sphinx mysql command line

I’ve been searched for a method on how to work using sphinx from console, I needed to access the interface in order to see what’s indexed, how is stored and  also test my searches.

After my google searches i’ve ended up with this.

mysql -h --prompt 'SphinxQL>' --port 9312

Horizontal scaling using Db Charmer

I was looking for a way to scale horizontally a Ruby on Rails application, and i have tried several methods to scale it. A method would be using a MySQL cluster, but that would require some serious database administrator skills, which unfortunately i don’t have.

dbreplication173Mainly i have an application that is read intensive (80% reads vs 20% writes) so, i have considered to use a MySQL master – slave configuration. The problem is that there is nothing about it in Rails documentation, however, after a short look in ruby-toolbox.com I have discovered that I am not the only one who encountered this problem.

I have tried octopus as my first choice, but i have soon discovered that is not fit for my application. For some reasons, not all my “read” queryes were passed to my slave connection. I have tried to see why, but because I was kind of pressed by time, i have dismissed this gem, even if i love the simplicity of the models.

After dismissing octopus, I have tried db charmer gem, which is pretty active. This is yet another Active Record Sharding gem that offers you the possibility to split database reads and writes.

The method i have chosen for my first try was to split my actions that were 100% reads, and push them to a slave. That was pretty simple using a before filter in my rails controllers.

class ProfilesController < Application
  force_slave_reads :only =>  [ :show, :index ]

This action allowed me to scale the application by keeping the same amount of servers, but the main effect was a drop in the response time of the applications.

The second action i have taken was to get all the heavy queries like counts out of the mysql master server and move them to slave.

class User < ActiveRecord::Base
  def some_some_heavy_query
    self.on_slave.joins(:profile, :messages).count(:group => ['messages.thread_id'])

In my enthusiasm of having a mysql slave I have thought that it would be nice to have “ready” 3 slave instances in my config. I have later realised that this “optimisation” caused problems because those 3 connections multiplied by the number of max_child in my apache configuration and also multiplied by the number of the servers exceded the number of the max_connection on my mysql slave server.

After a small fix in my database.yml files I was back online with a more performant application.

Database optimization technique and how storage engine works

Today i was discussing with a colleague of mine about a mysql database optimization technique, and some of the things i have learned about how mysql works on big table datasets. This article is not about “mysql”, is not about “Mongodb”, but more about the principle behind of it.

I will try to give you an idea about this using i using a book library example.

In any database engine you might have tables / collections that have million of rows, which needs to be sorted, ordered, filtered using conditions, and every time is a pain finding the right solutions.

I have indexes, so is not a problem for me.

Well, if you’re reading this, then you might be wrong. First of all, if it is a query (and by query i mean any SQL / NoSQL statement meant to retrieve informations from the table) that is not ran on a constant basis, then most probably, your database engine does not have it cached. This means every time you’ll run that query the database engine will actually look in the store and actually filter all of your statements.

Yes, but i am running statements using indexes …

Well … imagine that you’re in a library, and you ask the librarian to give you a certain book, knowing just the author. He will leave the office and go to search your book, knowing just a fragment of the specs. He will know that he needs to go to a certain row, but will not know which shelf, or what is the position of the book, so he will lose some time to read all those authors in that shelf, until he provides you the book.

This is happening in any Database engine, when you ask for a book that is not read too often.

Now imagine that you’re asking a book that was just returned by another reader. Now, the librarian might have not put the book back the shelf, so, he might give you faster because is just few meters away.

Yeah… but the database engine is using “index tables”… there are pointers … etc

Well … yeah and no … imagine that you already know that the book is on a row and a shelf… you’ll still lose time by moving there, and also, searching on that shelf. The story is more like the one above.

Ok … might get it … but how can i get a random record?

Well, in MySQL and other SQL languages you might do something like this :

SELECT @v:=RAND() * (SELECT MAX(id) FROM your_table)


SELECT * FROM your_table WHERE  id > @v AND  the_rest_of_your_condition = 'something';

in MySQL RAND() function will always return a real number between 0 and 1, which means that any number will get generated will be actually between record 1 and your MAX(id).

This kind of hack is useful when your need to perform a query like…

SELECT * FROM your_table WHERE your_condition = 'something' ORDER BY RAND();

At this very moment i don’t know how could you achieve that in a NoSQL engine.

Well my problem is a pagination issue.

Some pagination ...

Good luck with that! Maybe you have something like:


to find the number of rows that you have in your database… Well That’s bad … a mysql database optimization technique would be to use CALC_FOUND_ROWS function call which might solve your issue like this:

SELECT CALC_FOUND_ROWS * FROM my_table WHERE my_conditions = '1' LIMIT 0, 10;

Then you can use


However, your problem is still not fixed … On big datasets, you will still have a MySQL issue when using OFFSET or LIMIT. A query like:

SELECT * FROM my_table LIMIT 10000, 10;

is similar with

SELECT * FROM my_table OFFSET 10000 LIMIT 10;

Which in MySQL, will mean that the storage engine will actually load in memory and read 10010 rows. And once the offset gets higher, and higher, so the time waiting for the response will increase.

Those are just 2 optimisations that you can perform at least on MYSQL to get a faster response. There are more ways to optimize your mysql queries, but, because is not a “one size fits all” solution, i cannot speak about them, as i did not encountered them … YET.

Mysql field types and their charsets

When having to administer a big database having tables that contains millions  of rows, a big issue emerges. Optimizing the tables and the data stored.

A short version would tempt me to say: Don’t use the utf8 encoding for all the table. Use it for the damn fields where you really need it.

Longer version would require some calculus to be performed, and that would lead to a better understanding of what happens in a MySQL server.

When varchar is it used, the storage engine, allocate exactly the amount of bytes required to store the value, no matter the charset. However during the creation of the temporary tables and internal buffers (during joins, ordering, basic reading from the storage and other operations), the maximum amount of bytes for the column is allocated.

This means is you have a column that is Varchar(255) utf8, and you write a single letter in it, the disk will need just 2 bytes  to store it: 1 byte for its length and 1 byte to store the value. The issue becomes trickier, when you need to fetch that value, as MySQL not knowing what it is stored in that field, will allocate 768 bytes (you have read right 768… 256 bytes the length of the field * 3 bytes per utf8 character).

Assume you have a table that contains 2 columns of type varchar(255) utf8 and you might run a query like this:

SELECT column1, column2 FROM table1 ORDER BY column1 DESC;

If you don’t have index on column1 and the table has 1 Million rows, MySQL would require to create a temporary table to do the sorting, which would lead to a temporary table of something like 1.43 Gb. The calculus is symple:

768 (Bytes) *  2 (columns) * 1.000.000 (rows) = 1536000000 Bytes  which would give a roughly 1.43 Gigabytes of temporary table on disk.

If you have values that are actually that long, it makes sense, but if the maximum length is around 30 – 50 (let’s consider a username or a password or even an email field) the temporary table would be roughly around 290 Megabytes

150 (Bytes) * 2 (columns) * 1000000 (records) = 300000000 Bytes which is a give a roughly 290 Megabytes.

It is obvious that 290 Megabytes are faster to read than a 1.43 Gigabytes.

Of course this could be pushed a little bit more, and if you don’t need utf8 encoding, you could  switch to latin1 which would mean around 95 Megabytes.

The above stuff is valid even you have a table of 20 Megabytes in size on disk.

Lessons that i am trying to express:

a) use the charset you need!
b) use the length of varchar in a responsible matter, don’t make it just “as much as possible”

Read more about: MySQL Storage Requirements

This post is written with help from my friend: Rene Cannao