So far, so good. But considering using replication for backup purposes, which is something I guess is quite common, the binlog of changes to the database is necessary to be able to do a point in time recovery. To get that binlog, you have to add log_slave_updates. Unless you have backup of the binlog files from the master, of course, but the point was using the replicate for backup, wasn’t it?

If you are unsure, compare the size of the binlog files on the master and the backup replicate slave.

In this release, I have merged changes from the upstream rpm, and added native systemd support for f17 and rawhide. It also builds nicely for epel5 and epel6, providing packages quite similar to those available from the varnish project repo.

As epel does not allow changes in a package API after release, varnish-3.0.2 won’t be available through epel5 or epel6, so use the varnish project repo, or my precompiled packages for epel 4, 5 and 6 available here.

As always, feedback is very welcome.

lsof +L1 | egrep 'bin/|lib/'

Now, on later Linux distributions, the usage of prelink has changed this, so one usually gets a lot of false positives, and could never trust the result of that good old lsof output.

Finding running executables is possible using some perl magic (Yes, I’m pretty sure you perl guys can write this more compressed) along the lines of this, at least on RHEL5:

perl -e ' for $i (glob "/proc/[1-9]*/exe") { $f=readlink $i; if ( $f=~ /([^\0]+)\0.*deleted/ ) { print "$1\n" }} ' | sort | uniq

But this won’t help us finding what service to restart after a dependency library was updated. So I finally wrote this script to help me. My boxes are mostly Fedora and RHEL, so it uses the fact that installed binaries on Red Hat based systems have installation time stored in the rpm database (rpm tag %{INSTALLTIME} ). This script abuses rpm heavily, and may use some time to finish on a busy system.

http://users.linpro.no/ingvar/check_newlibs

Test run:

[root]# /home/ingvar/check_newlibs
Warning: Needs restart: /sbin/agetty, pids 6067
Warning: Needs restart: /usr/bin/tail, pids 7315
Warning: Needs restart: /usr/bin/vim, pids 19759
Warning: Needs restart: /usr/sbin/sendmail.sendmail, pids 10645 10637
Warning: Needs restart: /usr/sbin/acpid, pids 5259
Warning: Needs restart: /usr/sbin/crond, pids 5567
Warning: Needs restart: /bin/bash, pids 26074 17731 16848 15718 30753 6120 32704
Warning: Needs restart: /sbin/mingetty, pids 6071 6069 6076 6068 6072 6070
Warning: Needs restart: /sbin/portmap, pids 5082
Warning: Needs restart: /usr/sbin/smartd, pids 20948
Warning: Needs restart: /sbin/multipathd, pids 20170
Warning: Needs restart: /usr/sbin/atd, pids 5969
Warning: Needs restart: /usr/sbin/sshd, pids 19863
Warning: Needs restart: /usr/libexec/mysqld, pids 17775

This fixes it:

# echo 'supports_quiet=false' >> /etc/default/grub
# update-grub

Now, the HP iLO 3 should support IPMI, so after a bit of fiddling around, I came up with this, and it actually works. The following was executed on RHEL5.

First find the LAN channel

# for i in `seq 1 14`; do ipmitool lan print $i 2>/dev/null | grep -q ^Set && echo Channel $i; done

Channel 2

So, on this system, channel 2 is the LAN channel.

# ipmitool lan print 2

Set in Progress         : Set Complete
Auth Type Support       :
IP Address Source       : DHCP Address
IP Address              : 0.0.0.0
Subnet Mask             : 0.0.0.0
MAC Address             : c0:ff:ee:c0:ff:ee
BMC ARP Control         : ARP Responses Enabled, Gratuitous ARP Disabled
Default Gateway IP      : 0.0.0.0
802.1q VLAN ID          : Disabled
802.1q VLAN Priority    : 0
Cipher Suite Priv Max   : Not Available

Okay, so if you have a DHCP server on your management network, you may be content with this, and just give it an address by DHCP. I wanted to set a static address, though.

# ipmitool lan set 2 ipsrc static
# ipmitool lan set 2 ipaddr 192.168.42.36
# ipmitool lan set 2 netmask 255.255.255.0
# ipmitool lan set 2 defgw ipaddr 192.168.42.1
# ipmitool mc reset cold

That’s it actually. Exchange the LAN channel and network addresses with your own, of course.

After four hours, the result exceeded all expectations! Happy advent everybody!

Update: The Friendly Fredrik posted some more pictures of the building process.

Parts ready for the oven

That's a lot of parts

Assembly by melted sugar. Hot stuff!

AT-AT Walker #1

AT-AT Walker #2

AT-AT Walker #3

AT-AT Walker #4

AT-AT Walker #5

AT-AT Walker #6

So if you wonder, @ingvarha – that’s me.

I recently built jemalloc-2.2.5 for fedora and epel6. Updates are submitted to bodhi, and will trickle down to their respective test (and later stable) repos in a few days.

Because of som ppc specific lacking functionality, jemalloc is still not updated for for epel5. epel5 supports ppc/ppc64 and needs those fixed to get a working update. For details, see Jason’s answer in April. As stated before, I have access to 32 and 64 bit ppc hardware, and would be happy to test that if the code appears.

Because of this, I have built x86_64 and i386 packages of jemalloc-2.2.5 for rhel5 and clones. The packages are available here:

http://users.linpro.no/ingvar/jemalloc/2.2.5/el5/

I found sharkcz‘ method to install fedora 16 on a ppc64 system running older versions of Fedora or RHEL, and was inspired to dig out two old 32bit ppc boxes from the attic.

I started out with my good old IBM B50. The 7046-B50 is a CHRP 604r (32bit ppc) running at blazing 375MHz. It has that 16MB-should-be-enough-for-everybody OpenFirmware limit, so it can’t boot newer Fedora install media. It may be able to boot Karsten’s Fedora 16 ISO image once he gets a 32bit version available :-)

The latest Fedora version that installs through normal media is Fedora 9. So I installed a minimum package set, and yum-upgraded through f10 and f11 to f12. From there I ran sharkcz’ installroot trick, and got a working f16 environment using the mash package repo.

To get a clean boot, better add lvm2 and e2fsprogs to the rootfs. Also, bind-mounting /sys, /boot, /dev and /proc into the installroot while installing makes it possible to chroot into it, and run yum self-hosted.

Not that hard, but it took a few hours to loop through the upgrade cycles. I had to run dracut with -H to get an initramfs small enough to boot.

Blinded by success, I took an Apple G3 iBook that I got for free. This white laptop is equipped with a 750FX CPU running at 700MHz and 384MB RAM. It had no PSU, but I found one at DX that worked fine.

The iBook installed without problems with the standard f12 network installer CD. Booted again with the standard f12 CD1 in rescue mode, and shrinked the root lv to make place for another fedora instance. Then did the yum installroot trick again as described above. Finally found that the new root would not boot. After a lot of silly debugging, I found that the “pmac” IDE driver that I needed is not probed automatically via udev, as it does not support libata. So, how do you make dracut probe another driver? Via the boot option rdinsmodpost, and I guess you had never heard about that either.

After a bit of trying and failing, I got a working XFCE desktop. Now, if I only had an airport slot-in unit, I could have used this on the go. If you are content with epiphany for browsing and abiword and gnumeric for office apps (or just use emacs for everything), this is actually working.

The iBook running Fedora 16

varnish-3.0.0 was released some weeks ago. I have built packages for Fedora and epel4/5/6. Packages may be found at the usual http://users.linpro.no/ingvar/varnish/. The rhel packages require some dependencies pulled from epel.

Varnish Software produces their own packages, based on the specfile I maintain for Fedora. The changes from their rpm spec are mostly cosmetic to fit better to Fedora’s packaging standards.