FC4t2 no good without LILO

Paul Iadonisi pri.rhl3 at iadonisi.to
Tue Apr 12 18:10:37 UTC 2005


On Tue, 2005-04-12 at 10:21 -0700, Mike Bird wrote:
> > Which of the 82 Grub bugs prevents it from being used for serious work?
> 
> 1) That basic Bugzilla queries yield 82 Grub bugs versus 15 Lilo bugs is
> illustrative of their relative levels of reliability.

  Not really.  I demonstrates that grub is used more since it's been the
default in RHEL/FC for some time.  It's entirely predictable that lilo
will have fewer bug reports if it's not used by as many RHEL/FC users.
Frankly, because of that fact alone, I don't think it demonstrates
anything either way.

> 2) The general flakiness of Grub makes it unusable for serious work. 
> One simply cannot afford plane tickets (or even trouble tickets) every
> time Grub fails.  Lilo is much more reliable than Grub.  Grub may be
> appropriate for a newbie on a home PC, but Grub does not replace Lilo.

  You're speaking in generalities.  How is it flaky?  I do agree that
grub is flaky when it comes to getting it installed and working on some
systems.  But once up and running, even upgrading a kernel is more
reliable.  For remote, mission critical systems where a plane ticket
costs to much, I would *always* pick hardware that had a serial bios and
redirect the grub output to the serial port.  I would also be sure it
was hooked up to a remotely cycle-able power strip (or there was 24hr
staff available at the data center to power cycle for me).
  With that kind of setup, even if you have a bogus grub.conf entry, you
can drop to a grub command line and boot with the proper options and
filenames.  Try that with lilo.  Especially if something gets messed up
and lilo isn't run after installing a new kernel (though I believe
that's done in new-kernel-pkg now, however it could conceivably fail).
No need to rerun grub-install when installing a new kernel, so the boot
block is not getting updated.
  So lilo seems more reliable in terms of getting it installed initially
(via anaconda or even manually), but after getting it working (which on
many systems *is* out of the box), it tends to be *less* risky than lilo
(i.e.: no updating of the boot block).  I would never upgrade the OS (or
even just grub) remotely on a critical system, anyhow, so the difficulty
in getting it on (some) systems is less of a concern than ending up with
a bogus config (lilo has no command line to use to recover ... you must
be local and boot from a rescue CD to fix it).

> 3) One of the most serious ongoing problems with Grub has been the
> flakiness of the software RAID support.

  Ah, now we're talking!  That problem has aggravated me for quite some
time.  I don't remember if I ever tried lilo in that config, but I've
had exactly that problem FOREVER (er...ever since grub replaced lilo as
the default, anyhow) on two of my systems.  Fortunately, as mentioned
above, that problem only crops up during an upgrade (or clean install).
I usually have to boot to rescue mode, muck with /etc/mtab to fake out
grub into thinking that my disks are /dev/sd[ab][123] instead of /dev/md
[012], and run /sbin/grub-install /dev/sda and then muck with /etc/mtab
again and do /sbin/grub-install /dev/sdb.  Ugly, but it works.  Or at
least, boots now.  I have no idea if it'll work if I yank /dev/sda
(though I suspect if /dev/sdb is gone, it'll work).

>   This was first added in 0.90
> (July 2001) but has ever since been a constant source of Bugzilla
> reports - reports which are not included in the list of 82 because they
> were closed when the March 16th patch was written.  Now maybe that patch
> will solve all known problems, but I'm certainly not going to bet the
> farm on it without months of testing.

  Um, I'd agree, but aren't we actually *doing* months of testing here?
This is fedora-test-list, after all, and this is Fedora Core 4 *Test* 2
we are talking about.

>   Why is Fedora betting its farm? 
> Lilo's RPM takes only 547k on the CD.

  I doubt disk space is the reason in this case.

> 4) And this brings us again to the incredible lack of judgment
> manifested by the Fedora Core team.  First do no harm - if you don't
> understand the issues relating to Lilo and Grub don't mess with them.

  Well, I think it's probably triage.  Grub is the future due to some
important benefits over lilo (the command line being one significant
one).  Does it have issues?  Yes.  But not with the majority of systems
out there.  I don't think it's a lack of judgment.  It will cause some
pain for some systems.  I doubt it'll make it *impossible* to use on any
system (okay, maybe a few), but it will make it a challenge to get
working on some.
  I do hope the software RAID 1 issue is fixed, myself.  But I'm not
willing to go back to lilo to 'fix' it for myself.  Let's get grub
working for all systems (at least all modern systems) instead.


-- 
-Paul Iadonisi
 Senior System Administrator
 Red Hat Certified Engineer / Local Linux Lobbyist
 Ever see a penguin fly?  --  Try Linux.
 GPL all the way: Sell services, don't lease secrets




More information about the fedora-test-list mailing list