Today I received an email asking to some extent best practices on development with SQL Server Integration Studio (SSIS) and Business Intelligence Developer Studio (BIDS) in a team environment. Here is part of the email:
Me and another DBA belong to the same team, we have a SQL server with SSIS running. We use the SSIS transfer data among multiple data sources. In SQL 2000 DTS, both of us can save the package on the server and open/edit it in the enterprise manager. In SQL 2005, I can see the package on server, but can’t open it directly. We came out a solution: create a shared folder on the server called ‘SSIS Projects’, both of us can access to it. We run the ‘SQL Server Business Intelligence Development Studio’ on local PC, to open the project in that shared folder. When done with the change, save the package to the SSIS server. Now, we have more than 50 packages in a project. Problem is: it’s very slow when open a project, ‘Business Intelligence Development Studio’ tends to open/verify every single package inside a project, takes up to 10 mins and getting worse. We really miss the SQL 2000 DTS, but we have to turn to SQL 2005.
Are we doing the right thing? Is there any better solution for SSIS developing in a team environment?
When open a project, does ‘Business Intelligence Development Studio’ has to open/verify every package?
This got me thinking, and I figured instead of write an email back, it would be good info for a blog post. So here is what I think and some things I have done that have worked.
First, yes, SQL 2000 DTS allows you to just edit on the server, do more than SSIS, is just way better than SSIS. Wait, what? Well, yeah some people will say that, because it does one thing that might be a little rigmarole in SSIS, but no, SQL 2000 DTS is not better than SSIS, just wanted to clear that up.
So, the is meant to be a starting point, by no means all encompassing, and as always, YMMV.
One thing that I first thought about is this: Yeah, if BI devs and SQL devs have never really worked in a team environment, developing software, how would they know what to do, or best practices? They would just go about “making it work” until everything breaks or who know what.
So how to develop Microsoft Business Intelligence Solutions in a team environment?
1) Standardize on Versions
First, figure out what “versions” you are going to support, and what you are going to use, and get standardized on them. I am guessing majority of BI devs right now are on the 2005 stack. Yeah, there is still probably a bit of 2000 legacy stuff out there, and some people are now getting into the 2008 stuff, but 2005 is pretty much the norm from what I see, at least at this point.
So, 2005. Get all your dev’s on 2005 on their machine – same patch level, etc. Get BIDS up to the same level. Get BIDS helper installed everywhere. Strive to get all your ETL packages in SSIS 2005, get all your cubes to SSAS 2005, etc, etc. Come to a consensus on things like config files for SSIS, naming conventions, within your development and on disk – folder structure is key! With a smaller number of versions of things floating around, it makes it easy for anyone on the team to open up a solution and start hammering away without tons of setup.
2) Get Source Control
This is crucial! I have talked about source control in the past, and also about some that aren’t so great. Really it doesn’t matter what you use, I prefer SVN. I install Tortoise SVN, SVN proper (to do scripting etc if I need to using cmd line) and also purchase Visual SVN, an add on to Visual Studio that integrates with SVN. for 50 bucks you have your source control system. Visual Source Safe works but is outdated, honestly I hate it. Team Foundation Server is good, but expensive. Other solutions might be using something like GIT, etc. Whatever you do, just get a source control system going, and learn it well. Learn how to create repos, commit, update, revert, merge, etc. Set up a user for each BI dev and make sure they commit often, and make sure they leave comments in the source control log when they commit, history is your lifeline to go back to something if you need to! Note: exclude .suo, bin, obj directories, .user files, etc. Anything that changes every time you build, open, etc, you want to exclude from source control.
3) Development Box
You now have your version standardized, and your source control setup. You can get most of your work done on your machine, but you need somewhere to test deployments, run scenarios, etc, etc. Make sure you have a comparable box to your production server. Set it up the same, same software etc. Make sure its backed up. Let all the devs know its a dev box, it can be wiped at any time for any reason if need be. It can be rebooted 5 times a day if need be. Its a dev box! But you can test and develop and tweak and change settings to your hearts content and not have to worry about breaking Mr. Executives reports.
4) Developing, Merging, Committing, Collaborating, Communicating.
So now you have your setup, well.. setup. Start creating stuff. SSIS Packages, ETL’s, SSAS Cubes, SSRS Reports, the whole MSFT BI Solution. This is where stuff can start to get tricky in a team environment though. SSIS/SSAS/SSRS isn’t as clean cut as something like C#/VB.NET, etc. Everything is in some form of XML behind the scenes, and with graphical based editing, you can move stuff around and it changes the files. Things like that are going to be your enemy. This is why you need to collaborate and communicate. Usually one person should be working on one project at a time. You can get really good at communicating and then in SSIS at least have multiple people working on different packages. Also in SSAS dimension editing and stuff can be done by multiple people at the same time as long as the dim is already hooked up to the cube. But you want to make sure that you communicate, “Hey, I am checking this in, you might want to do an update”, or “Is anyone working on this or are they going to? I want to modify something, and I will check it in so you all can see it”
You want to make sure you have your folder structure, and solution/project structure set up well. C:Projects .. and then maybe a folder for each major project “CompanySales” and under that, “ETL”, “Cube”, “Reports” and have a solution under each with 1 project of each type. You can also have a generic SSRS solution with many projects, which might work well for you. In any case, just come up with a standard and stick to it. Trust me it will make your life easier. The question from the email above, it sounds like they have every package in one solution, one project. Sounds like it needs to be split, multiple solutions, multiple projects.
5) Deployment Scenarios and Strategies
Now that you have everything developed, tested, checked in, what do you do?
Personally for SSIS I like xcopy deployments. One folder on the server, not on the C drive, but another drive, lets say “E:SSIS” under that a folder for each project. Put your dtsx and configs in the same folder. 99% of the time you are going to call the dtsx from a SQL Agent Job, and most likely you are going to run into a scenario where you need uber rights to execute it, so learn how to create a proxy/credential in SQL security so you can run the step as that. Once you have this folder and subfolders setup, you can use something like Beyond Compare to compare the folder on the server to the one you have locally that matches. Remember to copy files from the bin directory of your project after you build it, not the files directly on your project. As far as BIDS validating every package, there are workarounds out there you can do, here is one.
For SSAS, I try to lean towards using the Deployment Wizard that comes with SQL Server. You can use BIDS deployment, but if you start doing anything advanced with roles, partitions, etc, you are going to run into trouble. Take control and use the deployment wizard. I usually like to deploy, and then process manually when developing. And then later use SQL Agent or and SSIS package to actually do processing when it comes to a scheduled processing.
SSRS, I have become used to the auto deployment from Visual Studio. To really do this though, you need a project for every folder in SSRS, which can become a pain. You can always just upload the .RDL file and connection and do it manually, but if you start off right with using the deployment from BIDS, it can make your life easier.
So that is just a 10 minute overview of everything to kind of get started. Everything depends on your infrastructure and the way your team is setup, etc. But I think the biggest thing to take from all of the above is to standardize on things. If you standardize on as much as possible, SQL versions, setup of machines, naming conventions, layouts, design patterns, etc, everyone can do things faster and pretty soon it will start running like a well oiled machine!