The current convert-html-to-pdf split (Worker fetch HTML + per-image proxy, browser html2pdf) fits Cloudflare Workers Free, with CPU and daily-request multipliers as the real ceilings.

Date: 2026-09-06 13:15 Summary: The current convert-html-to-pdf split (Worker fetch HTML + per-image proxy, browser html2pdf) fits Cloudflare Workers Free, with CPU and daily-request multipliers as the real ceilings. Visibility: people

[question] 分析目前这种模式，cloudflare free能否适用

Tool links (English)

https://onlinefreetools.org/tools/convert-html-to-pdf Chinese: https://onlinefreetools.org/zh/tools/convert-html-to-pdf

[try to solve]

Context

The tool loads a URL through a Worker (GET /api/tools/convert-html-to-pdf?url=), proxies some images (/asset), then runs html2pdf.js in the tab. The question is whether that design stays inside Cloudflare’s free allowances (Workers Free + R2 free tier), not whether every live site will convert.

Official limits used here: Workers limits (updated 2026-09-05) and R2 pricing / free tier.

Process

Mapped each hop to a quota:

PDF itself never runs on the Worker. html2canvas / jsPDF stay in the browser. That avoids Free’s 10 ms CPU / request and Browser Rendering’s 10 minutes/day.
HTML API does one fetch (plus up to 5 redirects), caps the body at 2 MB, rewrites CSS/image URLs with HTMLRewriter, does not download CSS or images in that same invocation. Subrequests stay far under Free’s 50 / invocation.
Asset API is one Worker request per image (client cap 80, concurrency 4), 1.5 MB each. Subrequests = 1. CPU is mostly arrayBuffer + magic-byte sniff.
R2 only stores prerendered tool HTML. Conversion responses are not written to R2.

Root cause / analysis

Free is usable because work is split across many short invocations and the heavy screenshot stays on-device. A single-invocation “fetch page + all images + render PDF” Worker would fail Free (CPU, 50 subrequests, and Browser Rendering quota).

Risks that remain on Free:

10 ms CPU on a large HTML rewrite (Cloudflare notes parse-heavy work often uses 10–20 ms). Waiting on fetch does not count; HTMLRewriter + TextDecoder on a 2 MB page might. Occasional overage is tolerated; consistent overage returns Error 1102.
100,000 Worker requests / day counts the HTML fetch and every /asset call. One image-heavy convert can be ~1 + 80 invocations, plus normal page views through the same Worker/R2 path.
No rate limit on these two APIs today; a scraper would burn the daily cap for the whole site (same Worker).
Sites that block datacenter IPs still fail (Could not load that URL). Paid Workers does not give residential IPs.
Using the pair of endpoints as a general open proxy can trip abuse review; SSRF checks exist, but quota and ToS are still the operator’s problem.

R2 free (10 GB-month, 1M Class A, 10M Class B, free egress) is not the bottleneck for this tool.

Solution

Yes, Cloudflare Free applies to the current pattern for small-to-moderate traffic, as long as PDF stays in the tab and the Worker only does short fetch/rewrite/proxy. Upgrade to Workers Paid (~$5/month) when: HTML rewrite hits 1102, daily 100k invocations, or you need a real headless browser. Do not move html2pdf onto the Worker on Free.

Notes / boundaries

Workers Free ≠ Cloudflare website Free; this analysis is Workers + R2 free tiers on the existing custom domain.
Workers AI on the same Worker has a separate 10k Neurons/day cap; this tool does not use it.
Paste-HTML-only conversion uses zero of these fetch APIs.

[actions]