Remove /etc/varnish from our packages

[dridi]

From #105:

The fact that we maintain a default.vcl that is a copy of example.vcl is in my opinion something we should kill at some point in the weekly branch and keep maintaining for the 6.x series. Much like we simplified our packaging by no longer generating a secret and letting varnishd generate one, I'd like to replace the -f /etc/varnish/default.vcl option with -b localhost:8080 and let users decide how they operate their VCL.

The only reason why we ship things in /etc/varnish and especially a default.vcl is tradition. There's nothing in varnish-cache itself that warrants such a default configuration. I suggest we make this change to the weeklies after the Sep 2018 release.

[gquintard]

I'm a bit on the fence about this, I clearly follow the desire to clean things up, but:

• -b won't get you very far
• people will add -f and immediately get a conflict error because of -b
• default.vcl was here, ready to be modified, I assume example.vcl will (have to) be copied over?

[lkarsten]

I don't think this is a good idea.

Significant part of the value you get from Varnish is VCL.

I can't think of a single dynamic site that won't end up in return(pass) because of cookies everywhere, if someone tried to use -b today. This will create another step anyone new to Varnish must accomplish to get it going/caching, and that is just wasted effort instead of doing it at the packaging level.

[dridi]

This came up again during a discussion and I still think we should ship an empty /etc/varnish directory. If the only argument I get is the need for an extra configuration step (because so far it comes down to that) then I'm willing to ignore it.

We already ship a default ExecStart that requires a manual change either by dropping a new varnish.service somewhere in /etc or by using a systemd override. Unless of course if -s malloc 256m is something we recommend users to stick to.

At the very least, moving from -f to -b doesn't change the fact that today 1) you have a working service out of the box and 2) you need to do some sysadmin work to get it in a production-ready state.