It’s been a while since my last blog, but IMHO this is one of important blog that I’ve ever made that can be useful pointer for every aspiring developer / companies that will/already implement application integration. Below contains a list / summary of best practices I’ve acquired in my more or less 10 years working with BizTalk, SSIS, WCF and .NET integration (primarily Microsoft Technology) but this I think is also applicable to other technology.
Design / Development:
#1 K.I.S.S (Keep it simple, stupid). Don’t get me wrong, I like complex integrations, from multiple message correlations, requirement to aggregate certain line items with certain condition in the message to having a direct binding with delivery notification set to transmitted (multiple send port / send port groups, this will result in some zombie messages). Anyone from novice to expert BizTalk Developer can build a solution, it can be done & solve in different ways but the key here is providing a simple and effective solution for a complex integration.
#2 Isolation with maximum reusability. Isolation for me means that the application can be deployed without having too much dependency to other BizTalk artifacts that are shared. During the design it’s always a struggle on how to seperate artifacts, on which solution / projects it should be placed to avoid having to uninstall all referencing application before you can refresh the dependency (GACing works in some situation but not for all, you’ll be surprise how BizTalk deal with it internally). Once identifying or isolating a certain integration flow whenever there’s a new requirement, we should always check whether it fits on already existing solution (extend) without having to build a new integration flow. in this case we will have maximum reuse of existing solution.
#3 Consistency. Every solution that will be build should look as if it’s build / coded by one person. Having a naming convention on BizTalk artifacts, namespace, solutions, project & folder structures should be define first. Based on experience providing a proper name is always a challenge since if you define it incorrectly, it will be a costly mistake later on.
#4 Nuget – all common code / components / libraries should be referenced from a local nuget repository. Having a local nuget repository not only simplifies referencing dependencies but also saves you once you start using TFS Build Server (autorestore feature). This means that shared assemblies is not required to be checked in together with the source codes.
#5 .NET over XLST in complex mapping – Whenever there’s a requirement for complex mapping like grouping, aggregation, unique numbering over a certain combinations or constructing a hashtable in the mapping, I always go for .NET component. Using .NET component it is easier to maintain and tested (unit testing + mockup) vs doing everything in XSLT (in which can be forgotten in time).
Build / Deployment:
#1 Release Pipeline – It’s important to define strategy on how you would build and be able to do ‘Build once’ and deploy to multiple environments. In our case, part of our strategy is to minimize branching and always build from MAIN. In TFS, we’ve implemented a Gated Checkin which means that every checkin will trigger a release pipeline. During the checkin, a build will be triggerred (MSBuild and BTDF build), it will run all unit test and copies the binaries to drop folder (only If all steps are successful). From this point the Ops can just enter a command (application name) and environment and then the automated deployment will take over..
#2 Configuration over Code – for normal deployment most of time you would see that a solution contains a powershell script for deployment, this is not bad but i think this only works on sample codes or tutorials but a problem in Enterprise deployment. For this you need a good deployment framework (compilation of scripts) to do the deployment and during the deployment preparation just replace the values in the templates like servernames, shared folders and etc. I’ve posted some blogs earlier how to do this:
Centralize Powershell Repository: