Tips for a good ESB integration system

1. Build SOA Environment –

Share common logics around processes with exposed services.

Benefits:

Changes will be easy – If a common logic gets changed, then you would only need to deploy a service again and each process would work with the new logic.

Services logic would be shared around ALL applications – 

These days, almost every develop tool or language is able to consume a web-service. That means you can share logic easily around an organization. Use it in order to minimize duplicate code and maintenance.

2. Less-Code and Out Of The Box solutions –

More code = More complexity & maintenance. 

Use products for minimum coding, reliable solutions and support from professional vendors.
 

 

3. End-To-End Tracing –

A good BPM would always have tracing and information regarding messages being transferred in the wire.
 
Suppose message doing this route -
Application A --> Integration System --> Application B, Application C, Application D.
Make sure there's a field to identify the message in each step and each application writes to the same trace log.

Application A writes that message 12203 was created in a folder.
Integration System writes that message 12203 was taken and process started.
Integration System writes that message 12203 is now sent to Application B, Application C, Application D.
Application B writes that message 12203 was received.
etc.

Write important business and technical process data.

Many organizations use SQL Server for tracing but the innovative solution is to use ElasticSearch which includes free-text search on a huge mass of data.

Building reports for customers (based on those traces) is always a good thing because they can help with maintenance.

 

4. Cache layer for lookups –

Reading and writing to and from memory is faster than any I/O.

If many processes uses configurations or lookup values stored in SQL Server then consider an appropriate cache layer which uses memory to store data.

Redis & Memcached are classic for caching, but you can also use NoSQL solutions which stored data in memory first before being persisted (e.g couchbase).

5. XSLTs for Mapping -

DON'T write C#, Java or whatever language code in order to transform XML's.

Develop XML maps ONLY with XSLTs which can be reused in any integration tool which deals with XML transformation.

 

6. Use Stored Procedures for queries -

It's better for your application to execute stored procedure for queries, rather than sending those as text to the DB.

It's easier for a DBA to control & manage organization logic, and to avoid "BAD" queries.

 

7. Dynamic configuration –

Think deeply on every process configurations you should be able to modify on runtime without changing any code.

Don't exaggerate with dynamic configurations because it has a maintenance price. You do not want an interface with a huge number of configurations, if most of them are values which will be remain unchanged. 

 

8. Well-defined schemas –

It's important that schema nodes, elements and attributes would have real limitations (Min/Max occurs, Value types etc.).

It's possible to restrict schema field value types in many ways (int, string, regular expressions, enums etc.).

Benefits:

Data integrity – Well, that's the obvious reason. Data is more reliable on each point of entire process.

Less complexity in process – Suppose a process need to get an e-mail address from a request message.

If schema would enforce a valid e-mail regex in that field, then there is no need
to perform this validation on process.

 

9. ASync web-services is better than Sync web-services –

If there is a possibility of consuming asynchronous web-service – ALWAYS prefer it over a synchronous web-service.

Synchronic method (Two-Way) would open a session, send a request, wait for a response and only then closes the session.

Asynchrony method (One-Way) would open a session, send a request and closes the session immediately without waiting for a response.

More sessions = More CPU and more memory usage on server.

Take this tip in mind especially with offline interfaces. In those kind of processes the execution doesn't has be immediate so entire interface can last longer. 

 

10. Avoid direct-connection between Production & Test environments –

Some organizations tend to send messages from production to test\development environments.

That's really bad because if a resource isn't available in test\development then production would be affected.

 

 

11. Identical backup environment (or clustered in an active-passive mode) –

Maintain a backup (not-activated) environment for production (an exact copy of it).

Make sure backup environment is always up to date with the latest changes of the WORKING production environment.

 

12. Retry mechanism when using resources –

Consider using retry mechanism on a process which uses resources (SQL, Web-Services etc.).

Take notice you can implement retry only on OFFLINE processes/ASync services (i.e. when no one waits with an open session for an answer from that service).

 

13. Centralized production environments -

With the major communication progress during the last few years, geographic distance is not an issue as it used to be.

Less environments = Less maintenance.

A good centralized environment contains distributed servers which share data processing.
Some integration tools like "BizTalk" includes load balancing mechanism out of the box, but if your tool doesn't - a load balancing software would help.

 

14. Health and performance monitor system –

Monitor & alert crucial resources which is used by processes to early identify issues. "SCOM" and "Nagios" are popular products. Use "Watcher" for elasticsearch logs.

For instance: 

If a server CPU stays on 100% for the last 30 seconds – send an e-mail. 

If files are piling up in a directory which should be empty – send an SMS.

 

14. Archive source messages\requests –

Gives power for an integration system to rerun messages in case process went wrong and to be self-dependent from end-to-end.

 

15. Persistence –

Some integration tools (like BizTalk, which is DB-based) have persistence functionality built-in as part of their engine.
Products that put an emphasis on performance doesn't have it (e.g TIBCO) but it's possible to implement persistence logic inside a process.
Analyze each process before implementing and think when persistence is necessary if everything breaks down.

 

16. Parallel processing –

Use parallel processing if process logic allows it but beware from over-complex.
Keep in-mind that code should be thread-safe.