Normally, you debug, and when that’s done, you profile to speed things up.

That said, today I was confronted by a problem where profiling actually helpped the debugging.

A page, that should be working 100% from memcache was hitting mysql, 2 million times a day.

This page was on a serveur that couldn’t be debugged remotly. Two options were available, I could either put “print”s everywhere, and hoped to get lucky, or I could run the pager through xdebug’s profiler.

Loading the resulting profile into KCacheGrind, I had a detailled run down of every code path taken available to me. All that was rest was to lookup all mysql calls and work back up from caller to caller to identify the bug.

If you have a high bandwidth site, and are seeing high load, lots of time waiting for io and nginx processes in the D state a lot, here’s something you could try:

First of all, install iotop and make sure nginx is the culprit (lots of IO for nginx, not much for the rest)

Then, go and have a look at your NGinx configuration, (nginx.conf), especially the worker_processes

You can try pushing the value up to (and maybe even over) the number of cores you have in your system. This spreads the load more evenly, and allows you to dramatically increase disk throughput (doubling isn’t uncommon).

This worked for me (going from 3Mb/s to 7Mb/s peak). That said, YMMV.

Feel free to leave a comment to say if/how this was (un)helpful

That said, you’re ruining the fun for everyone else, so please go play on someone elses lawn.

It doesn’t cost me much to spam your comments, and to change the challenge response, so don’t think I’ll just give up.

Regards

My deepest apologies .

To understand what happened, here’s a quick lowdown on how my setup works:

I’m self hosting this blog as well as my mail on a converted EasyGate that I got as a beta tester. The machine is connected onto my internet router (a Trio 3D antique) and then onto the internet. My ISP provides a temporary IP address, and as such, I need to use a dynamic DNS record to be able to keep things updated.

Now, the Trio 3D (as with most of it’s features) has a half assed (excuse my French)  DynDNS system built in. I say half assed because it doesn’t accept any provider except the historic DynDNS and because it quite often forgets to update. Last week it completely stopped updating, and as an added bonus, I couldn’t get back into DynDNS to do things manually.

As a work around, I set up OVH (who are my registrar) to route my DNSes manually until I got a permanent fix. That worked until the IP changed Tuesday or Wednesday.

Now things are sorted out. I’m using OVH’s own DynDNS servers (much simpler) and am using a home built client to avoid all the dependancies other clients bought in.

The client is written in bash and depends on bash, sed and wget to work. It’s best used in a cron job, but can also be run on interface specific events if need be.

It uses a web based IP fetcher to work out your IP adresse, and if it’s changed since you last run the script, it updates your DynDNS records for you.

Without further ado, here’s the code:

#/bin/bash

USERNAME='myusername'
PASSWORD='mypassword'
HOSTS=('first_url.test.com' 'second_url.test.com')
SERVER="www.ovh.com"

#
#
#

CURRENT_IP=`wget -O - 'http://checkip.dyndns.org/' 2>/dev/null | sed -e 's/.*Address: \([^<]*\).*/\1/'` if [ -f /tmp/my_ip ]; then 	OLD_IP=`cat /tmp/my_ip`; fi # # Don't update unless ip has changed # if [ "x"$CURRENT_IP == "x"$OLD_IP ]; then 	exit; fi; echo "IP changed from " $OLD_IP " to " $CURRENT_IP echo "Saving new IP" echo $CURRENT_IP > /tmp/my_ip

for HOST in ${HOSTS[@]};
do
	echo "Updating " $HOST " to " $CURRENT_IP
	wget -O - "http://${USERNAME}:${PASSWORD}@${SERVER}/nic/update?system=dyndns&hostname=${HOST}&myip=${CURRENT_IP}&wildcard=NOCHG&mx=NOCHG&backmx=NOCHG" 2>/dev/null
	echo ""
done

Et voila, I hope this helps someone out there

Note: For the rest of this post:

• A square is the un-devidable unit containing 1 number
• A section is a 3×3 grid of squares
• A column is a 10 square vertical strip
• A row is a 10 square horizontal strip
• A board is the 9×9 sudoku board

There a few simple rules that go into solving these puzzles. Namely :

• What are the possible values for this square (based on row, column and section)
• What are the possible places for this number (in this section)

The two questions aren’t identical, since a numbers position can be influenced by the positions of numbers in other columns or rows that share the section.

The first question comes down to :

1. Take the ensemble of possible numbers
2. Remove all that are already in the current row
3. Remove all that are already in the current column
4. Remove all that are already in the current section
5. if there is only one left, put it in the square
6. lather
7. rinse
8. repeat

The second question comes down to:

For each number not entered in each section

1. Check for number in the rows that pass through this section, eliminate the squares that belong to them if found
2. Check for number in the columns that pass through this section, eliminate the squares that belong to them if found
3. if only one square is left, put the number in it
4. …
5. profit

OK, mabey that one’s a bit vague, I’ll try to make it clearer with a drawing

Example of a row and column problem

In the above image, concentrate on the bottom middle section. If you look, you’ll notice that there are 4s in the top 2 rows of this section, we can conclude that our 4 must be on the bottom row. We also notice that there is a 4 in the left most column of this section, which only leaves the bottom middle square as a possibility for this 4.

By combining the 2 rules stated above, we should be able to solve any sudoku puzzle.

Any how, that’s all for now. Proof by code will be comeing in an update.

(No deb instructions this time since I’ve passed on to arch, which has a perfectly functional package).

A couple of thing seem to have changed though, namely:

• Bitrates are now in bytes/seconde rather than kbytes/seconde (unless otherwise stated)
• AAC is no longer an encoder module (use libfaac)
• mp4/amr containers don’t seem to work to well with libfaac, use 3gp instead to be safe.

Which gives:

 ffmpeg -i input.avi -vcodec h263 -b 409600 -s 352x288 -acodec libfaac -ab 65536 -ar 48000 -ac 2 -f 3gp output.3gp

Happy encoding

Go to your browser, type in www.google.com, you get the google home page. Everythings normal there.

Now, change your url, ad a dot (so that it reads www.google.com.) . The page still loads.

This is risky, since it can lead to a duplicate content malus on google, but is perfectly normal DNS behaviour.

The actual FQDN is www.google.com. . It’s just that, to make our lives easier, the last dot is optional.

Unfortunatly, there’s no way to get rid of them (even 301s seem flaky).

That’s all, just so that the next time you see bizzars urls, you have a reason for them all ready for your boss

For once, the new machine is less powerful than the old.

This blog is now running on an opengate with 512Mb ram and 512Mb of disk (!)

The small install size (systme + data <400Mb) is obtained by using a basic Debian install, and by taking a rm to most of the documentation and things like that.

I might even try to get it down to under 300Mb soon.

The new test is a multiplication, but it isn’t in math notation, so it should hold out a little better