On Sun, 2015-05-10 at 17:11 +0200, Vincent Bernat wrote: > The disease is that actual servers running actual free software can > break at each boot because we cannot have both a persistent naming > scheme and use the eth* prefix is worse that the cure because old > versions of Novell ZENworks may stop to work on upgrade? Speaking as someone who runs Debian on his servers and laptop, I don't care about what you guys choose. I don't care on the laptop mostly because of the reasons Josh Triplet pointed out earlier. I manage the network through GUI interfaces, and it doesn't care what the interfaces are called. For servers the interfaces are all assigned fixed, well known names. The reason is their configuration is completely automated through 100's of scripts. In all instances I've seen it done like this well known interface names are used. It's not difficult to understand why if you've done it - you can sort out what NIC is supposed to be doing at install or boot up and rename it accordingly, or you can do it in every script that deals with the network. The choice is obvious. You have the "disease" wrong. These scripts have been around for over a decade. In that time various fashions in interface naming have come and gone. This is yet another one. The disease isn't the kernel choosing different names on boot up, it's people inventing new interface naming schemes every few years, just as being done here. Everyone who has had to support many servers over a long period gets sick of this and writes yet another script that does in the renaming in the way they want, in a way that will be stable forever more. Thus they don't care what choice is made here - as they have already given up relying on Debian to do it, because Debian isn't stable enough. That said when writing our own scripts most of us long ago came to the conclusion the bus path to the interface was the most useful way to identify it. Again the reason is straight forward - if you have an image tuned for a piece of hardware that you have deployed en-masse the one thing that is the same on all of them is the paths to the NIC's. Note that when, long ago, the kernels actually managed to repeatability name the NIC's eth0, eth1 etc, it was because the buses were enumerated in a repeatable order so the NIC's got seen in a repeatable order. So in effect the name was determined by the bus path. Ergo, nothing has changed in 20 years. I get the distinct feeling some people posting here consider ifup/down "old fashioned". Granted it doesn't have a nice GUI, but from the point of view of someone who deploys lots of similar machines a GUI of any sort is a negative, and it has a far nicer property - it is easily scriptable. In fact underneath the hood it's driven by scripts. If there is a network configuration it isn't capable of setting up, I haven't seen it. In my very brief look at networkd it didn't provide anything like the same amount of flexibility.
Attachment:
signature.asc
Description: This is a digitally signed message part