In fetch/747 it was suggested that browsers should send Accept-Encoding: identity along with range requests, else some servers ignore the range and return a 200. So, this is a test!

data.json contains all the URLs in HTTP archive that end in mp3 or mp4, de-duped by host. Warning: These are media files from the internet, so many will be not safe for work.

Here's the query:

SELECT ANY_VALUE(url) as url FROM `httparchive.runs.2018_02_15_requests`
WHERE REGEXP_CONTAINS (url, r'\.(mp4|mp3)$')
GROUP BY req_host

The test makes requests to those URLs with the following headers:

Range: bytes=0-
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_13_4) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/66.0.3359.181 Safari/537.36

It makes a request without an Accept-Encoding header, then makes additional request for each of these headers:

Accept-Encoding: identity;q=1, *;q=0
Accept-Encoding: identity
Accept-Encoding: gzip, deflate, br

And records the status code & Content-Encoding header for each response, or {err: true} if the request timed out or failed.

The results are in results/out.json, but I recommend the interactive html results (depends on modern features like transform streams, so may only work in Chrome).

Note: The data is ndjson. Also, some servers may be unstable and return different results each time.

Running the test

Requires Node 10.2.1+.

# Generate results/out.json from data.json
node index.js
# Print the results for a particular URL
node test-url.js https://www.narita-airport.jp/files/bg.mp4
# Print results where the return is different depending on Accept-Encoding
node log-interesting-results.js