如何解决Heroku上的“语言c权限被拒绝”错误？

Question

I have a project , which calls a Postgres function gen_random_bytes , which is declared like this:

CREATE OR REPLACE FUNCTION core.gen_random_bytes(integer) RETURNS bytea AS '$libdir/pgcrypto', 'pg_random_bytes' LANGUAGE c STRICT;

It worked fine, as long as I used it on my local machine. But I'm not able to create that function on Heroku - I get the following error:

[WARNING  ] CREATE OR REPLACE FUNCTION core.gen_random_bytes(integer) RETURNS bytea AS '$libdir/pgcrypto', 'pg_random_bytes' LANGUAGE c STRICT
ERROR:  permission denied for language c

It seems like only the superuser can declare functions in C. I found another answer , which recommends using the command

UPDATE pg_language SET lanpltrusted = true WHERE lanname LIKE 'c';

I don't have permission for that command, either:

ERROR: permission denied for relation pg_language
SQL state: 42501

How can I get my code work on Heroku?

I have at least following options:

1. Rewrite core.gen_random_bytes in the Postgres language.
2. Use some existing alternative for core.gen_random_bytes , which doesn't require superuser right.
3. Switch from Heroku to another hosting provider, which would give me superuser rights (eg Digital Ocean, Amazon EC2 etc.)

Right now, I'm just prototyping this project. Which option is the easiest one (including something that I didn't mention) ?

Answer 1

The answer is: you can't unless you run Postgres yourself on a bare instance (which I'm not sure is even an option on Heroku).

The downside to all the Postgres hosted services (like Heroku and Amazon RDS) is that internally they need to be able to run multiple Postgres databases in a single virtual server. That means they can't allow Postgres to do anything that could possibly affect things at the OS level. Untrusted procedure languages (such as C, plpythonu and plperlu, but NOT plperl) are able to do anything at the OS level that the OS user running Postgres can. That's a security hole that's unacceptable in a shared environment. (Just as an example, it'd be easy to go read raw Postgres data for every other customer on the server.)

Your only option here is to install and manage Postgres yourself. On Amazon that's pretty easy; just grab a raw EC2 instance with the OS of your choice and slap Postgres on it. You lose some of the ease of maintenance that RDS gives you, but if you know Postgres it's really not that big a deal. I tend to push my clients to do exactly that, so that I have the ability to use extensions that haven't been blessed by the RDS team (something else that's unsafe in a shared environment).

I'm not as familiar with Heroku's offerings, but as long as they have something comparable to a bare EC2 instance you can do the same thing.