Tangential, but I'm curious about the installation command:
curl https://clickhouse.com/ | sh
It seems like it just detects curl('s user agent?) and serves up the script, serving HTML otherwise (ex. `wget -O- ...` would print HTML). I wonder what the design decisions here were.
Yes, it is detecting curl by the user agent - by using the "page rules" in Cloudflare. Cloudflare is good for traffic management rules.
I've created this distribution channel as an experiment and for my own need - I use it to download and install ClickHouse without the need of package manager. So I don't need to run "apt update" to install ClickHouse. It works if the package manager is broken, and it works well on very old servers, and rare Linux distributions.
It is using the fresh build from the master branch. I have also created a website for my team to monitor the status of the continuous integration system: https://aretestsgreenyet.com/
We also have deb, rpm, tgz and Docker as wrappers for the release builds.
> I've created this distribution channel as an experiment and for my own need - I use it to download and install ClickHouse without the need of package manager. So I don't need to run "apt update" to install ClickHouse. It works if the package manager is broken, and it works well on very old servers, and rare Linux distributions.
Makes perfect sense! Thanks for the explanation :D
Interesting - for me broken package manager is indicator of that server to be fixed or reprovisioned, I'd be wondered to see some DB running on such kind of server - highly likely it's missing updates as well
Or another example - folks in a large company successfully running Ubuntu 12.04 in production for 10 years and cannot upgrade it because some Perl dependencies of their amalgamation of microservices working slightly differently.
Sketchy would be detecting bash and serving a different page depending on that; this is only a bit hackish.
Now that I’m thinking about it, normal browsers usually ask for text/html or text/* as a preference don’t they? While curl and wget just say Accept: *. Then content negotiation could achieve the same result per the actual HTTP semantics, without any hacks involving the User-Agent header.
There’s nothing in that link that says "I’m a download link" and I don’t think you would remember that the site you have open in your browser also serves as the download link if you curl it with the right user-agent.
