17.2 Non-Volatile Databases

Some information is so important that you cannot afford to lose it. Consider the name and password for authenticating users. If a person registers at a site that charges a subscription fee, it would be unfortunate if his subscription details were lost the next time the web server was restarted. In this case, the information must be stored in a non-volatile way, and that usually means on disk. Several options are available, ranging from flat files to DBM files to fully-fledged relational databases. Which one you choose will depend on a number of factors, including:

• The size of each record and the volume of the data to be stored

• The number of concurrent accesses (to the server or even to the same data)

• Data complexity (do all the records fit into one row, or are there relations between different kinds of record?)

• Budget (some database implementations are great but very expensive)

• Failover and backup strategies (how important it is to avoid downtime, how soon the data must be restored in the case of a system failure)

17.2.1 Flat-File Databases

If we have a small amount of data, sometimes the easiest technique is to just write this data in a text file. For example, if we have a few records with a fixed number of fields we can store them in a file, having one record per row and separating the fields with a delimiter. For example:

Eric|Cholet|[email protected]
Doug|MacEachern|[email protected]
Stas|Bekman|[email protected]

As long as we have just a few records, we can quickly insert, edit, and remove records by reading the flat-file database line by line and adjusting things as required. We can retrieve the fields easily by using the split function:

@fields = split /\|/, $record;

and we can put them back using join:

$record = join '|', @fields;

However, we must make sure that no field uses the field separator we have chosen (| in this case), and we must lock the file if it is to be used in a multiprocess environment where many processes may try to modify the same file simultaneously. This is the case whether we are using mod_perl or not.

If we are using some flavor of Unix, the /etc/passwd file is a perfect example of a flat-file database, since it has a fixed number of fields and most systems have a relatively small number of users.^[1] This is an example of such a file:

^[1] Disregard the fact that the actual password is stored in /etc/shadow on modern systems.

root:x:0:0:root:/root:/bin/tcsh
bin:x:1:1:bin:/bin:
daemon:x:2:2:daemon:/sbin:
adm:x:3:4:adm:/var/adm:
lp:x:4:7:lp:/var/spool/lpd:

: is used to separate the various fields.

Working with flat-file databases is easy and straightforward in plain Perl. There are no special mod_perl tricks involved.

17.2.2 Filesystem Databases

Many people don't realize that in some cases, the filesystem can serve perfectly well as a database. In fact, you are probably using this kind of database every day on your PC—for example, if you store your MP3 files categorized by genres, artists, and albums. If we run:

panic% cd /data/mp3
panic% find .

We can see all the MP3 files that we have under /data/mp3:

./Rock/Bjork/MTV Unplugged/01 - Human Behaviour.mp3
./Rock/Bjork/MTV Unplugged/02 - One Day.mp3
./Rock/Bjork/MTV Unplugged/03 - Come To Me.mp3
...
./Rock/Bjork/Europa/01 - Prologue.mp3
./Rock/Bjork/Europa/02 - Hunter.mp3
...
./Rock/Nirvana/MTV Unplugged/01 - About A Girl.mp3
./Rock/Nirvana/MTV Unplugged/02 - Come As You Are.mp3
...
./Jazz/Herbie Hancock/Head Hunters/01 - Chameleon.mp3
./Jazz/Herbie Hancock/Head Hunters/02 - Watermelon Man.mp3

Now if we want to query what artists we have in the Rock genre, we just need to list the files in the Rock/ directory. Once we find out that Bjork is one of the artists in the Rock category, we can do another enquiry to find out what Bjork albums we have bought by listing the files under the Rock/Bjork/ directory. Now if we want to see the actual MP3 files from a particular album (e.g., MTV Unplugged), we list the files under that directory.

What if we want to find all the albums that have MTV in their names? We can use ls to give us all the albums and MP3 files:

panic% ls -l ./*/*/*MTV*

Of course, filesystem manipulation can be done from your Perl program.

Let's look at another example. If you run a site about rock groups, you might want to store images relating to different groups. Using the filesystem as a database is a perfect match. Chances are these images will be served to users via <img> tags, so it makes perfect sense to use the real path (DocumentRoot considerations aside) to the image. For example:

<img src="/images/rock/ACDC/cover-front.gif" alt="ACDC" ...>
<img src="/images/rock/ACDC/cover-back.gif"  alt="ACDC" ...>

In this example we treat ACDC as a record and cover-front.gif and cover-back.gif as fields. This database implementation, just like the flat-file database, has no special benefits under mod_perl, so we aren't going to expand on the idea, but it's worth keeping in mind.

Too Many Files There is one thing to beware of: in some operating systems, when too many files (or directories) are stored in a single directory, access can be sluggish. It depends on the filesystem you are using. If you have a few files, simple linear access will be good enough. Many filesystems employ hashing algorithms to store the i-nodes (files or directories) of a directory. You should check your filesystem documentation to see how it will behave under load. If you find that you have lots of files to store and the filesystem implementation won't work too well for you, you can implement your own scheme by spreading the files into an extra layer or two of subdirectories. For example, if your filenames are numbers, you can use something like the following function: my $dir = join "/", (split '', sprintf "%02d", $id)[0..1], $id; So if you want to create a directory 12345, it will be converted into 1/2/12345. The directory 12 could become 0/0/12, and 124 could become 0/1/124. If your files have a reasonable distribution, which is often true with numerical data, you might end up with two-llevel hashing. So if you have 10,000 directories to create, each end-level directory will have at most about 100 subdirectories, which is probably good enough for a fast lookup. If you are going to have many more files you may need to think about adding more levels. Also remember that the more levels you add, the more overhead you are adding, since the OS has to search through all the intermediate directories that you have added. Only do that if you really need to. If you aren't sure, and you start with a small number of directories, abstract the resolution of the directories so that in the future you can switch to a hashed implementation or add more levels to the existing one.

17.2.3 DBM Databases

DBM databases are very similar to flat-file databases, but if all you need is to store the key/value pairs, they will do it much faster. Their use is much simpler, too. Perl uses tie( ) to interact with DBM databases, and you work with these files as with normal hash data structures. When you want to store a value, you just assign it to a hash tied to the DBM database, and to retrieve some data you just read from the hash.

A good example is session tracking: any user can connect to any of several mod_perl processes, and each process needs to be able to retrieve the session ID from any other process. With DBM this task is trivial. Every time a lookup is needed, tie the DBM file, get the shared lock, and look up the session_id there. Then retrieve the data and untie the database. Each time you want to update the session data, you tie the database, acquire an exclusive lock, update the data, and untie the database. It's probably not the fastest approach, and probably not the best one if you need to store more than a single scalar for each record, but it works quite well.

In Chapter 20 we give some important background information about DBM files and provide a few examples of how you can benefit from usingDBM files under mod_perl.

17.2.4 Relational Databases

Of course, the most advanced solution is a relational database. But even though it provides the best solution in many cases, it's not always the one you should pick. You don't need a sledgehammer to crack a nut, right?

Relational databases come in different implementations. Some are very expensive and provide many tools and extra features that aren't available with the cheaper and free implementations. What's important to keep in mind is that it's not necessarily the most expensive one that is the best choice in a given situation. Just as you need to choose the right database structure, you need to choose the right relational database. For example, ask yourself whether you need speed, or support for transactions, or both.

It makes sense to try to write your code in such a way that if later in the course of development you discover that your choice of relational database wasn't the best, it will be easy to switch to a different one.

mod_perl greatly helps work with relational databases, mainly because it allows persistent database connections. We'll talk extensively about relational databases and mod_perl in Chapter 20.