mmaslano

Developer Conference - day 2

Wed, 22 Feb 2012 14:43:32 GMT

Lukáš Czerner had a one of the best talk here. Benefits of btrfs are a scalability more than 16EiB (ext4 can reach 16TB). It should be possible to do a very fast file system checking, incremental backups and snapshots. It should be easy add/remove drives or add subvolumes. The main deficiencies at the moment are: fsync (not fully functional), virtualization is quite slow and also encryption is not implemented. It sounds to me like not ready as default for Fedora, but it might be good choice in the future.

Later that day presented Edward "Joe" Thornber and Zdeněk Kabeláč Thin provisioning and snapshots in device mapper. Joe showed the theoretical part of thin provisioning and Kabi showed how does it work in few easy commands. Lvm commands are often seen as too complex and hard to use by an average user, but the howto showed it doesn't have to be true.

Bryn Reeves had his second talk - How to lose data and implicate people, which was the best talk from the whole conference.
He made a summary of common scenarios how to lose data. His presentation skills are awesome, so better look on the video, but the most funniest quotes were:

How long is an ohnosecond? Elapsed time between making a catastrophic error and realising what happened.
Backups? We've heard of them...

Milan Brož had a talk about the disk encryption. He mentioned the benefits of a full disk encryption, where the key removal also means easy data disposal. Milan listed the possibilites of full disk encryption: hardware - encryption on disk controller (usb2sata on disk) - if the board broke, it's hard to save data. And the software encryption: truecrypt, loop-AES with external store for key, bitlocker (proprietary, windows), luks/dm-crypt - based on a strict separation of the disk encryption engine (dm-crypt) and the key-management (LUKS).

Don't forget to vote for your favourites.

Developer Conference - day 1

Tue, 21 Feb 2012 12:53:46 GMT

Radovan started the event with a short speech and also announced the Open House in Red Hat Brno office on April 4th as the next Red Hat related event.

František Řezníček from MRG (Messaging, Realtime, Grid) spoke after him. His talk was called Towards Unified Messaging, where he presented their work, mentioned protocols used in their project - qpid and amqp. He shortly explained the asynchronous messaging and how the cooperation works between server and users.

Bryn Reeves from support of storage had a lovely talk called - Supporting the Open Source Enterprise. He mentioned few examples from his area of work, but it's probably the same for all open source projects. Problems with support of a project in various branches, support of older releases, which are unsupported by upstream and so on. What has changed in last few years are tools. Git made things easier, but there are also other tools worth looking at like OpenGrok or cscope.

Jan Hutař presented "Software Robot Competition around the world". He made an overview of different competitions in different disciplines and showed the most funniest videos. For my colleagues was important the Red Bot competition based on FIbot, where he announced the winner of the internal competition. Jan promissed the best videos from the Red Bot on youtube.

Stanislav Kozina spoke about What can Linux learn from others. I felt that this talk was about how cool is Solaris, namely zfs and dtrace. I heard many times praise of both and they probably have many cool features. But I would prefer a comparison of programs or what are the deficiencies of these two. I guess drawbacks are everywhere.
Phil Knirsch made an overview of rpm and yum. I assume they spent a lot of time on looking into "yum is so slow" issues and aimed their work on the most irritating bugs. According to their research is rpm+yum only 4x slower than untar, so it's very good because rpm+yum are doing installation and scriptlets, SElinux, depsolving and many more. Some projects claimed to be faster, but in that case they are "cheating", because they are skipping some work. The most important future plan is using libsolv by SUSE as the depsolver. On the top of depsolver will be a new project, which will merge rpm+yum functionality into one tool, which will be introduced in Fedora 18.

Jarda Škarvada presented news from the Power Management. He announced the test day, where anyone can bring his/her own laptop into Red Hat Brno office for measurement or do the testing at home. They introduced new release of tuned 2.0, which have new profiles for power saving. He proposed a little controversial change to stop desktop application to run their flash/javascript/whatever during their minimalization. It would require patching for all Desktop Environments.

Harald Hoyer and Kai Sievers presented their new look on FHS called - A streamlined and fully compatible Linux Filesystem Hierarchy. IMHO it's another change for a change, which will probably create more problems for Fedora maintainers. I'm not quite sure what this change should bring to the user or the developer. Audience was arguing and they were arguing with Harald and Kai. We spoke afterwards, but I can't say I'm convinced...

ruby 1.9.3 in EL-6 branch

Mon, 16 Jan 2012 14:28:10 GMT

The Ruby repository contains ruby-1.9.3, rails-3.0.10, and group of essential gems. 
Everything from collection can be installed by:
yum install stack_rails_3.0

Now, when the collection is installed, the rails version can be verified.

stack enable rails_3.0 'rails -v'

As mentioned above, the only difference between running commands inside and 
outside stack is using the stack command. If you want run your own rails3 
application, then you simply install the Ruby stack. In the directory, 
where is located your application run:

stack enable rails_3.0 'rails new app_name'

Now you can develop in this directory, but everything must be executed within 
the stack environment. For running rails server you can use the command:
stack enable rails_3.0 'rails server'
But you can also run a new shell:
2/ stack enable rails_3.0 bash
where the Stack Ruby will override the system Ruby. In the new shell will be used
the Ruby from Stack, but also all gems and everything, what is provided by this
software collection.

Notes to installation: The Ruby software collection was built with the old utility and macros, so it's needed to install stack instead of scl-utils from EPEL before you start. In the future will be all software collections using dsc prefix and scl-utils from EPEL repository.

Dynamic Software Collections

Thu, 12 Jan 2012 08:01:02 GMT

Collections should provide new or old version of software packaged in rpm. Imagine you have for example EL-6 and you want use new features from Perl 5.14.x. Update of system Perl is not possible, because it would bring changes incompatible with other system packages. Or you have application running on Ruby 1.8.x and in Fedora 17 will be Ruby 1.9. Collections are groups of packages, which are using different paths for installation, so those packages shouldn't influence the system environment. It should be possible work with both versions. The system one is linked to application like vim or netsnmp and the collection package could be executed in his own shell and used only there.

Currently, there are prepared few repositories of Ruby and one repository of Perl for testing purposes.

As the maintainer of Perl I'll show all examples on perl 5.14.2, which is prepared for EL-6.

How does it work:
* install scl-utils. This package will provide macros, which are needed for running software collection. The scl-utils-build package is needed only as build requirement.
* pick a collection, in this case Perl. yum install perl514* Perl collection includes upstream tar ball, DBI, FCGI.

The installed version can be verified by:
scl enable perl514 'perl -V'

Setting SElinux rules should be easy by using same contexts on new directories:
semanage fcontext -a -e /usr /opt/rh/perl514/
restorecon -R -v /opt/rh/perl514/

If you run new session, you can use there new version only:
scl enable perl514 bash

For example execute perl -V, which shows information about your Perl. Let's look at paths specific for this collection:
    @INC:
    /opt/rh/perl514/root/usr/local/lib64/perl5
    /opt/rh/perl514/root/usr/local/share/perl5
    /opt/rh/perl514/root/usr/lib64/perl5/vendor_perl
    /opt/rh/perl514/root/usr/share/perl5/vendor_perl
    /opt/rh/perl514/root/usr/lib64/perl5
    /opt/rh/perl514/root/usr/share/perl5
    .

Ruby collection is using rails_3.0 instead of perl514 in path. They started creating their collection sooner than me, so they are using old macros. For working with their ruby stack, you need install macros and enable the repository for Fedora or EL-6.
yum install stack_rails_3.0 should install whole Ruby stack and stack enable rails_3.0 'rails -v' should verify the path.
Update: Next article should be about the functional installation of Ruby 1.9.x on EL-6.

Fedora elections

Sat, 04 Jun 2011 19:34:51 GMT

Did you know that Fedora elections already started? End of voting is 8th June for FESCo and 9th for Board. Don't forget to vote for your favourites!

Howto create RPM filters

Fri, 03 Jun 2011 18:56:25 GMT

Rpm 4.9 brought change of macros. They do not have much documentation, but I was able to create something with help of rpm guys. Thanks.

Now you can define and use your own generator of dependency, which could be handy if your packages have specific needs. New macros have only extended regular expressions, not Perl regular expressions :(

1. create filter for your favourite package:
/etc/rpm/macros.myfavourite

2. define your own generator
%global __favourite_provides /usr/lib/rpm/favourite.prov
%global __favourite_requires /home/user/favourite_inprocess.req

3. filter *.so, which don't have to be provided
%favourite_default_filter %{expand: \
%global __provides_exclude_from %{_share}/favourite/.*\.so|%{_sbindir}/.*\.so
}

4. change it to filter *.so and also unwanted require MyPage
%favourite_default_filter %{expand: \
%global __provides_exclude_from %{_share}/favourite/.*\.so|%{_sbindir}/.*\.so
%global __requires_exclude favourite\\\$MyPage\\\$
}

5. use it in your favourite package before prep
%description
This is favourite package..

%{?favourite_default_filter}

%prep

6. Provides and requires added manually before prep section are not filtered, because they are added after filtering.

And example for perl-DBI package. In case there is more provides to be filtered, it's getting uglier.

# filter provides, which you don't want in all Perl packages
%perl_default_filter
# don't overwrite exclusion of provides, just add another
%global __provides_exclude %{?__provides_exclude}|perl\$DBI\$
%global __requires_exclude %{?__requires_exclude}|perl\\(RPC::

Perl is used everywhere and 5.14 is out

Thu, 31 Mar 2011 18:15:14 GMT

Probably, not only I have to answer again and again that Perl is life and widely used. Gabor Szabo had nice idea to make a social net, where are stored list of Perl products, companies and employers. At the moment there is growing list of companies using Perl. I was very happy to hear that even CERN is using Perl to manage their grids, and not only them. There is also list of applications, which are written in Perl, web-pages like www.imdb.com or BBC and frameworks like Catalyst, Dancer, and many more.

New Perl 5.14 was released few weeks ago and I'd like to bring it as a feature into F-16. Hopefully, this time will be possible to do rebuild of all Perl related packages in more automatic way and the whole process will take less time. So, for Perl in F-16 we are preparing:
* better generating of provides and requires by external generator
* using new rpm macros, which works since rpm4.9 and there will be in updated guidelines
* and script for automatic rebuilds based on dependencies

systemd in Fedora

Fri, 03 Dec 2010 13:18:21 GMT

Lennart's systemd doesn't replace only sysvinit/upstart, but also make chkconfig redundant if sysvinit rc scripts are thrown away and only systemd units are used.

There is the systemd packaging guideline draft, which suggest use only systemctl command which can set up correct units symlinks.

But what should be done with chkconfig, which is used for upstart/sysvinit is still not known. I'm running my rawhide machine with upstart, so I rewrote cronie scriptlets with both options for the mean time. Other possibility would be to:

package all old scriptlets into new-package, probably upstart-initscripts. Upstart maintainer has to be contact for every change.
create sub-packages for every daemon with legacy initscripts.

This should be decided by guys from Fedora packaging guidelines probably. It's sure that rawhide will be broken more than usual ;-)

# scriptlets in cronie
%post
/sbin/chkconfig --add crond
# systemd
/bin/systemctl enable crond.service >/dev/null 2>&1 || :

%preun
if [ "$1" = "0" ]; then
    service crond stop >/dev/null 2>&1 ||:
    /sbin/chkconfig --del crond
    # systemd
    /bin/systemctl disable crond.service >/dev/null 2>&1 || :
    /bin/systemctl stop crond.service > /dev/null 2>&1 || :
fi

%postun
if [ "$1" -ge "1" ]; then
    service crond condrestart > /dev/null 2>&1 ||:
    # systemd
    /bin/systemctl try-restart crond.service >/dev/null 2>&1 || :
fi

Another interesting change is that some (suggested) deamons runs in foreground (crond -n). Systemd's initscript is completely different from sysvinit's.

# cronie.systemd
[Unit]
Description=Command Scheduler
After=syslog.target

[Service]
EnvironmentFile=/etc/sysconfig/crond
ExecStart=/usr/sbin/crond -n $CRONDARGS

[Install]
WantedBy=multi-user.target

Thanks

Mon, 29 Nov 2010 18:29:13 GMT

Thanks to all of you who voted for me. I will try to do my best.

Results:
http://lwn.net/Articles/417600/
(...and thanks for fixing my name ;-) ).

Fedora elections

Sat, 20 Nov 2010 12:57:11 GMT

Today started Fedora elections. I candidate for FESCO, because I believe there should be represented also non-desktop developers.

Statements, questionnaire and IRC log with answers could be found on wiki pages.

You can vote here until the 28th November.

PSPad

Wed, 17 Nov 2010 20:44:37 GMT

Do you need from time to time good editor for working with documents or source codes on Windows? For this exceptions I always used PSPad. Even it isn't working on Linux, it' still great because:

it's freeware
it was written by Czech guy
and it can switch between different encoding and languages

fedpkg broken for new packages

Mon, 04 Oct 2010 11:26:11 GMT

Today I wonder what's wrong with fedpkg. If you are trying to create new package and you have asked for master branch only, you'll be surprised.
[marca@localhost perl-Convert-UU]$ fedpkg new-sources Convert-UU-0.5201.tar.gz
Traceback (most recent call last):
File "/usr/bin/fedpkg", line 1086, in <module>
    args.command(args)
File "/usr/bin/fedpkg", line 569, in new_sources
    mymodule = pyfedpkg.PackageModule(args.path)
File "/usr/lib/python2.6/site-packages/pyfedpkg/__init__.py", line 938, in __init__
    self.distval = self._findmasterbranch()
File "/usr/lib/python2.6/site-packages/pyfedpkg/__init__.py", line 886, in _findmasterbranch
    return(int(fedoras[-1].strip('f')) + 1)
IndexError: list index out of range

You can fix it manually by:

$ echo "Convert-UU-0.5201.tar.gz" > .gitignore
$ md5sum "Convert-UU-0.5201.tar.gz" > sources
$ curl -k --cert ~/.fedora.cert \
  -F "name=perl-Convert-UU"
  -F "md5sum=f60f49d15770503efa5ed0c81296ef2f" \
  -F "file=@Convert-UU-0.5201.tar.gz" \
  https://pkgs.fedoraproject.org/repo/pkgs/upload.cgi
$ git add .gitignore sources perl-Convert-UU
$ git commit
$ git push
$ koji build dist-f15 \
  "git://pkgs.fedoraproject.org/perl-Convert-UU#$(<.git/refs/heads/master)"

Pay attention to @ before source file in curl options. Without it you can wonder why is source missing forever. All credit goes to ppisar, who created this nice workaround.

Renaissance of Perl

Thu, 02 Sep 2010 20:09:03 GMT

This year was European conference YetAnotherPerlConference in Pisa with subtitle Renaissance of Perl. Finally, I found link on videos from there. My favourite advertiser of Perl and the maker of Perl IDE Gabor Szabo had a speech about Padre. But it was also about how start an open source project and how find community of developers.
Perl programmers could have fun with Dave Cross Vogue. Other people probably won't understand what's so funny ;-)
I was looking forward to speech about changes in the latest release - perl5.12. Very interesting was presentation about migration from old to new release of software by Allison Randal. She was speaking about different approaches by projects like apache or python and why it is not possible to push users to new releases.
Many speeches were about Perl6: here and here and here, which brings me back to Padre, which has plugins for programming in Perl6.

Perl6 is popular topic these days, so hopefully we will have Rakudo star in Fedora 14. Gerd Pokorra wants it as a feature and I'm reviewing the package. It should bring some new tools for Perl6 and replace the old rakudo.

Fedora moved to git

Fri, 30 Jul 2010 10:50:25 GMT

Hurray, Fedora is finally using git. And because maintainers are used to cvs branches, there exists handy tool fedpkg, which makes our lives easier. The full list of new fedpkg and old cvs/make commands could be found here. It looks like everything is working fine if you have the latest fedora-packager (0.5.1.0) from koji. With older releases wasn't possible to access branches with fedpkg by

fedpkg switch-branch f13

which will be my favourite command.
Also there are inconveniences like local builds

fedpkg local

because it doesn't show the build process until the build has finished. Also file .build-your-favourite-package-nvr is created after the build has finished.

Update: proposed patch and srpm.

perl 5.12.1 release candidate

Fri, 14 May 2010 16:00:07 GMT

I didn't have much time lately for further testing, but at least I tried rebuild of the latest (second) release candidate of perl 5.12.1. All tests passed, but I didn't have time to install it and try more. Brave ones can try rebuild 'rpmbuild --rebuild perl-5.12.1-123.fc14.src.rpm' and experiment on their rawhide machines.

Hopefully with help of perl-SIG will be all modules ready and it will be possible to put them into rawhide, future F-14. I suppose that in F-14 will be 5.12.1 and 5.12.0 stays only in testing buildroot.

comment to never-ending thread with hall monitor

Fri, 07 May 2010 21:28:10 GMT

Jarda at his blog and Karel, Matěj and others in mailing list nicely sum up, what I have on my mind and what's bothering me for last few weeks. I can only add censorship is simply bad and it's really sad if it has to be explained. Especially if one of the main Fedora values is freedom.

Suggested reading for this weekend: George Orwell - 1984 ;-)

Update of Perl to 5.12

Fri, 07 May 2010 21:14:47 GMT

Last few weeks I was working on update of perl to the latest stable version. I was using release candidates and prepare rebuild of all modules. This time was rebuild quite painful, because some things were deprecated. Perl changed way of releasing new versions and now will be every spring released new stable version and during the year will be released bug fixes version. I believe this helps maintainers and also perl. Some things will be deprecated and new features will be added, therefore it should help perl to keep up with other languages.

I've started mass rebuild of perl modules in my own test buildroot. I didn't find any script in fedora-packager for it, which surprised me a lot. For a next update I have to write something more sofisticated, which can handle dependencies at least for perl. This time was used only rpmdev-bumpspec, which can update spec file with defined strings. Next surprise was that chain build doesn't work at all. Anyway in many cases was needed fix specfile manually, or update version or apply patch, so chain builds can't help much.

The most common issues were (and still are):
- missing Class::ISA which has been deprecated. That was easy to fix by creating rpm. Perl is printing out that deprecated module is used, so it's at least easy to find out what's the problem.
- test t/99_pmv.t which is checking version is broken at the moment. The problem is somewhere in Perl::MinimumVersion and maybe perl(version) which can't handle new release number. But that's also easy to fix for the meantime by removing this particular test from build.
- some other modules are using deprecated syntax (e.g defined with hash %).
There are some other strange warnings, but these probably will disappear after fixing major dependency. At the moment
I found some that can't be fixed because they are old, unmaintained, but I wait with final list after I succeed with most of the easy ones.
One of the complicated is Regexp::Copy as Ian Arnell spotted. It can't handle regexps with 5.12, which means we have to left it out. So that's one module which can't make it. Hopefully not many ends up the same.

Annoying restarts of KDM

Tue, 06 Apr 2010 13:47:37 GMT

Few releases ago I started to be annoyed with kdm, because it can't be killed easily. After every killed process there was new one started. KDM is now executed with nodeamon option and it's started as an upstart "job". All configuration job files are stored in /etc/init/.

# cat configuration file for prefered DM
cat /etc/init/prefdm.conf

start on stopped rc RUNLEVEL=5

stop on runlevel [!5]

console output
respawn
respawn limit 10 120
exec /etc/X11/prefdm -nodaemon

The job can be stopped in command line by simple "stop prefdm" as root. Nice. Upstart brought many changes that are well documented, but sometimes is hard to find where to start looking.

Perl dual-lived modules in Fedora

Tue, 30 Mar 2010 16:27:41 GMT

Last few weeks we were discussing on perl-sig mailing list, how should we create better updates of perl core modules. Modules in core are e.g. ExtUtils::MakeMaker and this could be a blocker for many new interesting packages which we want see in distribution. So we had no other choice than update the whole perl package, which doesn't make no-one happy. Users complain about big and often updates and maintainers about broken tests etc.

The solution is quite simple - create dual-lived modules. Now we are in process of creating cvs branches for separated modules and hopefully we'll be able to update easily and more often. The process of creating dual-lived module is desribed here and list of blocker bug here, which tag currently targeted modules.

Moreover Chris Weyl added new macros into perl.macros, which provide better filtering of unversioned provides/requires in specfile. And he's also working on tool for automatic updates of packages from CPAN. And today he had updated database of perl modules in Fedora which looks cool and for Perl maintainers it is more handy than Fedora package database.

saving with micromiser?

Tue, 26 Jan 2010 15:40:42 GMT

I heard about a program called micromiser which should be doing the same stuff as tuned in Fedora. It's closed source
and they offer they package as binary for different distros. The only thing which is visible to user is output
in /var/log/messages, something like:
Jan 25 09:48:32 arrakis micromiser[2260]: Estimated energy saved since MicroMiser start: 0.001049kWh (31.98%)

I was suspicious how can be so much saved, so I tried for myself. From what I saw, the "saved energy" changed
and was written into log after type of work was changed e.g. from compiling to reading. So I suppose they are
doing almost the same things as we in tuned (hdparm, sysctl values, ...).

On their web-pages is a short documentation. So, they are reading from /proc and changing sysctl values according to
their findings. I tried to boot in level 3, switch off cron which could create some sort of activity and in this idle
mode measure by wattmeter with/out micromiser.

With micromiser on: 16 hours - 1.125kWh and fantastic numbers like:
Jan 25 17:35:06 arrakis micromiser[15205]: Estimated energy saved since MicroMiser start: 0.001107kWh (33.74%)
Jan 25 17:38:18 arrakis micromiser[15205]: Estimated energy saved since MicroMiser start: 0.002211kWh (33.45%)
Jan 25 17:44:42 arrakis micromiser[15205]: Estimated energy saved since MicroMiser start: 0.004403kWh (33.15%)
Jan 25 17:57:30 arrakis micromiser[15205]: Estimated energy saved since MicroMiser start: 0.008907kWh (33.48%)
Jan 25 18:23:06 arrakis micromiser[15205]: Estimated energy saved since MicroMiser start: 0.018063kWh (33.91%)
Jan 25 19:14:18 arrakis micromiser[15205]: Estimated energy saved since MicroMiser start: 0.035965kWh (33.74%)
Jan 25 20:14:18 arrakis micromiser[15205]: Estimated energy saved since MicroMiser start: 0.057724kWh (34.14%)
Jan 25 21:14:18 arrakis micromiser[15205]: Estimated energy saved since MicroMiser start: 0.078934kWh (34.09%)

But without it my computer consumed: 16 hours - 1.125kWh. Intriguing.

Stepan's speech

Thu, 21 Jan 2010 11:52:57 GMT

I found a great video from my colleagues presentation about changes in Perl in Fedora 13 in English. Video is also funny if you know some or all of used language (Czech, Slovak, Russian).
Perl changes in F-13

new Padre arrived into rawhide

Thu, 14 Jan 2010 14:42:57 GMT

Finally I found time to update Padre, which is a great IDE for programming mainly in Perl. Authors aim to create a tool for writing code and learning Perl but it could be also used for other languages. Padre provides functions like installation of modules from CPAN, plugins (e.g. Perl6 syntax) and localization into many languages (including Czech).

Padre has a friendly community on irc: irc.perl.org channel: #padre and they are working on new cool features e.g. debugger. The whole application is written in Perl5 using Wx library (perl-Wx). See more at their homepage and test it in rawhide or at least check the older version in Fedora 12 (perl-Padre).

Fun with rice

Mon, 14 Dec 2009 14:29:11 GMT

Ok, kernel rice might be, but imported by Gnome?

RPM scriptlet

Sat, 28 Nov 2009 14:59:16 GMT

The main part of rpm is clear. There is some prep, build, install, clean, check and files
and functionality of these are obvious. But sometimes the specfile includes also
scriptlets.

The easy ones are:
%pre - run before an installation of a package,
%post - run after an installation,
%preun - run before a package is removed
%postun - run after a package is removed.

Nice and clean, but what if an update is needed? Now they are executed scriptlets from both releases of package:
%pre - from a new package
%post - new
%preun - old
%postun - old
Since %preun/postun are used, the spec might break an update
e.g. if %postun is touching a configuration file, which was already configured in post by new the release.

Let's check these scriptles on an example - on my favourite package cronie.
%prep
# not needed here

%post
# usual operation - adding a daemon into services
/sbin/chkconfig --add crond

%post anacron
# all changes in scriptlets should be checked
[ -e /var/spool/anacron/cron.daily ] || touch /var/spool/anacron/cron.daily
[ -e /var/spool/anacron/cron.weekly ] || touch /var/spool/anacron/cron.weekly
[ -e /var/spool/anacron/cron.monthly ] || touch /var/spool/anacron/cron.monthly

%preun
# test is a must. The return value tells whether the daemon must be stopped.
if [ "$1" = "0" ]; then
service crond stop >/dev/null 2>&1 ||:
/sbin/chkconfig --del crond
fi

%postun
# different check, but it also checks the return value.
# In this case is controlled whether daemon was up or not before the update.
if [ "$1" -ge "1" ]; then
service crond condrestart > /dev/null 2>&1 ||:
fi

In these parts are often controlled return values of daemons, but the application has to
return correct LSB compliant values. Otherwise this check is useless. Another necessity is
command ||: Something unexpected could happened and the failure would be printed on
standard output, which could upset an user.

And now the triggers. They need an option - package with/out specified version.
The most common example is, when one package is obsoleted by another and simple
adding provides isn't enough:
Provides: vixie-cron = 4:4.4
Obsoletes: vixie-cron <= 4:4.3
# the old configuration file has to be removed:
%triggerun -- cronie < 1.4.1
# After every update from a rpm smaller than cronie-1.4.1, create a backup of crontab and remove a content
# if it matches the standard configuration.
cp -a /etc/crontab /etc/crontab.rpmsave
sed -e '/^01 \* \* \* \* root run-parts \/etc\/cron\.hourly/d'\
-e '/^02 4 \* \* \* root run-parts \/etc\/cron\.daily/d'\
-e '/^22 4 \* \* 0 root run-parts \/etc\/cron\.weekly/d'\
-e '/^42 4 1 \* \* root run-parts \/etc\/cron\.monthly/d' /etc/crontab.rpmsave > /etc/crontab
exit 0

%triggerun -- vixie-cron
# For every update from vixie-cron to cronie, copy the lock, because the daemon
# shouldn't be restart without a really good reason.
cp -a /var/lock/subsys/crond /var/lock/subsys/cronie > /dev/null 2>&1 ||:

# if the lock exist, the daemon must be restarted (it was running in the past).
# add new daemon into chkconfig everytime, when cronie was upgraded from vixie-cron
%triggerpostun -- vixie-cron
/sbin/chkconfig --add crond
[ -f /var/lock/subsys/cronie ] && ( rm -f /var/lock/subsys/cronie ; service crond restart ) > /dev/null 2>&1 ||:

The order of triggers how they are executed:
%pre - of new package
%post - make some things to newly created files
%triggerin - all triggerins which are part of other rpms
%triggerin - of this new package
%triggerun - all triggeruns which are part of other rpms
%triggerun - of this new package
%preun - of old release of this rpm
%postun - of old release of this rpm
%triggerpostun - of old release of this rpm
%triggerpostun - of this new package

And nice example again. There exist packages A and B. They share one configuration file.
The A-1 contains:
%triggerin -- B which fill config
%triggerun -- B which removes content of config
%preun which removes content of config

The B-1 contains:
%post which fill config
%preun removes content of config

Every update of B calls triggerun from A which means removing content of the config. On the first look,
it could be solved by removing triggerun from A. But what happen every after update is:
rpm -Uvv A-2 B-2
update files from A-2
update files from B-2
triggerin from A-2 -> fill config
triggerun from A-1 -> remove content of config

So, the update to A-2 break things and next update to A-3 will be working flawlesly. And this is
downside of triggers. One mistake can cause problems which can't be fixed correctly during one update.
It needs another workaround with new scriptlets.

Remove triggerun from A for next succesful updates. In B must be made bigger changes.
%post
# backup of config
cp -p config config.backup

%triggerun -- A < 2
# rewrite backup with previous
mv config.backup config

%preun
if [ $1 -eq 0 ] ; then
remove_content_of_config()
fi
[ -f config.backup ] && rm -rf config.backup || :

This solution fix the issue immediately after update, but there is lot of dependencies and it could
create unexcpeted problems in the future. Also it must be updated both packages together in one command.
Otherwise the config will be empty again.

Using triggers is easy, but nex time I'd rather pass on fixing them ;-)

Linuxalt conference

Sat, 07 Nov 2009 22:25:02 GMT

Linuxalt is a conference held annually in Brno. The great thing is, that people can vote for talks in advance and the "winners" are invited to talk.
For the first time, one lecture was in English. One member of Fedora community, Hans de Goede, travelled to our country to speak about the state of gaming in Linux. I didn't see it because I attended lecture of Michal Schmidt about Trusted Computing in Linux. I wasn't aware of these possibilities and as it was promised, I was even more confused after the lecture. The "Trusted computing" is based on hardware - on a coproser which works as a slave for confirmation of data of your Bios/OS. This is a very short (and wrong) explanation. The topic is very wide and it was quite hard to explain it in details during one hour.
The main reason, why I went there, was Milan Brož's lecture called Data Encryption. Milan had a fantastic lecture about dmcrypt with practical examples previous year, so this year it was a theoretical one. He gathered articles about popular attacks as "Cold boot" or "Evil maid" and about restoring data on old discs.
Cold boot means that attacker can steal your encryption password from computer if it still has the password in the memory. It can be used, when the computer was powered off for just 1-2s or suspended in RAM. The memory can be scanned and the password is easily obtained, because a RSA key has different entropy than the rest of data. Evil maid means stealing password by installing some spyware in your computer. All these techniques have fancy names but I don't believe that especially the evil maid attack is real in common live. What we can really do for our data is encrypt the whole disc and to switch off our laptop if we leave it somewhere for a while. The other thing is, discs are sold or thrown out as old or broken. Milan told us about experiment when were bought discs on eBay and then the data on them were restored. Huge part of it were sensitive personal or company data, which could be safe, if users really erase or encrypt them. The most important thing is, that encrypting home partition isn't enough. The password could be easily obtained from swap. Milan found some bugs during his preparation for the talk, which were already fixed in the latest version, so his talk had effect not only on listeners but also on the quality of code ;-)
The next day was speaking Dan Horák about Development of Linux distro, where he introduced the development cycle, tools, problems of packaging and also services which Fedora project offers. He made a review of the whole process for common users.
Milan Brož was speaking again, this day about LVM2 - data recovery. His speech was even more lively than the day before, because LVM is his cup of tea. He mentioned parts of file system under the hardware and started speaking about possible problems with LVM. The best part was, when he showed us "broken" LVM and with a few commands was able to fix it.

(I might be inaccurate in technical details, so I recommend reading the original source).