Re: Finally, success (WAS Re: Mandrake install troubles with DAC960 RAID controller...)


Subject: Re: Finally, success (WAS Re: Mandrake install troubles with DAC960 RAID controller...)
From: civileme (civileme@mandrakesoft.com)
Date: Tue Jan 08 2002 - 08:59:22 AKST


On Monday 07 January 2002 09:32 pm, Joshua J.Kugler wrote:
> civileme, et al -
>
> This update is late, the server was up and running last Thursday night
> (1/3).
>
> So, I was finally able to get through the install by using the second disk,
> which booted with a 2.2.x kernel. This kernel was able to correctly
> initialize the DAC960. During the install, I was able to select the 2.4.x
> kernel, and it didn't complain, and set up lilo correctly. I rebooted,
> compiled 2.4.17, and away we went. Thanks for the tips!
>
> Now, for the complaints, suggestions. :)
>
> I realize strcmp("install", "upgrade") != 0, but it would be nice if the
> install routine would look for existing configuration info. I was
> prepared, so I had written down pertinant info, but it would be nice if
> install looked for an existing /etc, and gleaned partition information, IP
> addresses, DNS servers, etc. It would make the install go that much more
> smoothly.

Well, the first part of that is in place for 8.2 -- we needed a tool to
compress config files and put them out of the way for a "vandal upgrade"
which would scrub the old packages and links and reinstall then restore the
config files. Update between distros is not really very useful and tends to
leave artifacts. The same can be said about almost everyone's updates.

>
> I noticed during package selection that aspell would not install without
> the GB (Great Britain) dictionaries (depended on them), even if US
> dictionaries were selected. Why?
>
See the answer on kdeutils...

> What is the difference between bash and bash1? The importance on bash1
> says "maybe." Why two? It just seems a bit confusing to users, as there is
> nothing in the description to differentiate them.
>

You are not expected to know all the packages--we have 3000 of them in the
downloadable distro. bash1 is a contriibuted package which is a variant of
bash that some users felt important enough to contribute.

> kdeutils depends on efax? Really? Why can't I install kdeutils without
> installing efax? This is a server, after all.

You can, just do it after install and use --nodeps. Some dependencies like
that occur because we of necessity follow a rule of not packaging the same
tarball more than once (otherwise imagine the fun in repackaging every time a
tarball changed), so efax and kdeutils both use something packaged in the
efax package.

>
> There still needs to be some tweaking done on the startup configuration in
> the install program. I'm sorry, I didn't write everything down, but I know
> there were several services I selected to start at boot that didn't get
> included in /etc/rc.d/rc3.d/. I had to symlink them manually.

Well you do have sysVinit loaded so you don't have to do them manually, but
it is possible they were set for runlevel 5 rather than 3.... I will pass
that to install team.

>
> Another problem: Today, I discovered my server was not responding at all.
> No ping, no ssh, nothing. Went to the server closet and was going to
> login, but there was nothing on the console. Keyboard wouldn't respond.
> Num lock wouldn't toggle, etc. Ctrl-Alt-Del wouldn't work. Had to hit
> reset on the box (not something I like to do on a Linux box, much less a
> server). When it came up, I looked through all the logs, and nothing
> looked amiss (other than no entries after 12:40AM today). Does this ring a
> bell, or do I need to start trouble shooting on my own? After bringing the
> server back up, fsck'ing the partitions, and restarting everything, it was
> fine. We'll see if it happens again.
>
> Thanks again for all your help. It is much appreciated.
>
> j----- k-----

System crashes are flabbergasting... Way way outside the experiences of QA.

Update does preserve the configuration, as you would expect, but is mostly
useful for adding things later or restoring after experimenting with new
configurations.

The only problems even near the crash behavior you described that I have seen
are Reiser vs postfix or Reiser vs knfsd (nfs). I have taken to recommending
XFS to nearly everyone as the filesystem of choice for servers, ext3 is way
way too slow compared to the others... My benchmarks put them this way...

ext3 .67
ext2
Reiser
XFS all 3 approximately 1

JFS about 1.3

XFS is slow to delete a large batch of files, snappy on other functions
JFS uses significantly more storage for the same data and had some stability
problems at release time
Reiser still did not like nfs mounts and has been reported to corrupt mailbox
files made by postfix.

Civileme



This archive was generated by hypermail 2a23 : Tue Jan 08 2002 - 08:48:40 AKST