Nicolas' Blog

It is with great happyness that I learned Ubuntu 14.04 was now trimming SSD by default. Wondering how it was done, I’ve started digging to find out how it was done.

The choice of the Ubuntu devs has been to go with a weekly cron job calling /sbin/fstrim-all, a bash script.

I wanted to trim right away and not wait for the next cron deadline so in a terminal I went ahead and ran a
sudo fstrim-all

To my great surprise, the script ran almost instantly, weird when you know a trim usually lasts a few minutes.

So I digged out my favorite text editor and read ahead. An extract follows:


HDPARM="`hdparm -I $REALDEV`" 2>/dev/null || continue
if [ -z "$NO_MODEL_CHECK" ]; then
if ! contains "$HDPARM" "Intel" && \
! contains "$HDPARM" "INTEL" && \
! contains "$HDPARM" "Samsung" && \
! contains "$HDPARM" "SAMSUNG" && \
! contains "$HDPARM" "OCZ" && \
! contains "$HDPARM" "SanDisk" && \
! contains "$HDPARM" "Patriot"; then
#echo "device $DEV is not a drive that is known-safe for trimming"
continue
fi
fi


So I rewinded a bit and read this “explanation” at the top of the file.


# As long as there are bugs like https://launchpad.net/bugs/1259829 we only run
# fstrim on Intel and Samsung drives; with --no-model-check it will run on all
# drives instead.
if [ "$1" = "--no-model-check" ]; then
NO_MODEL_CHECK=1
fi


So, what is happening is simple, either you have an Intel, a Samsung, a SanDisk, a Patriot or an OCZ SSD in which case the ext3, ext4, xfs and btrfs filesystems on the drive will be trimmed or you have another brand of SSD and Ubuntu wont trim it for you “yet”.

You can force the trim of course if you edit /etc/cron.weekly/fstrim and remplace the following

exec fstrim-all

Adding the –no-model-check flag to it:

exec fstrim-all --no-model-check

As most of the issues preventing generalized trimming happen under high I/O, my recommendation however is to run the trimming at a time you decide to do it (i.e. when you leave the machine idle for a while).

I just got my hands on the Rasberry Pi model B, a tiny PCB with great computing capabilities embarking a 700Mhz ARM11 CPU on a Broadcom BCM2835 SoC with 512Mb of RAM. I got it from a local retailer in Toulouse: Snootlab with convenient free pick-up avoiding shipping costs.

What you get is a micro-USB powered device capable of running any distribution of Linux built for the ARM architecture. Dual USB ports, Ethernet connectivity, HDMI/Analog output as well as a stereo audio jack.

I also got the following:

• Lexar SDHC Premium 8 Gb
• D-Link DWA-121 Wireless USB dongle
• Black Raspberry Pi case

What was my goal? Replacing a disfunctional HTC Dream + Android Airplay receiver app that I was using to get music in my Kitchen/Dining Room. The problem with the old Android Phone hooked up to a pair of good speakers setup was that for some reason, the app would sometime close and I would need to manually re-launch it making the setup inefficient in practice.

I started by following this guide. I adapted it slightly:

• Never hooked a display or keyboard, got the IP the Pi was assigned by my DHCP server and SSHed yo it directly
• Configured the wireless through wpasupplicant.conf rather than using the GUI
• Didn’t go for a USB soundcard as recommended

Despite what I have read, Audio out of the Pi is not that crappy. It is not the best one could expect of course but for talk shows or low bitrate streaming, it’s more than enough.

Once I had the Pi setup for Airplay, I couldn’t just stop there and had to go ahead and setup a few other ways to stream music to it, specifically, from non-apple devices.

I’ve tried BubbleUPNP which is an excellent piece of software, one of the only UPNP software I saw that could query my Linksys UPNP compatible wireless router to discover UPNP services on other subnets. The only downside to BubbleUPNP is the “closed source” aspect which in the end made me look for “open” alternatives.

The only feature I wanted was to be able to control the audio playing on the Raspberry Pi from my Android mobile phone. I decided to setup an MPD instance on the Pi and combine that with the excellent MPDroid app that does just what I needed. Now, it turns out I already had an MPD deamon running elsewhere on my home network and MPDroid is unable (or I couldn’t find how) to use two different servers on the same Wifi SSID. For now, I have resigned to having two MPD player apps on my phone, one for each server.

Once I had it all setup, I gave the sdcard to fsarchiver and saved that to a backup location, just in case the SD card goes dead sometime and I have to change it.

What do I want to do next? Get another one and get XBMC on it to replace/complement my current MythTV HTPC setup. I might get a Model A this time and replace the Model B I have as an Airplay receiver in the Kitchen to use it with XBMC, the 256Mb extra RAM will certainly come in handy.

After upgrading to Ubuntu Jaunty yesterday, I found that none of the pictures I had in F-Spot worked anymore. Ok, I have my pictures stored on a remote machine and I access them through SFTP but that shouldn’t be a reason for it to break so easily.

It appears that between Intrepid and Jaunty, the GVFS mounts done through the Places->Connect to Server… have changed mountpoint. Ideed, when they used to be mounted to /home/nicolas/.gvfs/sftp on someserver/ they are now mounted to /home/nicolas/.gvfs/sftp for nicolas on someserver/ and this broke F-Spot’s database.

In order to fix that, I went through the following:

1. Backed up /home/nicolas/.gnome2/f-spot in case I messed it up. 
2. sudo apt-get install sqlitebrowser 
3. sqlitebrowser /home/nicolas/.gnome2/f-spot/photos.db 
4. File-> Export -> Database to SQL file (data.sql in my case) 
5. Replaced all occurences of /home/nicolas/.gvfs/sftp on someserver/ 
with /home/nicolas/.gvfs/sftp for nicolas on someserver/ using 
a text editor

Removed two lines that would not allow me to restore the database:

CREATE TABLE sqlite_sequence(name,seq); 
INSERT INTO sqlite_sequence VALUES('photos',3306);

Restored the database:

sqlite3 -init data.sql data.db

Finally, overwrote photos.db with data.db

mv data.db photos.db

This fixed it and I got my photos back ;) In order to be able to import new photos however, I had to change the default location which was not found anymore as it still refered to /home/nicolas/.gvfs/sftp on someserver. This was done in the preferences window!

Happy F-Spotting

Subversion does not allow fine grained permission management natively through svnserve. For example, it is not possible to disable read/write access to a certain SVN directory for a given user.

However, you generally don’t want all your users to be able to write changes anywhere else than the branch they’re working on until they finish their work and prove that they’re able to “not screw everything up” ;)

This can be achieved by using pre-commit hooks in subversion that will make transactions fail in directory where users don’t have permissions.

Read more »

Ant is a great tool for working with Java projects.

In fact, maybe the need for it does not come as an evidence to you if you’ve been using an IDE like eclipse which does all the path resolution and compilation by some kind of magic! However, when not all developers on a project use the same IDE or if you want to have your source on an automated build system, you’ll find ant very powerfull. With ant, I’m able to have an automated build system compile all my sources, run all my jUnit tests, generate Javadoc, several JARs (with javadoc, with source, without javadoc, …) and run a CheckStyle utility on the source ! Moreover, I can run all this by typing a single command: ant

Read more »

In order to setup an ssh tunnel between port 8080 on the local machine and port 80 on the SSH server, you can go and type:

ssh -L8080:localhost:80 machine.example.org -l yourUsernameAtMachineDOTexampleDOTorg

But if you follow the law of the least effort, you won’t want to type that in everytime you need to establish the tunnel, you would rather type:

ssh tunnelName

and have SSH do everything by itself for you! This can be acomplished by inserting the following lines in your ~/.ssh/config file.

Host tunnelName        
User yourUsernameAtMachineDOTexampleDOTorg        
Hostname machine.example.org        
GatewayPorts yes        
LocalForward 8080 localhost:80

Notice the GatewayPorts yes directive, this is optional for the tunnel to come up but allows you to use the tunnel from other machines than just the one that establishes it!

For some of the latest projects I’m working on, it is required that I quite frequently access various machines that are located behing a corporate firewall, on a privately adressed subnet inside a company.

One option I had but got fed up with quite fast was :

 outside$ ssh nicolas@gateway 
 gateway$ ssh root@box1 
 box1# 
or  
 ssh -t nicolas@gateway ssh root@box1

Once again, SSH came in handy and I found that everything was there for me to have that first hop and re-ssh done automatically. Just added the following to my ~/.ssh/config

 Host gateway       
 Hostname gateway.example.org       
 User nicolas 

 Host box1       
 Hostname 192.168.0.1       
 User root       
 ProxyCommand ssh gateway nc -w 1 %h 22

This, coupled to ssh’s passwordless authentication made my day a lot easier ;)