Fetching Instagram Data at Scale in Python: From One Request to Async

SHINO — Thu, 23 Jul 2026 08:20:45 +0000

I collect Instagram data for <私が収集しているもの・用途例: hashtag research for my content projects>, and at some point "just call the API in a loop" stopped being fast enough. Here's how I went from a single request to concurrent fetching in Python, using HikerAPI — a REST API for Instagram data (auth is one header, from $0.001/request, 100 free requests to start).

The starting point: one request

Everything begins with a plain synchronous call:

import requests

headers = {"x-access-key": "YOUR_KEY"}
r = requests.get(
    "https://api.hikerapi.com/v2/hashtag/medias/top",
    params={"name": "photography"},
    headers=headers,
)
print(r.json())

This returns the top posts for a hashtag as JSON. Simple. But if you have <私のボリューム例: a few hundred hashtags or usernames> to go through, a sequential loop means each request waits for the previous one to finish. At ~0.5–1s per round trip, hundreds of items turn into many minutes of waiting for what is mostly network idle time.

Option 1: threads (smallest change to your code)

Since this is I/O-bound work, ThreadPoolExecutor gets you most of the win with almost no rewrite:

from concurrent.futures import ThreadPoolExecutor
import requests

headers = {"x-access-key": "YOUR_KEY"}
HASHTAGS = ["photography", "travel", "food"]  # your list here

def fetch_hashtag(name):
    r = requests.get(
        "https://api.hikerapi.com/v2/hashtag/medias/top",
        params={"name": name},
        headers=headers,
        timeout=30,
    )
    r.raise_for_status()
    return name, r.json()

with ThreadPoolExecutor(max_workers=10) as pool:
    for name, data in pool.map(fetch_hashtag, HASHTAGS):
        print(name, len(data))

Ten workers roughly means ten requests in flight at once — a 10x wall-clock speedup on a big list.

Option 2: asyncio + aiohttp (better for large volumes)

For bigger batches I switched to asyncio, mainly because a semaphore makes concurrency control explicit:

import asyncio
import aiohttp

HEADERS = {"x-access-key": "YOUR_KEY"}
CONCURRENCY = 10

async def fetch_hashtag(session, sem, name):
    async with sem:
        async with session.get(
            "https://api.hikerapi.com/v2/hashtag/medias/top",
            params={"name": name},
        ) as r:
            r.raise_for_status()
            return name, await r.json()

async def main(hashtags):
    sem = asyncio.Semaphore(CONCURRENCY)
    async with aiohttp.ClientSession(headers=HEADERS) as session:
        tasks = [fetch_hashtag(session, sem, h) for h in hashtags]
        return await asyncio.gather(*tasks)

results = asyncio.run(main(["photography", "travel", "food"]))

The semaphore is the important part: without it, gather fires everything at once, which brings us to…

Rate limits: don't hammer the API

A few things I do to stay polite and keep runs stable:

Cap concurrency. I stay around 10 in-flight requests. More than that mostly increases error rates, not throughput.
Back off on 429/5xx. Retry with exponential backoff instead of failing the whole batch:

async def fetch_with_retry(session, sem, name, retries=3):
    for attempt in range(retries):
        try:
            return await fetch_hashtag(session, sem, name)
        except aiohttp.ClientResponseError as e:
            if e.status in (429, 500, 502, 503) and attempt < retries - 1:
                await asyncio.sleep(2 ** attempt)  # 1s, 2s, 4s
                continue
            raise

Save as you go. At $0.001/request the cost of a batch is small, but re-running a half-failed batch is still wasted money. I write each result to disk (JSONL) as it arrives, so a crash only re-fetches what's missing.

Takeaway

One-off calls: plain requests is fine.
Hundreds of items: ThreadPoolExecutor, 10 workers, done in minutes.
Ongoing/large collection: asyncio + semaphore + retry, writing results incrementally.

The nice thing about having Instagram data behind a plain REST API is that all the usual Python concurrency patterns just work — no session juggling or scraper babysitting.

How do you handle batch collection in your projects — threads or asyncio? And what concurrency do you find APIs generally tolerate?