I'm trying to run
docker-compose --env-file <(cat file1.env <(echo -e "\n\n") file2.env) config docker-compose expects --env-file to be a file. I need to use (concatenate) two files.
Running docker-compose --env-file file_any.env config works well.
Running cat file1.env <(echo -e "\n\n") file2.env separately outputs valid result.
But it somehow doesn't work with docker-compose.
What am I doing wrong?
You don't need an additional process substitution. The outer process substitution captures the standard output of all the commands it wraps, so you aren't limited to a single cat command.
docker-compose --env-file <( cat file1.env; printf '\n\n'; cat file2.env) config Unfortunately, docker-compose requires the argument to --env-file be a real file. The value of the argument is passed directly to Environment.from_env_file, which makes an explicit check via env_vars_from_file:
def env_vars_from_file(filename, interpolate=True): """ Read in a line delimited file of environment variables. """ if not os.path.exists(filename): raise EnvFileNotFound("Couldn't find env file: {}".format(filename)) elif not os.path.isfile(filename): raise EnvFileNotFound("{} is not a file.".format(filename)) ...
--env-file <(cat file.env) does
--env-file <(cat file.env) doesn't work, then docker-compose expects a seekable file handle and you can't use process substitution, period.
docker-compose is only a wrapper and passes its options down to docker so it can further spawn containers. It makes sense that it doesn't like what process substitution produces, what I suppose to be some kind of ephemeral file descriptor.dotenv enough to see if that's a necessary restriction, or if it actually could read from a /dev/fd file (it may be restricting itself too much).
docker-composedoesn't issue any errors in cases there's no file or it is empty, only can judge on the output not having variables defined in .env files.