Re: [gentoo-dev] [RFC PATCH 04/19] www-servers/nginx: add nginx-r5.initd

[Michael Orlitzky <mjo@gentoo.org>]

On Wed, 2024-07-17 at 15:05 +0300, Zurab Kvachadze wrote:
> 
> +NGINX_CONFIGFILE=${NGINX_CONFIGFILE:-/etc/nginx/nginx.conf}
> +pidfile=${NGINX_PIDFILE:-/run/nginx.pid}

OpenRC allows you to run multiple instances of the same service by
creating symlinks in /etc/init.d. For example, I might symlink nginx-
internal to nginx, and then run,

  # rc-service nginx-internal start

The two configurable variables above make this possible, but require
the user to create a new /etc/conf.d/nginx-internal file to override
them. It would be a small improvement to use $RC_SVCNAME instead of
"nginx" as the default. That way, for the new service, they would be
/run/nginx-internal.pid and /etc/nginx/nginx-internal.conf.

In fact, I would delete the NGINX_PIDFILE variable entirely, leaving it
at /run/${RC_SVCNAME}.pid. There's no reason for anyone to change it.
You can force the daemon to use that path with -g "pid ${pidfile}",
relieving you of the responsibility to coordinate with the config file.



> +depend() {
> +	need net
> +	use dns logger netmount
> +}

I don't think "need net" is right here because nginx starts fine on
localhost or 0.0.0.0 until you configure it to use a specific address.

There's a section called "Be wary of 'need net' dependencies" in,

  /usr/share/doc/openrc-*/service-script-guide.md.bz2

that talks about this.


> +start_pre() {
> +	checkpath -d -o root:root /var/tmp/nginx || return 1
> +}
> +

The old script had,

    if [ "${RC_CMD}" != "restart" ]; then
        configtest || return 1
    fi

here. If "nginx -t" produces better error messages that plain nginx,
that might be the reason. But otherwise it's redundant.