Guides · Python · requests

Use the 24-hour render cache to cut costs in Python with requests

Every response carries an X-Cache header, and it is worth wiring into your logs on day one: HIT means the bytes came from the 24-hour cache — served in milliseconds and not counted against your quota. MISS means a real browser rendered the page and one render was metered.

Below is a complete, runnable Python program using requests (pip install requests). It reads your API key from the SNAPDOK_KEY environment variable — free keys take about thirty seconds and need no card.

import os
import requests

resp = requests.post(
    "https://snapdok.io/v1/render",
    headers={"Authorization": "Bearer " + os.environ["SNAPDOK_KEY"]},
    json={
        "url": "https://your-app.com/report/42",
        "format": "png",
        "width": 1280,
    },
    timeout=90,
)
resp.raise_for_status()

with open("shot.png", "wb") as f:
    f.write(resp.content)
print(resp.headers.get("X-Cache"), "# HIT = free, MISS = rendered and metered")

The cache key is the normalised request body: URL plus every rendering parameter. Same URL at a different width, format or scale is a different entry; alias spellings (device_scale_factor vs device_scale) are folded together first, so they share an entry. Failed renders are never cached and never metered.

Two design consequences. First, idempotent retries are safe: a queue worker that re-fires the same job inside a day costs nothing extra. Second, if you need a fresh render of a page that just changed under the same URL, vary the request — the pragmatic trick is a throwaway query parameter on the target URL (?v=deploy-id), which changes the fingerprint and forces a MISS. Rate limiting still applies to HITs (it protects the endpoint, not the renderer), so keep an eye on X-RateLimit-Remaining in tight loops.

Same endpoint, one more trick: if the page you are rendering has a form on it, adding "pdf_forms": true to a PDF render brings it back with real, fillable AcroForm fields — a PDF people can type into, not a picture of one. How that works.

Notes for requests

Three requests-specific details the sample already handles. resp.content is the binary body — resp.text would decode PDF bytes as text and corrupt them. raise_for_status() turns a JSON error response into an exception instead of letting it reach the open(..., "wb") call. And requests has no default timeout — without the explicit timeout= a hung connection blocks forever, which matters for a call that drives a real browser and legitimately takes seconds.

Response headers worth reading

Header	Meaning
`X-Cache`	HIT or MISS.
`X-Quota-Remaining`	Metered renders left this month.
`X-RateLimit-Remaining`	Requests left in the current second.

Use the 24-hour render cache to cut costs in Python with requests

Notes for requests

Response headers worth reading

Related guides

Same task, other stacks

More with Python + requests

In the wild