- Introduction
- File Transfer
- Shared Database
- Remote Procedure Invocation
- Messaging
This chapter is from the book
This chapter is from the book
This chapter is from the book
File Transfer
by Martin Fowler
An enterprise has multiple applications that are being built independently, with different languages and platforms.
How can I integrate multiple applications so that they work together and can exchange information?
How can I integrate multiple applications so that they work together and can exchange information?
In an ideal world, you might imagine an organization operating from a single, cohesive piece of software, designed from the beginning to work in a unified and coherent way. Of course, even the smallest operations don't work like that. Multiple pieces of software handle different aspects of the enterprise. This is due to a host of reasons.
-
People buy packages that are developed by outside organizations.
-
Different systems are built at different times, leading to different technology choices.
-
Different systems are built by different people whose experience and preferences lead them to different approaches to building applications.
-
Getting an application out and delivering value is more important than ensuring that integration is addressed, especially when that integration doesn't add any value to the application under development.
As a result, any organization has to worry about sharing information between very divergent applications. These can be written in different languages, based on different platforms, and have different assumptions about how the business operates.
Tying together such applications requires a thorough understanding of how to link together applications on both the business and technical levels. This is a lot easier if you minimize what you need to know about how each application works.
What is needed is a common data transfer mechanism that can be used by a variety of languages and platforms but that feels natural to each. It should require a minimal amount of specialized hardware and software, making use of what the enterprise already has available.
Files are a universal storage mechanism, built into any enterprise operating system and available from any enterprise language. The simplest approach would be to somehow integrate the applications using files.
Have each application produce files that contain the information the other applications must consume. Integrators take the responsibility of transforming files into different formats. Produce the files at regular intervals according to the nature of the business.
)
Have each application produce files that contain the information the other applications must consume. Integrators take the responsibility of transforming files into different formats. Produce the files at regular intervals according to the nature of the business.
An important decision with files is what format to use. Very rarely will the output of one application be exactly what's needed for another, so you'll have to do a fair bit of processing of files along the way. This means not only that all the applications that use a file have to read it, but that you also have to be able to use processing tools on it. As a result, standard file formats have grown up over time. Mainframe systems commonly use data feeds based on the file system formats of COBOL. UNIX systems use text-based files. The current method is to use XML. An industry of readers, writers, and transformation tools has built up around each of these formats.
Another issue with files is when to produce them and consume them. Since there's a certain amount of effort required to produce and process a file, you usually don't want to work with them too frequently. Typically, you have some regular business cycle that drives the decision: nightly, weekly, quarterly, and so on. Applications get used to when a new file is available and processes it at its time.
The great advantage of files is that integrators need no knowledge of the internals of an application. The application team itself usually provides the file. The file's contents and format are negotiated with integrators, although if a package is used, the choices are often limited. The integrators then deal with the transformations required for other applications, or they leave it up to the consuming applications to decide how they want to manipulate and read the file. As a result, the different applications are quite nicely decoupled from each other. Each application can make internal changes freely without affecting other applications, providing they still produce the same data in the files in the same format. The files effectively become the public interface of each application.
Part of what makes File Transfer simple is that no extra tools or integration packages are needed, but that also means that developers have to do a lot of the work themselves. The applications must agree on file-naming conventions and the directories in which they appear. The writer of a file must implement a strategy to keep the file names unique. The applications must agree on which one will delete old files, and the application with that responsibility will have to know when a file is old and no longer needed. The applications will need to implement a locking mechanism or follow a timing convention to ensure that one application is not trying to read the file while another is still writing it. If all of the applications do not have access to the same disk, then some application must take responsibility for transferring the file from one disk to another.
One of the most obvious issues with File Transfer is that updates tend to occur infrequently, and as a result systems can get out of synchronization. A customer management system can process a change of address and produce an extract file each night, but the billing system may send the bill to an old address on the same day. Sometimes lack of synchronization isn't a big deal. People often expect a certain lag in getting information around, even with computers. At other times the result of using stale information is a disaster. When deciding on when to produce files, you have to take the freshness needs of consumers into account.
In fact, the biggest problem with staleness is often on the software development staff themselves, who frequently must deal with data that isn't quite right. This can lead to inconsistencies that are difficult to resolve. If a customer changes his address on the same day with two different systems, but one of them makes an error and gets the wrong street name, you'll have two different addresses for a customer. You'll need some way to figure out how to resolve this. The longer the period between file transfers, the more likely and more painful this problem can become.
Of course, there's no reason that you can't produce files more frequently. Indeed, you can think of Messaging as File Transfer where you produce a file with every change in an application. The problem then is managing all the files that get produced, ensuring that they are all read and that none get lost. This goes beyond what file system–based approaches can do, particularly since there are expensive resource costs associated with processing a file, which can get prohibitive if you want to produce lots of files quickly. As a result, once you get to very fine-grained files, it's easier to think of them as Messaging.
To make data available more quickly and enforce an agreed-upon set of data formats, use a Shared Database. To integrate applications' functionality rather than their data, use Remote Procedure Invocation. To enable frequent exchanges of small amounts of data, perhaps used to invoke remote functionality, use Messaging.