Hongdam หงส์ดำLINE

Last mile · · 4 minutes to read

One Cloudflare Pages line moved a domain: 147 of 147 URLs in one hop, POSTs become GETs

ย้ายทั้งโดเมนด้วยบรรทัดเดียวใน Cloudflare Pages: 147 จาก 147 URL ถึงปลายทางในฮอปเดียว ส่วน POST ของฟอร์มกลายเป็น GET

One _redirects line sends each old path and its query string to the new address. Left without a code, the line is a 302. A 301 drops a form's POST. A 16-line check, and what Google's Change of Address tool asks for.

A site that changes its address needs each old URL to answer with a redirect to the same page at the new address. On Cloudflare Pages that is one line in a text file. This week we moved a 147-page site of ours from its own domain, defiant.to, to a folder on GitHub Pages. Measured today: 147 of 147 old URLs reach their new page in one hop. The same line turns a form's POST into a GET, and the line's default status code is the temporary one.

One line carries each path and its query string

Cloudflare Pages reads a file named _redirects in the published folder. Each line is a source, a destination and an optional status code 1. An asterisk in the source matches the rest of the path, and :splat in the destination puts it back 1:

/* https://example.github.io/newsite/:splat 301

A Pages project whose only file is this one, with the old domain attached as its custom domain, is the whole redirect server. Three facts from Cloudflare's page, read today 1:

  • The code is optional and defaults to 302, a temporary redirect. A line without 301 moves visitors and says the move is temporary.
  • A file takes 2,000 static lines and 100 lines with a splat or placeholder. A whole-domain move uses one of the 100.
  • Redirects in the file do not apply to requests that a Pages Function (server code inside the project) answers.

Cloudflare's pricing page says requests that invoke no Function are free and unlimited on the free and the paid plans 2. A project with one text file invokes none.

What the line does with awkward requests, measured 21 November 2026 with curl, one request each:

Request to the old domainAnswer
/about/301 to /newsite/about/
/about/?utm_source=x&id=7301, query string intact
/a%20b/%E0%B8%81/ (a space and a Thai letter, encoded)301, encoding intact
/no-such-page/301; the new host then answers 404
//about/301 to /newsite/about/, the double slash folded to one
HEAD /about/301
http:// in place of https://301 to https:// on the old domain first, then the line

The 301 carries no Cache-Control header. RFC 9110 lets a browser cache a 301 on its own judgement 5, so a wrong destination shipped for an hour can sit in visitors' browsers after the fix.

147 of 147 land in one hop; a missing slash costs a second hop

The new site's sitemap (its list of pages) has 147 URLs 3. We requested each one on the old domain and followed the redirects: 147 answered 200 at the matching new URL after one hop. Then the 146 URLs other than the home page, with the final slash removed, as people type them: 146 answered 200 after two hops, because GitHub Pages adds the slash with a 301 of its own.

Seconds to load /about/, redirects included · lower is better
New address, direct0.073 s
Old domain, one hop0.262 s
Old domain, no final slash, two hops0.283 s
Median of 20 runs each, curl, measured 21 November 2026 from Thailand. The redirect answer alone: median 0.182 s.

Google's page on site moves advises a redirect straight to the final URL, says Googlebot follows up to 10 hops, and says to leave the redirects in place for at least a year 4.

A 301 turns a form's POST into a GET

A form or a webhook (a message one service posts to another) sends a POST: a request with data in its body. RFC 9110 says of the 301:

a user agent MAY change the request method from POST to GET

and names the 308 as the code that forbids the change 5. Measured today, a POST to a made-up path on the old domain:

ClientAfter the 301Final answer
curl, following redirectsGET, body dropped404
curl with --post301POST405 from GitHub Pages

GitHub Pages serves files and answers a POST with 405, so a 308 changes nothing when the new host is static. The form needs a receiver of its own and the page needs to post to it by full address. Our site's sign-up form posts to a Cloudflare Worker on a separate hostname; the redirect line plays no part in it. Before a move, the list to make is each place outside the site that posts to the old domain: payment webhooks, form services, apps.

Google's tool accepts a folder as the destination, and takes each hostname separately

Search Console's Change of Address tool tells Google that a site moved. Its help page, read today 6:

  • The old site is a domain or subdomain. A folder cannot be the starting point, and a folder on another domain can be the destination.
  • www does not move with the bare domain. Each hostname variant of the old site is filed on its own.
  • The tool checks for 301s on a few pages before it accepts the request.
  • Google forwards signals for 180 days. After that it treats the old site, if still up, as unrelated.
  • The page recommends paying for the old domain for at least a year, so that nobody else buys it.

What GitHub Pages allows

GitHub's limits page, read today: a published site of at most 1 GB, a soft limit of 100 GB of traffic a month, and a use policy against using it as free hosting for an online shop or a paid software service 7. A shop or a paid app goes on a different host; the redirect line is the same there.

The check, 16 lines

The script reads the new site's sitemap, requests each page at the old domain, and prints each URL that does not reach its new page with a 200 in one hop.

import re, subprocess, sys, urllib.request

new, old = sys.argv[1].rstrip("/"), sys.argv[2].rstrip("/")
xml = urllib.request.urlopen(new + "/sitemap.xml").read().decode()
locs = re.findall(r"<loc>([^<]+)", xml)
bad = 0
for loc in locs:
    out = subprocess.run(
        ["curl", "-s", "-o", "/dev/null", "-L", "-w",
         "%{num_redirects} %{http_code} %{url_effective}", old + loc[len(new):]],
        capture_output=True, text=True).stdout.split()
    if out != ["1", "200", loc]:
        bad += 1
        print("BAD", old + loc[len(new):], *out)
print(len(locs), "checked,", bad, "bad")
sys.exit(1 if bad else 0)
python3 move_check.py https://example.github.io/newsite https://old-domain.example

On our move it printed 147 checked, 0 bad in 43 seconds. It reads the new sitemap, so an old URL that has no page on the new site is outside it; the old site's own sitemap or server log, saved before the switch, covers those. Moving a site between hosts, each page checked, is a service we sell 8.

Sources

  1. Redirects · Cloudflare Pages docs developers.cloudflare.com
  2. Pricing · Cloudflare Pages docs developers.cloudflare.com
  3. docs/sitemap.xml · NaNoBotCo/defiant github.com
  4. Site Moves and Migrations | Google Search Central developers.google.com
  5. RFC 9110: HTTP Semantics rfc-editor.org
  6. Change of Address tool - Search Console Help support.google.com
  7. GitHub Pages limits - GitHub Docs docs.github.com
  8. Move your website to a new host · Hongdam hongdam.net

Hongdam Studio services on this topic