySafe);
  }
};

---

5. Debounced API Calls

Avoid excessive API calls (e.g., search inputs).

const useDebounce = (value, delay = 400) => {
  const [debounced, setDebounced] = useState(value);

  useEffect(() => {
    const t = setTimeout(() => setDebounced(value), delay);
    return () => clearTimeout(t);
  }, [value, delay]);

  return debounced;
};

---

6. Request Deduplication

Prevent duplicate API calls across components.

const inFlight = new Map();

export const dedupedFetch = (key, fn) => {
  if (inFlight.has(key)) return inFlight.get(key);

  const promise = fn().finally(() => inFlight.delete(key));
  inFlight.set(key, promise);
  return promise;
};

---

Examples

Example 1: ML Prediction with Cancellation

const controllerRef = useRef(null);

const handlePredict = async (input) => {
  controllerRef.current?.abort();
  const controller = new AbortController();
  controllerRef.current = controller;

  try {
    const result = await fetchWithBackoff(() =>
      apiClient("/predict", {
        method: "POST",
        body: JSON.stringify({ text: input }),
        signal: controller.signal,
      })
    );

    if (!controller.signal.aborted && controllerRef.current === controller) setOutput(result);
  } catch (err) {
    if (err.name === "AbortError") return;
    if (controllerRef.current === controller) setError(err.message);
  }
};

---

Example 2: Debounced Search

const debouncedQuery = useDebounce(query, 400);

useEffect(() => {
  if (!debouncedQuery) return;

  const controller = new AbortController();

  searchAPI(debouncedQuery, { signal: controller.signal })
    .then(setResults)
    .catch((err) => {
      if (err.name !== "AbortError") {
        setError("Search failed. Please try again.");
      }
    });

  return () => controller.abort();
}, [debouncedQuery]);

---

Example 3: Optimistic UI Update

const deleteItem = async (id) => {
  const previous = items;

  setItems((curr) => curr.filter((item) => item.id !== id));

  try {
    await apiClient(`/items/${id}`, { method: "DELETE" });
  } catch (err) {
    setItems(previous);
    setError("Delete failed. Please try again.");
  }
};

---

Best Practices

• ✅ Centralize API logic in a dedicated layer
• ✅ Normalize errors using a custom error class
• ✅ Always handle loading, error, and success states
• ✅ Use AbortController for request cancellation
• ✅ Retry only transient failures (5xx)
• ✅ Use debouncing for input-driven APIs
• ✅ Deduplicate identical requests

---

Anti-Patterns

• ❌ Retrying 4xx errors
• ❌ No request cancellation (memory leaks)
• ❌ Race-condition-prone state updates
• ❌ Swallowing errors silently
• ❌ Global loading/error state for multiple requests
• ❌ Calling APIs directly inside components repeatedly

---

Common Pitfalls

Problem: UI shows stale data

Solution: Use cancellation or guard against outdated responses

Problem: Too many API calls on input

Solution: Use debouncing + cancellation

Problem: Duplicate requests from multiple components

Solution: Use request deduplication

Problem: Server overload during retry

Solution: Use exponential backoff

Problem: State updates after component unmount

Solution: Use AbortController cleanup

---

Limitations

• These examples use vanilla JavaScript patterns; adapt them to your framework's data-fetching library when using React Query, SWR, Apollo, Relay, or similar tools.
• Do not retry non-idempotent mutations unless the backend provides idempotency keys or another duplicate-safe contract.
• Do not expose privileged API keys in frontend code; proxy sensitive requests through a backend.

---

Additional Resources

• https://developer.mozilla.org/en-US/docs/Web/API/AbortController
• https://react.dev
• https://axios-http.com

---