[gentoo-soc] Gentoo Clustering CD Progress Report X+1

[gentoo-soc] Gentoo Clustering CD Progress Report X+1

[Eric Thibodeau]

[-- Attachment #1: Type: text/plain, Size: 3602 bytes --]

_*Whuh?
*_Yeah, I know, radio silence for too long, I forgot I had exam/lab 
correction assigned to me last week, sorry about that folks (that also 
knocked me off schedule as I was at ease reading up flame wars about how 
system configs should be done before that realization...more below)

_*Configuration automation:*_

I spent the past weeks investigating how my scripts would be integrated 
into portage in a /_durable fashion_/ and it seems I keep hitting 
friction from /EX/soon to be ex/still active/ devs concerning what an 
ebuild should/can do to a filesystem. I am crawling to the conclusion 
that all the automation should be accomplished with a mix of 
CONFIG_PROTECT_MASK (CONFIG_PROTECT="") and addition of my scripted 
functions to pkg_config() sections of meta/seed ebuilds... I figure this 
is the Gentoo way to do things. Down the line, some of the pkg_config 
stuff could probably be integrated into upstream as part of an 
configuration backend...but I'm diverging.

_*Evolution:*_

/*net-dns/dnsmasq auto-config*/: is almost complete, the script will 
still require some form on configuration input but very little compared 
to ldap-setup.sh, most of the network configuration is deduced from the 
use of /*net-misc/sipcalc*/ applied to the eth? on the private net.

Caveat:
* Assumes the private eth is already configured with a static IP 
address...don't know how I will pull that one off. I could also force 
the static IP in /etc/conf.d/net ...but that also assumes I force eth1 
to be the private nic.

/*catalyst:*/
I was able to rebuild the stage tarballs. This time I didn't such 
specific arch settings as nocona and left it to a generic x86_64 setting.

_*Regression:*_
I have been hitting a circular dependency hellish problem while 
attempting building livecd-stage1. I though the problem was due to my 
switch of the snapshot so I decided to rebuild the entire stages1-3:
- stage3-desktop won't build because openssh pulls in 
/*net-nds/openldap*/ (it's now on by default apparently) and, for 
another reason, openldap is being merged _before_  /*sys-apps/shadow*/ , 
which breaks everything as described by  
http://bugs.gentoo.org/show_bug.cgi?id=142637 leading to 
http://bugs.gentoo.org/show_bug.cgi?id=53269 ... And to quote the mentor:

20:12 < dberkholz> the "fix" is what i mentioned above, plus a patch 
across the whole tree to fix it
20:13 < dberkholz> first add new eclass, second change all ebuilds and 
eclasses running enew{user,group}, third drop from eutils
20:13 < dberkholz> the third could wait a while because it'll break old 
binary packages from before portage cached the whole environment

...which means we might add "new eclass" on the list of things to do.

_*Comming up:*_

* Finish proof of concept script for dnsmasq auto-configuration.
* Figure out how eth1 will be auto-configured within /etc/conf.d/net
* NFS root, how will it be created?!:
** catalyst?
** quickconfig?
**  DonnieMatic (http://dev.gentoo.org/~dberkholz/ltsp/setup.txt)?
** All of the above :)

_*Time killer:*_

I expect the circular dependency problem with the liveCD to take up much 
of my time tomorrow (very unfortunate). For those of you that are 
curious (or even read this O_o!) http://pastebin.ca/1059646

_*Spying:*_

I checked out http://pareto.uab.es/mcreel/PelicanHPC/ as a reference 
(linked to it by http://idea.uab.es/mcreel/ParallelKnoppix/ which seems 
to be dropping the ball). I was wondering how the configuration was 
implemented and saw quickly that choices are made for the user and there 
is a configuration phase to be accomplished...

Eric

[-- Attachment #2: Type: text/html, Size: 4578 bytes --]

[gentoo-soc] Gentoo Clustering CD Progress Report X+1

[Eric Thibodeau]

[-- Attachment #1: Type: text/plain, Size: 2721 bytes --]

Picking up where we left off:

_catalyst:_
We're back on track, LiveCD now able to build again, the circular 
dependency is being caused by the fam USE flag (disabled in the spec 
file for now)

_Hit and run:_

I hit 2 bugs today during the LiveCD build and they also got fixed, 
hurray for the quick commits and interaction from the devs!

    http://bugs.gentoo.org/show_bug.cgi?id=229103
    www-client/mozilla-firefox-bin-2.0.0.14 needs to DEPEND on pango, no
    officially fixed in the tree, manually fixed in my snapshot

    http://bugs.gentoo.org/show_bug.cgi?id=206753 (again)
    "gnome-base/gnome-desktop-2.20.3 - parser errors during emerge"

_Coding_

** Finished, as in, it works: **

* net-nds/ldap-auth [1]
     Available versions:  0.1 {authmaster autoconfigure}
     Homepage:            
http://code.google.com/soc/2008/gentoo/appinfo.html?csaid=79DB497BA0D95148
     Description:         This meta ebuild automates the configuration 
of OpenLDAP as the authentification backend

[1] /usr/local/portage/soc2008

The whole ldaps-setup.sh script thing I had going is now transformed 
into an ebuild with all the work done in pkg_config because:

1- It lets me fsck the filesystem
2- I get to use my original code almost unchanged

USE flags added:
autoconfigure: "Configure the package with NO user interaction, you get 
what you pay for ;)"
authmaster:      "Differentiate between the authentication server and a 
simple workstation that needs to authenticate"

usage:
emerge net-nds/ldap-auth
emerge --config =net-nds/ldap-auth-0.1

** Started **

* sys-cluster/beowulf-head [1]
     Available versions:  0.1
     Homepage:            
http://code.google.com/soc/2008/gentoo/appinfo.html?csaid=79DB497BA0D95148
     Description:         This is a meta ebuild that pulls in the 
packages and some config
scripts/files for building the Gentoo Clustering Master node

Work stared on it, dnsmasq complete but not verified, same for exportfs 
generation and pxelinux.cfg/default file

** git.overlays.gentoo.org **

    FINALLY! My work is now hosted on git.overlays.gentoo.org. I cleaned 
up my sandbox and attempted to organize the stuff as coherently as I 
could. Everyone can now look at my monstrous ebuilds :P

_DOC_

I didn't think I was going to jump on this this quickly but there is 
already a HOWTO in my git repo to reproduce my environment. I am 
currently testing it on a "virgin" machine and all seems to be working 
fine. Speaking of which, check this out:

http://wiki.neuralbs.com/~kyron/soc2008/htopSunFireX4600BuildingGentooClusterCD.jpg

^^^^ This was taken while catalyst was building "something"

Nuf said, go look at the git and the bug reports...this has been a long 
day ;)

Eric

[-- Attachment #2: Type: text/html, Size: 4002 bytes --]

[gentoo-soc] Gentoo Clustering CD Progress Report X+1

[Eric Thibodeau]

A frustrating day with lots of work, standard definitions but little done...

========================================
What I remember doing:

-> openldap built with minimal problem:

        -- Fixed by explicitly adding it to the liveCD-stage1.spec 
(forced rebuild with the correct USE flag)
        -- read 
/usr/portage/profiles/default/linux/amd64/2008.0/desktop/../../../../../targets/desktop/package.use 
for more details

-> discuss about IP configurations and defaults. We finally opted to 
give instructions in the MOTD about  running emerge --config on the 
specialized packages I have built.

-> Built LiveCDs x++ times to test the process and integration of the 
ebuilds, hit a few errors on the way but finally got something concrete.

========================================
Tomorrow:
-> Add instructions to the user for setting up the environement 
(beowulf-head/beowulf-head-0.1.ebuild) as well as some verification on 
the network configuration (make sure it IS at least configured)
-> Start testing dnsmasq config from the liveCD

-> Need to find a way to mimic the gentoo-livecd-release spec profile 
(so the user _can_ install from the clustering liveCD...)
...and lots more...

========================================
Heros of the day:

agaffney on #gentoo-cluster
pchrist  all over the place
dberkholz TCP/IP is fun huh! :P

Eric

PS:

The stats (now that I have git ;P):

kyron clustering-livecd # git log --stat |cat
commit 9cccd8ed2a77936ff5a57f4813597e20a33bc7e9
Author: Eric Thibodeau <kyron@neuralbs.com>
Date:   Thu Jul 3 02:41:38 2008 -0400

    Correct bgus in the ebuilds (incomplete variable changes)
    Add compensation for missing dirs on LiveCD (openldap ebuild not 
creating /var/run/openldap ...bug?)

 overlay/net-nds/ldap-auth/Manifest                 |    2 +-
 overlay/net-nds/ldap-auth/ldap-auth-0.1.ebuild     |   19 
+++++++++++++++++++
 overlay/sys-cluster/beowulf-head/Manifest          |    2 +-
 .../beowulf-head/beowulf-head-0.1.ebuild           |    2 +-
 4 files changed, 22 insertions(+), 3 deletions(-)

commit a1412d103896111cd75a2d9e2facd2dac56fb6a9
Author: Eric Thibodeau <kyron@neuralbs.com>
Date:   Thu Jul 3 01:48:21 2008 -0400

    "##DISPLAY_MANAGER" --> hardcoded to "gdm" in motd

 .../releases/2008.0/misc/clusterlivecd.motd.txt    |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

commit 838850ca3ff9669f22f1674841f2916b7ad2067c
Author: Eric Thibodeau <kyron@neuralbs.com>
Date:   Thu Jul 3 01:43:02 2008 -0400

    livecd-stage2 is now of type generic-livecd and does not start X by 
default
    Corrected a few config file path changes into /etc/gentoo/

 LiveCD-HOWTO                                       |    2 +-
 .../releases/2008.0/specs/amd64/livecd-stage2.spec |   10 ++++++----
 overlay/net-nds/ldap-auth/Manifest                 |    2 +-
 overlay/net-nds/ldap-auth/ldap-auth-0.1.ebuild     |    9 +++++----
 4 files changed, 13 insertions(+), 10 deletions(-)

commit 5256d54ac52486b5b8031a4e83d43707608bdd55
Author: Eric Thibodeau <kyron@neuralbs.com>
Date:   Wed Jul 2 22:18:23 2008 -0400

    Added to livecd: net-nds/openldap, sys-cluster/beowulf-head, 
net-nds/ldap-auth, sys-cluster/openmpi, app-admin/eselect-cblas, 
app-admin/eselect-blas, app-admin/eselect-lapack
    Added custom MOTD with basic setup instructions
    Imported a few template files and some scripts from releng to be 
modified shortly.

 .../releases/2008.0/misc/clusterlivecd.motd.txt    |   16 +++++
 catalyst/releases/2008.0/scripts/livecd.sh         |   68 
+++++++++++++++++++
 catalyst/releases/2008.0/specs/amd64/doitall.sh    |   12 +++
 .../2008.0/specs/amd64/nfsroot_stage4.spec         |   71 
++++++++++++++++++++
 overlay/sys-cluster/beowulf-head/Manifest          |    2 +
 .../sys-cluster/beowulf-head/files/cluster.conf    |   43 ++++++++++++
 profile/package.keywords/openmpi                   |    2 +
 7 files changed, 214 insertions(+), 0 deletions(-)

commit 1b505a9807658819d0cd02fd90ba834a4a09977a
Author: Eric Thibodeau <kyron@neuralbs.com>
Date:   Wed Jul 2 20:06:31 2008 -0400

-- 
gentoo-soc@lists.gentoo.org mailing list