My main workhorse station — my ThinkPad x220 — has been on Fedora 16 since before the Beta release. But today since I have a calendar free of meetings, I thought I would move my gigantic workstation to F16 as well. However, the workstation has a gigantic hard disk and contains a partial Fedora mirror. I didn’t want to move all that stuff to another station for purposes of installation, and I wanted to do a faster installation than the network would allow (even with a 10-15 Mbps and a relatively fast local Fedora mirror on the Internet at UVa or Virginia Tech), so I decided to see if I could use its own storage from which to do the installation.
Obviously it was very important for me in this situation to choose a Custom disk layout. I needed to make sure not to inadvertently wipe out the partition where my Fedora mirror material was stored. I carefully selected the appropriate options there.
Then came the hard part — having my station serve its own content to itself. I completed the installation steps up to the package selection. At this point the Anaconda installer only wants to see HTTP/HTTPS or FTP repositories for installation. So what I decided to do was to set up a minimal Apache web server to serve out content already on the hard disk to the installation program. I decided to only do an installation from the base repository for the original release, and then I’d yum update first thing after booting the new installation.
How’d I do that? Using the shell environment Anaconda provides for utility and debugging. If you hit Ctrl+Alt+F2 you can get the virtual console that holds a bash shell. Once there, you can either use rpm2cpio or manual copies to bring in the necessary content, since the installer environment is a highly stripped down Fedora installation. I did it somewhat manually, identifying a couple necessary libraries from apr and apr-util first by running ldd on the httpd binary, then bringing in httpd stuff and testing with the following command line until things worked:
httpd -e debug -X
As a result the package installation took about 3 minutes instead of 45 or more over the Internet connection (and well over an hour if I’d had to move all the mirror data around first).
I wouldn’t recommend this method unless you’re fairly comfortable with running Web servers already. And again, all of this is a highly unusual case where I was putting silly demands on the installer. But hopefully this is a nice illustration of why there’s always another way to skin a cat using Fedora. Or maybe it’s just a “stupid pet trick.” Either way, now that it’s memorialized I may be able to find this info next time I want to try this.
I’ll combine this with the information in a previous post to finish my installation shortly.
Very cool!
I wish Anaconda accepted file:\\ URIs in addition to HTTP, FTP, and NFS.
Could you have used this python command instead of httpd?
python -m SimpleHTTPServer 8000
Allen
What’s wrong with //docs.fedoraproject.org/en-US/Fedora/16/html/Installation_Guide/s1-begininstall-hd-x86.html ?
@Allen: Possibly, had I known it was there!
@Mat?j: That works for loading the installer image, and for package content if you have ISO images stored, not an exploded tree. (The Anaconda repo=hd option is similarly restricted, I believe.)
not mentioning that somebody should pack his luggage for onion-peeling-in-submarine mission (//joelonsoftware.com/articles/Unicode.html), I will try to ask for some magical solution.