To begin tinkering with SSIS in containers, you first need to install Docker. There are some prerequisites. I will not exhaust the prerequisites here. I strongly suggest you familiarize yourself with the requirements for Docker prior to attempting to install the software.
Since I use a PC (a Lenovo P51S) running Windows 10 Pro, I chose to use Docker for Windows.
I Needed a HyperVisor
I chose to run Docker for Windows with Hyper-V:
One reason I upgraded to Windows 10 was to work with containers. I read posts and articles that stated I could use Docker for Windows with VirtualBox, and I had been using VirtualBox for a long time. When I started using VirtualBox, it was the only HyperVisor that was:
Supported 64-bit guests.
I decided to switch to Hyper-V, though, and rebuilt my demo and development virtual machines in Hyper-V.
IWOMM; YMMV (It works on my machine; your mileage may vary…) 😉
Choose a Container OS Platform
Once Docker is installed you need to decide whether to work with Linux or Windows containers, but you can switch anytime:
One nice thing about switching is it’s fairly painless, as one may glean from the message that displays when I click “Switch to Linux containers…”:
Docker for Windows with Windows Containers
The cool kids are using Linux containers, especially the cool SQL Server kids. I’ve tinkered a little with SQL Server on Linux. I’m interested in SSIS, though, so I’ve been focusing on running Windows containers in Docker for Windows.
Containers are like lightweight virtual machines. They typically consume less disk space. Containers spin up relatively quickly. A developer can configure a container, persist it in some state, and then turn it off. It’s faster and lighter than working with a VM. There are other benefits that we will get to later in this series.
After installing Docker, your next step is pulling an image.
Pulling an Image
I can hear you thinking, “Umm, Andy… what’s an image?” I’m glad you asked. From the Docker Glossary:
An Image is an ordered collection of root filesystem changes and the corresponding execution parameters for use within a container runtime. An image typically contains a union of layered filesystems stacked on top of each other. An image does not have state and it never changes.
I can now hear you thinking, “Umm… what?” I think of an image as a pre-configured container. What’s configured in the pre-configuration? Well, the operating system and other software I may want to use.
For example, the hello-world image is relatively small and will test your installation of docker. In the animated gif below, I show how I:
Search for images named “hello-world”
Locate the name of an image labeled “Official” and named “hello-world”
Pull the “hello-world” image (sped up – my internet is not that fast out here in Farmville…)
Re-pull the “hello-world” image to show how things look when the image is up to date.
Run the image
As stated earlier, the hello-world image is a test.
Searching, Pulling, and Starting a Windows Container with SQL Server Installed
To search for, pull, and start a Windows container with SQL Server installed and configured, execute the following Docker commands:
I love the SSIS Catalog. It’s an elegant piece of data integration engineering and I cannot say enough positive things about it. Packaging SSIS Catalog deployments can be tricky, though.
The SSIS Catalog is a framework. Frameworks manage execution, configuration, and logging; and the SSIS Catalog handles each task with grace. Like I said, I love it!
(You knew there was a “but…” coming, didn’t you?)
A Tale of Two Audiences
There are two audiences for the SSIS Catalog, two groups of consumers:
I listed three. Because…
SSIS is often administered by database administrators, or DBAs. I admire DBAs. It’s often a thankless job – more like a collection of jobs all rolled into one (and subsequently paid as if its one job…).
I believe the SSIS Catalog interface presented to DBAs in SQL Server Management Studio is sufficient.
My complaint is the SSIS administrator has to expand a handful of nodes in Object Explorer and then right-click to open the SSIS project configuration window and then double-click each referenced SSIS Catalog environment to determine which value is configured for use when an SSIS package is executed.
Click the screenshot above to see what I mean. Configuring SSIS Catalog deployments in SSMS is challenging. I find it… clunky. Once I understood all the windows, what they meant and how to configure an SSIS package and project deployed to the SSIS Catalog, this made sense. But – in my opinion – this interface works against comprehension.
Does this interface work, though? It certainly does. When I teach people how to use the SSIS Catalog, I show them how to use the Object Explorer interface provided in SSMS.
I don’t stop there, however, because I built one solution to the problem. I call my solution SSIS Catalog Browser. If you click to enlarge this image you will note I am viewing the configuration of the same parameter displayed in the SSMS image above. I find this interface cleaner.
Do administrators still need to understand how to configure SSIS Catalog deployments and SSIS packages and projects deployed to the SSIS Catalog? You bet. There is no substitute for understanding. SSIS Catalog Browser surfaces the same metadata displayed in the SSMS Object Explorer. The only difference is Catalog Browser is easier to navigate – in my opinion.
SSIS developers and release managers (DevOps release teams) need more functionality. As I wrote in DILM Tiers for the SSIS Enterprise, an enterprise should have a minimum of four Data Integration Lifecycle Management (DILM) tiers to manage enterprise data integration with SSIS. Those tiers need to be:
Development – an environment where SSIS developers build SSIS packages and projects. SSIS developers need permission / rights / roles to utterly destroy the database instances in Dev. If the SSIS developers lack this ability, you have “an environment named Development but not a Development environment.” There is a difference.
Test or Integration – an environment where SSIS developers have permission / rights / roles to deploy, configure, execute, and view logs related to SSIS packages and projects.
UAT or QA (User Acceptance Testing or Quality Assurance or any environment other than Production, Test, or Development) – an environment that mimics Production in security, permission / rights / roles. Developers may (or may not) have read access to logs, source, and destination data. SSIS administrators (DBAs or DevOps / Release teams) own this environment. The person performing the deployment to Production should perform the deployment to UAT / QA / Whatever because I do not want the Production deployment to be the very first time this person deploys and configures this SSIS project.
Viewing the contents of an SSIS Catalog is not enough functionality to manage releases. Why, then, do I include developers? Because…
SSIS developers create the initial SSIS Catalog deployments in the DILM DevOps cycle.
I cannot overemphasize this point. Developers need an environment where they are free to fail to build SSIS. They aren’t free to succeed, in fact, unless and until they are free to fail.
Have you ever heard a developer state, “It works on my machine.”? Do you know why it works on their machine? Defaults. They coded it up using default values. The defaults have to work or the developer will not pass the code along to the next tier.
How does an SSIS developer know they’ve forgotten to parameterize values? How do they figure this out?
It’s impossible to test for missing parameters in the Development environment.
The answer is: SSIS developers must deploy the SSIS project to another environment – an environment separate and distinct from the Development environment – to test for missing parameterization.
To review: SSIS developers need a Development environment (not merely an environment named Dev) and they need a different environment to which they can deploy, configure, execute, and monitor logs.
Having the SSIS developers build and script the SSIS Catalog deployments eliminates 80% of deployment configuration errors (according to Pareto…).
Having SSIS administrators practice deployment to UAT / QA / Whatever eliminates 80% of the remaining errors.
Math tells me an enterprise practicing DILM in this manner will experience a 4% deployment error rate. (Want to knock that down to 0.8%? Add another tier.)
Packaging SSIS Deployment
I will not go into functionality missing from the SSMS Object Explorer Integration Services Catalogs node (nor the underlying .Net Framework assemblies). I will simply state that some functionality that I believe should be there is not there.
I don’t stop there, however, because I built one solution to the problem. I call my solution SSIS Catalog Compare. If you click the image to enlarge it, you will see a treeview that surfaces SSIS Catalog artifacts in the same way as SSIS Catalog Browser (they share a codebase). You will also see the results of a comparison operation, and the user initiating the packaging of an SSIS folder deployment by scripting the folder and its contents.
The result is a file system folder that contains T-SQL script files and ISPAC files for each SSIS Catalog configuration artifact:
You can use SSIS Catalog Compare to lift and shift SSIS Catalog objects from any environment to any other environment – or from any DILM tier to any other DILM tier – provided you have proper access to said environments and tiers.
This includes lifting and shifting SSIS to the Azure Data Factory SSIS Integration Runtime, also know as Azure-SSIS.
Zip up the contents of this file system folder, attach it to a ticket, and let your enterprise DevOps process work for data integration.
The SSIS Catalog is a framework, and a pretty elegant framework at that. Some pieces are clunky and other pieces are missing.
A great big THANK YOU to everyone who attended my presentations!
As I’ve shared (numerous times) in the past, I share my ratings not to boast but to let less experienced and would-be presenters peek behind the curtain.
The ratings were on a scale of 1 (bad) to 5 (good).
Intelligent Data Integration Precon
I had 107 attendees and received feedback from 33 attendees.
Rate the value of the session content: 4.45 How useful and relevant is the session content to your job/career? 4.24 How well did the session’s track, audience, title, abstract, and level align with what was presented? 4.42 Rate the speaker’s knowledge of the subject matter: 4.79 Rate the overall presentation and delivery of the session content: 4.45 Rate the balance of educational content versus that of sales, marketing, and promotional subject matter: 4.55
These are ok numbers. Not as good as Brent’s or Cathrine’s, but ok. I’m ok with these numbers but I want to do a better job connecting with the audience.
Selected Comments About the Presentation
Most of the comments were positive. I really appreciate the positive comments. I included all the non-positive comments below. Some of the negative comments were also instructional. Others… need instruction.
“I learned tips and tricks that justified the cost of the conference.” “Andy was an excellent speaker. Working through demos was integral in giving me exactly what I need to implement a solution.” “Great knowledge and very humble in the way he talks.” “Andy is awesome, his ego does not get in the way during the presentation, overall great.” “Andy is a wonderful speaker and I was able to easily understand the topics covered and feel confident that I can use the lessons learned in my profession.” “Excellent delivery of material. Always delivers humor along with the content. Picked up tips that would have taken a very long time to learn on my own. Would love to see another pre-conference session on ETL by Andy (whether SSIS or ADF).”
“We spent way too much time in Biml. ” “Andy knows his subject but while the Biml intro was interesting the examples of SSIS patterns were too simplistic, too ‘Hello world!’. Should have had at least one example were data source and destination were on different servers.” “Continual reference speaker made to how session content just ‘hobbying’/not production conscious, alluding to the ‘real’ stuff being something other than what was being presented. ‘I have a stand in the exhibitors arena’!!! The serious stuff!!! Cheapened the experience. Although pre-con by no means cheap! This inflection, re-iterated by the speaker was disappointing. Felt a superiority complex existed! Unsettling.”
I didn’t get much feedback on the Biml portion of the precon. Of the two comments that mention Biml, one was negative and another positive. I’m not sure if that’s because I covered Biml during the first part of the precon (and evaluations were filled out at the end of the precon) or if I did a worse job on Biml than the other topics. To me, it looks like a wash.
The last comment is… I don’t know. I think someone was having a bad day or I wasn’t their cup of tea or I just rubbed them the wrong way. I actually took 60 seconds to explain that PASS exhorts presenters to steer clear of sales-y talk in presentations. In the past, I’ve been vocal in my disagreement with PASS about these policies. While I believe I am right, I signed a contract to abide by the rules.
That said, the rules permit sponsors and exhibitors to share information about their for-pay products during presentations. In the 60 seconds, I shared my belief that someone – either the attendee or their company – had paid good money for the attendee to attend this precon. But I did not want to spend any of our time together selling them anything. I shared some of the free tools available from DILM Suite. I mentioned I sell other tools and solutions and that I was an exhibitor. If they wanted to learn more about products I sell, they could drop by the booth or email me.
A bunch of attendees did just that! Nick and I had some great conversations with people at our booth. I’m pretty sure we will be helping some folks with training and consulting, and that we sold some licenses to DILM Suite products.
I strive to cover the topics in as much depth as possible. In order to do that, I demonstrate “chunks” of functionality and patterns using demos that are not Production-ready. I believe more than one day is required to delve into data integration with SSIS with Production-ready code. I’ve led such training – and continue to deliver beginner and advanced SSIS training privately and occasionally, publicly. If you’re interested in learning SSIS from the ground up, I have a five-day course named From Zero To SSIS. I’d be honored to deliver this training for you or your team. Email me. If you are an experienced SSIS developer, I recommend attending a delivery of Expert SSIS presented in cooperation with Brent Ozar Unlimited.
To this attendee who I do not know (evaluations are anonymous when sent to speakers) and any other attendees who agreed with this attendee, I apologize and ask your forgiveness for coming across as superior and cheapening your experience.
I had 203 attendees and received feedback from 63 attendees.
Rate the value of the session content: 4.54 How useful and relevant is the session content to your job/career? 4.44 How well did the session’s track, audience, title, abstract, and level align with what was presented? 4.60 Rate the speaker’s knowledge of the subject matter: 4.83 Rate the overall presentation and delivery of the session content: 4.63 Rate the balance of educational content versus that of sales, marketing, and promotional subject matter: 4.83
These are better numbers. They trend about the same, though: I scored higher on speaker’s knowledge of the subject matter in both this session and the precon. My lowest score in both presentations is the usefulness of the session content for job and career. From this I glean I need to work on content.
Selected Comments About the Presentation
As before, most of the comments were positive and I included all the less-than-positive comments below:
“I now know that a hashbyte match for an etl will work. Ive wanted to try this but my clients use other methods to load data. Thank you very much.” “Andy is a good man and we are lucky to have him in our SQL Family.” “Very easy to sit through! Great content and answers.” “Andy is always entertaining. I learned quite a bit from this session and I plan to implement the things he spoke about.” “Great session, love your humor. Good and easy to follow demos. Good job!” “Andy was incredibly entertaining and the content was very good.” “Great session. He’s on my list when I come back.” “More sessions should have interpretive dances.”
“Thanks for the talk. Liked the time for questions. However I would have preferred more examples.” “Excellent presentation skills. Very interesting and entertaining. There were way too many questions about their own specific issues that would have been better if handled after the session. It felt as if we rushed through content because there were so many questions. I’m definitely going to look into Andy’s available content because I found his approach interesting and different and would like to see more tricks.” “Ok. Hard to understand some concepts.” “Good instructor, but deep subject matter (hard to follow).” “I should have read session description more carefully. It was beyond my skills. My own fault.”
I’m going to start at the bottom of the negative comments here and say: “God bless you sir or ma’am,” to the person who had the courage to share that the level of the session was beyond their skills. I had this session pegged as Level 300. I will increase that level going forward to 400 (minimum) based on feedback about the level. It wasn’t just this person, I own the “hard to follow” and “hard to understand” comments. I will correct that moving forward. I will also add a disclaimer about the pace of this presentation in response to the “rushed through content” comment – which is another reason to bump the level.
Regarding the interpretive dance comment, well… you had to be there. 😉
BI & Data Visualization Panel
This was a panel that included Mico Yuk, Melissa Coates, Meagan Longoria, Ryan Wade, and me. We had 111 attendees and received feedback from 20 attendees.
Rate the value of the session content: 3.75 How useful and relevant is the session content to your job/career? 3.90 How well did the session’s track, audience, title, abstract, and level align with what was presented? 3.90 Rate the speaker’s knowledge of the subject matter: 4.40 Rate the overall presentation and delivery of the session content: 3.90 Rate the balance of educational content versus that of sales, marketing, and promotional subject matter: 4.50
These numbers are not good but the comments help us understand why.
Selected Comments About the Presentation
The comments were evenly split between positive and less-than-positive:
“Fantastic and especially loved the diversity of the panel. Thank you!” “I valued the information about Data maturity – getting customers from data > information > knowledge.” “This was great. There was a lot of interesting discussion. Mico was a great host.” “Mico was a fantastic moderator and the panel (for the most part) were experts in their field.”
“Mico was great at moving the conversations along and keeping it on point…very well done. I guess I was just disappointed that I didn’t come away with more “definitive” material. I’ll admit, I didn’t know what to expect.” “Mico interrupted the panelists with her own comments too much and didn’t manage the room well. The panelists were fantastic.” “This session need a Plan B for when the audience did not contribute many (or any) questions. Not a worthwhile session for me.” “The panel could have been a little better balanced. They all(?) were consultants, it would have been nice to have someone on the panel who is just working for a company.”
It’s tough to tell towards whom the comments were aimed (except the comments that mentioned Mico which were evenly split positive and less-than-positive).
I’ll share that it’s tough to get high ratings from a panel presentation. Part of the reason is it’s technically impossible to prepare in advance for the questions from the audience. I believe that’s also an advantage for some attendees.
Of the 20 people who supplied evaluations, 4 ranked the panel session below 3.0. You don’t have to be good at statistics to realize those scores are going to tank the averages.
If you’re going to present, you should understand the dynamics of evaluations.
I shared with a friend – who happens to be a very good speaker and received some disappointing comments on his evaluations – with experience, we know when we’ve delivered as good a presentation as we can and when we’ve fallen short of that goal. For me:
I know I left it all on the field in my presentations at PASS Summit 2018. I did my very best. I practiced and the practice paid off. I planned and the planning helped – especially with timing demos in both the precon and Faster SSIS.
Because I know I delivered these presentations well, I’m happy with these numbers. Will I strive to do better? Will I incorporate this feedback into future deliveries of the precon and Faster SSIS? Goodness yes! Overall, though, I’m happy to have had the honor of delivering a precon solo at the PASS Summit and that I was selected to deliver a session and participate on a panel.
Frank (@Tableteer) and I just recorded a show in which I mention both this post and the game called “There’s Always One.” Check it out, along with other podcast recordings at DataDriven.tv!
I occasionally (rarely) read reviews at Amazon of books I’ve written. If I learn of a complaint regarding the book I often try to help. If a reader experiences difficulty with demos, I often offer to help, and sometimes I meet with readers to work through some difficulty related to the book (as I offered here). About half the time, there’s a problem with the way the book explains an example or the sample code; the other half the time the reader does not understand what is written.
I own both cases. As a writer it’s my job to provide good examples that are as easy to understand as possible. Sometimes I fail.
In very rare instances, I feel the review misrepresents the contents of a book – enough to justify clarification. This review afforded one such opportunity. None of what I share below is new to regular readers of my blog. I chose to respond in a direct manner because I know and respect the author of the review, and believe he will receive my feedback in the manner in which it was intended – a manner (helpful feedback) very similar to the manner in which I believe his review was intended (also helpful feedback).
Thank you for sharing your thoughts in this review.
I maintain successful technology solutions are a combination of three factors:
The problem we are trying to solve;
The technology used to solve the problem; and
The developer attempting to use a technology to solve the problem.
I believe any technologist, given enough time and what my mother refers to as “gumption” (a bias for action combined with a will – and the stubbornness, er… tenacity – to succeed), can solve any problem with any technology.
I find your statement, “It doesn’t really teach BIML but rather teaches a specific optional framework for those who want to do everything with BIML rather than increase productivity for repetitive patterns” inaccurate. Did you read the entire book? I ask that question because I most often encounter readers who state the precise opposite of the opinion expressed in this review (some of your fellow reviewers express the opposite opinion here, even). I disagree with your statement, but I understand your opinion in light of other opinions expressed to me by readers during my past decade+ of writing several books. I share one example in this blog post. The short version: we often get more – or different – things from reading than we realize.
There is a section in The Biml Book – made up of 3 of the 20 chapters and appendices – that proposes a basic metadata-driven Biml framework which is the basis of a metadata-driven Biml framework in production in several large enterprises. I wrote those 3 chapters and the basic metadata-driven Biml framework in question. Varigence has also produced a metadata-driven Biml framework called BimlFlex – based upon similar principles – which has been deployed at myriad enterprises. The enterprise architects at these organizations have been able to improve code quality and increase productivity, all while decreasing the amount of time required to bring data-related solutions to market. They do not share your opinion, although they share at least some of the problems (you mentioned ETL) you are trying to solve.
Am I saying Biml – or the framework about which I wrote or even BimlFlex – is the end-all-be-all for every SSIS development effort? Goodness no! In fact, you will find disclaimers included in my writings on Biml and SSIS frameworks. I know because I wrote the disclaimers.
Misalignment on any of the three factors for successful technology solutions – the problem, the technology, and/or the developer – can lead to impedance mismatches in application and implementation. For example, Biml is not a good solution for some classes of ETL or data integration (points 1 and 2). And learning Biml takes about 40 hours of tenacious commitment (which goes to point 3). This is all coupled with a simple fact: data integration is hard. SSIS does a good job as a generic, provider-driven solution – but most find SSIS challenging to learn and non-intuitive (I did when I first started using it). Does that mean SSIS is not worth the effort to learn? Goodness, no! It does mean complaints about simplifying the learning process are to be expected and somewhat rhetorical.
Architects disagree. We have varying amounts of experience. We have different kinds of experience. Some architects have worked only as lone-wolf consultants or as members of single-person teams. Others have worked only as leaders of small teams of ETL developers. Rarer still are enterprise architects who are cross-disciplined and experienced as both lone-wolf consultants and managers of large teams in independent consulting firms and large enterprises. Less-experienced architects sometimes believe they have solved “all the things” when they have merely solved “one of the things.” Does this make them bad people? Goodness, no. It makes them less-experienced, though, and helps identify them as such. Did the “one thing” need solving? Goodness, yes. But enterprise architecture is part science and part art. Understanding the value of a solution one does not prefer – or the value of a solution that does not apply to the problem one is trying to solve – lands squarely in the art portion of the gig.
Regarding the science portion of the gig: engineers are qualified to qualitatively pronounce any solution is “over-engineered.” Non-engineers are not qualified to make any such determination, in my opinion.
By way of example: I recently returned from the PASS Summit. Although I did not attend the session, I know an SSIS architect delivered a session in which he made dismissing statements regarding any and all metadata-driven frameworks related to ETL with SSIS. If I didn’t attend his session, how do I know about the content of the presentation? A number of people in attendance approached me after the session to share their opinion that the architect, though he shared some useful information, does not appreciate the problems faced by most enterprise architects – especially enterprise architects who lead teams of ETL developers.
My advice to all enterprise architects and would-be enterprise architects is: Don’t be that guy.
Instead, be the enterprise architect who continues to examine a solution until underlying constraints and drivers and reasons the solution was designed the way it was designed are understood. Realize and recognize the difference between “That’s wrong,” “I disagree,” and “I prefer a different solution.” One size does not fit all.
Finally, some of the demos and Biml platforms were not working at the time you wrote this review. Many have been addressed since that time. In the future, updates to SSIS, Biml, and ETL best practices will invalidate what this team of authors wrote during the first half of 2017. As such, the statements, “And much of the online information is outdated and no longer works. It’s a shame someone hasn’t found a way to simplify the learning process.” is a tautology that defines progress. Learning is part of the job of a technologist. I encourage us all – myself most of all – to continue learning.
I’ve been making smaller, more incremental changes to SSIS Catalog Browser – a free utility from the Data Integration Lifecycle Management suite (DILM Suite).
You can use SSIS Catalog Browser to view SSIS Catalog contents on a unified surface. Catalog Browser works with SSIS Catalogs on-premises and Azure Data Factory SSIS Integration Runtime, or Azure SSIS. It’s pretty cool and the price ($0 USD) is right!
The latest change is a version check that offers to send you to the page to download an update. You will find this change starting with version 0.7.7.0. Version 0.7.8.0 includes a slightly better-formatted version-check message. As I said, smaller, more incremental changes.
I’ve made statements about “two kinds of developers” for years. These statements are false inasmuch as all generalizations are false. The statements are not designed to be truisms. They are designed to make people think.
Last week – while presenting a full-day pre-conference session at the PASS Summit 2018 and again when delivering a session about Faster SSIS – I repeated the sentiment shown at the top of this post:
There are two kinds of developers: 1) Those who use source control; and 2) Those who will.
I follow up with: “Because if you do not use source control, you will lose code one day and it will break your heart.” Audience members laugh and the point is made.
More Than Two
There are myriad types of developers. And that’s a good thing. Why? Because there are myriad problems developers face in the wild and those problems need to be solved.
There are no one-size-fits-all solutions. If you attended the PASS Summit last week you likely saw some really cool demos. I know I did. And you may have thought – or even been told – that this, this right here is the answer for which you’ve searched your entire career.
There’s a possibility that the people selling you on whatever-this-is are absolutely correct. There’s a greater possibility that they are less than absolutely correct.
I write this not to disparage anyone’s solution (or my own solutions). Promise. I write this to dissuade the disparaging of anyone else’s solution (because that also happens).
The Bottom Line
Goldilocks. The bottom line is we all want the Goldilocks solution. We want the just-right solution that maximizes efficiency and minimizes complexity.
That’s just hard.
I can hear you thinking, “Why is maximizing efficiency and minimizing complexity hard, Andy?” I’m glad you asked. Solutions are a moving target, part art and part science, and the only way to learn where and when to draw the art-science line is experience.
A Moving Target
Maximizing efficiency and minimizing complexity is hard because it’s not at all the bottom line; it’s a line in the middle – a balancing of mutually-exclusive demands on your time, expertise, and energy.
Plus, it shifts.
Everything scales. Things scale either up and / or out or they scale down and / or in. In general (generality warning!), down and in is “bad” and up and out is “good.”
Experienced architects understand subtle nuances; the art part of the art / science of enterprise software. When you speak with an experienced architect, you may hear her say, “It depends,” often. Good architects will finish the sentence and share at least some of the things upon which “it depends.” Less-experienced architects will present naked scalars and polarized advice.
Naked scalars are numeric values in a vacuum. They are unsupported because most are unsupportable. In other words, they are lies. Now, “lies” is a pretty harsh word. I prefer an engineering definition for the word “truth” that sounds an awful lot like the oath witnesses are asked to swear in US courts:
“Do you promise to tell the truth, the whole truth, and nothing but the truth, so help you God?”
This oath covers falsehoods that are shared, yes; but it also covers omissions of the truth.
Examples of naked scalars:
“97% of engineers believe ____.”
“10% of the people I know have practiced ____ successfully.”
Polarized advice can be a special case of naked scalars, advice focused on 0% and 100%. Polarized advice may or may not include scalars (naked or otherwise).
Examples of polarized advice:
“I’ve never seen a good use case for ____.”
“You should always ____.”
Are naked scalars and polarized advice always bad and wrong? Nope. That would be a generality (and we covered generalities already, did we not?).
Managing the Risk of Inexperience
What exactly is a consultant communicating when they engage naked scalars or polarized advice? They are signalling a lack of experience. They are, in effect, stating, “I do not have experience with ____.”
How do you manage the risk of inexperience? You hire people – architects, especially – who understand there are good reasons systems are designed as they are. They will say things like, “I’m not sure why this was designed this way,” and mean it. It’s not a criticism; it’s an admission of curiosity. Trust me on this: You want curious consultants. They are more likely to identify a solution that solves the problem you are trying to solve in a way that doesn’t create new problems. Returning to the good reasons systems are designed as they are…
Are (or were) the good reasons, well, good? Sometimes.
Do the good reasons scale? Sometimes.
Do the good reasons stand the test of time? Sometimes.
Good architects discern the baby from the bath water. Their experience separates good architects from the crowd. Not-as-good architects are less flexible, less willing to learn, and loathe to admit mistakes.
Let’s face facts, though: All architects and developers know what they know and don’t know what they don’t know. Better architects recognize these uncomfortable truths and mitigate them.
One way to mitigate inexperience – the best way, in my opinion, is to work with others.
The Story Of Us
At Enterprise Data & Analytics, our consultants and architects work together as a team. Our diverse group is a strength, bringing perspective to bear on the problems you are trying to solve. Our experience levels vary, the software and tools with which we work vary, and our demographics vary. As owner, I am honored to lead a team from diverse cultural – as well as diverse technical – backgrounds.
I didn’t set out to build a cultural- / age- / gender-diverse team. I set out to find the best people – to do what Jim Collins describes as “getting the right people on the bus.”
I found, though, that focusing on getting the right people on the bus had the side-effect of building a cultural- / age- / gender-diverse team.
As an added bonus, people of different genders approach problem-solving differently. People of different ethnicity pick up on stuff – especially cultural stuff, including enterprise culture – that people of other cultures miss.
EDNA‘s diversity is a strength that emerged unintentionally, but emerged nonetheless. As Chief Data Engineer, it’s very cool to watch our less-experienced consultants growing into more-experienced consultants and architects, while at the same time watching our people interact and perform as a team – each member catching stuff and contributing ideas because of their unique perspectives.
I can hear some of you thinking, “We’re on a budget here. Don’t good architects cost more than less-than-good architects, Andy?” I feel you. The answer is, “No. Good architects cost less than less-than-good architects.”
It’s often accurate that good architects cost more per hour than less-than-good architects. Do you know why good architects charge more per hour?
Because they are worth more per hour.
But consider this: “Time will tell” is a tried and true statement. Like good wine, the likelihood a generality is accurate improves with age. If enterprises continue to hire an organization – like Enterprise Data & Analytics or any other firm – to help them solve the problems they are trying to solve, then the folks shouting them down may be doing so in an effort to compete. Competition is fine, but I never hire anyone who talks bad about other clients or the competition. Why? They’ve demonstrated the capacity to talk bad about me at some later date.
I love our team! I love our expertise! I love our diversity! I love that we always deliver value!
I’ll be at the PASS Summit 2018 next week. I’m delivering a full-day precon Monday, presenting Faster SSIS and participating in the BI & Data Visualization Panel Wednesday, and Enterprise Data & Analytics is exhibiting Wednesday through Friday.
Monday 5 Nov 2018, I’m delivering a full-day pre-conference session titled Intelligent Data Integration with SSIS. I’m going to cover everything listed at that link but there is an update about my precon content:
There will be Azure Data Factory content and demos!
Why this addition? Two reasons:
My presentation titled Faster SSIS was selected for Wednesday, 7 Nov 2018 at 1:30 PM PT in room 612. I usually include the three Faster SSIS demos in my precon. This time, you can just view the Faster SSIS session to see those demos. I’ve added some Azure Data factory to the precon in place of these three demos!
“How many Data Integration Lifecycle Management tiers (DILM tiers) do I need to deploy SSIS, Andy?” If I had a nickel for every time I am asked this question, I would have a lot of nickels. My answer is, “Four.”
I’m glad you asked. Before I answer, I have a question for you: Have you ever been told by a developer, “It works on my machine.”? Do you know why it works on the developer’s machine?
Developers build software that contains configurations information. In the first iteration of building the software, most developers do no externalize – or parameterize – all configuration values. Instead, they hard-code these values into the application, usually as variable defaults.
I can hear you thinking…
“But This Is SSIS, Andy”
And? Oh, you ask because SSIS stands for “SQL Server Integration Services.” I get that. It’s important you understand this:
SSIS development is software development.
SQL Server Integration Services suffers somewhat from its name which contains the name of a popular relational database engine (SQL Server). But SSIS is not a database, and SSIS development is not database development (and most definitely not database administration).
SSIS development is software development.
Software Best Practices Apply
Because SSIS development is software development, software development best practices apply. For example, SSIS developers should source-control and test their SSIS code.
Perhaps testing is that thing you do in Production right after deployment. I have this pithy saying:
While I freely admit this statement is pithy, it is also true. The remainder of this quote is even pithier: “Some software is tested by your soon-to-be-former largest customer.”
I don’t want you to lose customers – especially your largest customer. That’s why I advocate you test your software by developing it on one Data Integration Lifecycle Management – or DILM – tier and then promoting the code to at least two other DILM tiers before deploying to the Production tier.
Tiers. Or Tears. Your Call.
So which tiers do I recommend? I’m glad you asked! i recommend:
You must build the software somewhere. Wherever you build the software is your development environment.
Now let’s take a step back and describe the development environment for an SSIS developer. An SSIS development DILM tier shares some characteristics. In Dev, SSIS developers can:
The most important part of the SSIS Development DILM tier? SSIS developers can perform each action listed above without help from anyone else in the enterprise.
I can hear you thinking, “What if the SSIS developers blow something up, Andy? Huh? HUH?!?” Let me ask another question: What happens when things go awry or amiss in Production? Isn’t the answer to that question, “I fix it.”? If that’s true in Production…
Then you fix it. In Dev.
In fact (and I know this is a crazy thought) you could actually use a more-open Development tier to test and – perhaps, even – improve break-fix procedures for Production. After all, developers are notoriously excellent testers (I write this as a developer: I can break things like nobody’s business!).
What if the Dev tier becomes a place for you to test your recovery procedures for Production? If your HA/DR procedures work in Development where developers have too much access, then they should work in Production where the environment has been locked down. Am I correct?
Over the years I have found myself making this statement to enterprises: “You have an environment named Development but it is not a Development environment.” How do you know it’s a Development environment? If the developers can destroy it, it’s Dev.
One last thing before I kick this soapbox back under my desk. Consider that developers will be inconvenienced by the destruction of their Dev environment. It may slow them down. Heck, it may cost them a deadline. If that happens, you can take the opportunity to educate them, sharing what happened and why you believe it happened. Maybe they won’t do the dumb thing that burned down Dev again… Maybe.
Would having developers that better understand databases make your job a little easier in the long run?
Once the software works on the SSIS developer’s machine, they need another environment to which they can deploy the SSIS project. Why? Remember those defaults? SSIS packages will always execute in Dev because default values are aimed at local resources for which the SSIS developer has authorization and access.
Which resources? Required access (CRUD – create, read, update, delete) to files and directories, databases, servers, etc. No SSIS developer would claim an SSIS package is ready unless and until she or he successfully executes the package in the SQL Server Data Tools (SSDT) debugger. But a successful SSDT debugger execution does not a completed SSIS package make.
To suss out missing configurations (externalization, parameterization), one needs to move the software to another lifecycle management tier and execute it there. Only then can the developer be sure they’ve externalized required defaults.
I can hear you thinking, “But Andy, this is the same list of permissions as Dev!” Yes.
This part of the work must be completed. As professionals, we get to decide when and where the work gets done. Remember, it can always be done in Production. And it will be done in Production unless we are intentional about doing this work on another DILM tier.
Someone needs to do this part of the work. Do you want to do it? Or would you rather have the person most familiar with the code – the SSIS developer – do it? They are not going to catch everything that needs to be externalized, either; not on this pass. But the Pareto Principle informs us that they will catch 80% of the missing externalization.
Deployment to the DILM Test tier serves as the first deployment test. I want the SSIS developers bundling up SSIS deployments. Why? It’s good practice and practice makes perfect.
One last point regarding software testing: When estimating a software project I begin by estimating how much time it will take to build the software. After this step the calculation for how much time is required to test the software is really easy: It’s the same amount of time required to develop the software.
Andy, are you telling me it takes just as long to test software as to develop it?
“Cutting Testing to Save Time”
You can change this relationship by “cutting testing to save time.” But you can never shorten testing, only lengthen it. Decades of software development experience inform me that the costs of testing software increase by an order of magnitude for each tier of testing. The most expensive environment in which to test is Production.
Cutting testing never saves time.
Or cutting testing does save time; the time it takes you to deliver broken software. I can hear some of you thinking, “But can’t SSIS developers just develop better SSIS packages with fewer bugs?” The answer is the same as the answer to the physics question: “Can’t light just be either a wave or a particle, and not both?” Physics says no, light is both a wave and a particle. Andy says the same physics apply to software development along with the Theory of Constraints which tells us losses accumulate, gains do not.
As my friend Aaron Lowe used to say (before he passed away earlier in 2018 – man, I miss him, and others… we lost too many SQL Family members this year…), “Math is hard.”
You must include the cost of lost customers in the calculation.
Remember, all software is tested…
The QA (or UAT or whatever you call the environment sitting just this side of Production) environment should be locked down tighter than a tick. The security in place in this pre-Production tier should be identical to the security in the Production tier. Optimally, all hardware capability and storage capacity in Production will match QA.
If you understand the dynamics of SSIS Data Flow Task internals, it’s possible to test performance with less data on sub-optimal hardware. If you ask me to do it, I’m going to let you know there’s a possibility we will miss something. I’ll be able to tell you for sure after we run some performance test executions in Production (and we will need to do that to be sure, and that will cost both time and money, so how much money are you saving when you save money by buying sub-optimal hardware and storage for QA?).
Deployment to QA should be performed by the same team – optimally by the same individual – that will be performing the deployment to Production. Why? We do not want deployment to Production to be their first experience with deploying this SSIS project. The deployment to QA is another deployment test, this time for the Production release management person. Perhaps this is a DBA. Perhaps this is a Release Management team member. Regardless, this step is their opportunity to practice the deployment to Production.
I can hear you thinking, “What’s the big deal, Andy?” I’m glad you asked. The big deal is: No matter how you execute SSIS in your enterprise, there’s more to the solution than merely the SSIS packages. If you’re running from the file system (and most enterprises execute SSIS from the file system), there are configurations stored in dtsConfig files or database tables that also need to be promoted. Some of the values – such as connection string components – need to be edited in these dtsConfig files at each DILM tier.
The same holds for deployments to the SSIS Catalog. And… if you are deploying (or planning to deploy) SSIS to the Azure Data Factory (ADF) SSIS Integration Runtime, your only option is deployment to an SSIS Catalog. Externalization is accomplished via Catalog Environments, References, and Literals in the SSIS Catalog. Same stuff, different location. But both dtsConfig files and Catalog Environments must be managed.
Since editing will need to occur when it’s time to deploy to Production, I prefer the deploy-er practice the editing along with the deployment. I ask the developer to send the SSIS and scripts or files for managing external configurations metadata to the deploy-er, and I request these configuration artifacts be set up for the Test DILM tier.
If you’ve taken my advice and marched SSIS development through this process or a similar process, the Pareto Principle says you’ve identified at least 96% of the bugs prior to deployment to Production. My experience bears this out; about 4% of my Production deployments fail because some parameter slipped through the cracks walking through Dev, Test, and QA.
I can hear you thinking, “How does that happen, Andy?” Excellent question (and I am glad you asked!). parameters slip through the cracks because Dev and Test may be sharing data sources and/or destinations. It may be possible – desirable, even – to access Development data from the Test tier and vice versa.
Remember: I want QA to be identical to Prod, which means I want it equally isolated. Is Production behind a firewall? Is it impossible (without poking a hole in the firewall) to write from Test to Prod? If so, then it should also be impossible to write from Test to QA and from QA to Prod. If Prod has a firewall, QA should have a firewall.
If your enterprise sports this architecture you can beat Pareto’s 4% prediction by 3.2% (an additional 80% of bugs will be caught) and experience a deployment-to-Production success rate of 99.2% (a failure rate of 0.8%).
To quote Foghorn Leghorn, “figures don’t lie.”
Is this the only way to properly manage SSIS in an enterprise? Goodness, no.
You can add more tiers to your Data Integration Lifecycle as needed. You can technically get by with three tiers (Dev, Test, Prod) but I don’t recommend it unless you’re the person wearing all the hats. Even then, the more times you test deployment, the more bugs you will catch prior to Production, and the fewer bugs you will deploy to Production.
It’s always easier to deploy more bugs but it is never less-expensive and it never saves time. Plus it may cost you your soon-to-be former largest customer.
Has this ever happened to you? You are opening an SSIS project and… it won’t open. Instead of a Control Flow filled with awesome tasks and containers, you see a message in Solution Explorer telling you “The application is not installed.”
You know this isn’t right because you built the SSIS project on this same machine just a day or two ago! What gives?
Well, you may have seen a message informing you the Integration Services Projects 2.0 extension took too long to load:
The error notification / message reads:
Extension ‘Microsoft Integration Services Projects 2.1’ likely caused nn seconds of unresponsiveness. Disabling it may improve your experience. Disable the extension | Learn more | Never show this message again
If that happens, you are presented with three options:
Disable the extension
Ignore the message
Never show this message again
If you click option #1, you create this very issue.
How to Fix It
Open Visual Studio and click Tools–>Extensions and Updates:
When the Extensions and Updates window displays, select the Microsoft integration Services Projects entry and click the Enable button:
The Enable button text will change to Disable after the Microsoft Integration Services Projects extension is enabled:
Note: the change will not take effect until you restart Visual Studio.
Restart Visual Studio.
Depending on how you open the SSIS project, you may still get the same messages in Solution Explorer:
I am honored to join my friends at the Triad SQL PASS BI Group in Greensboro North Carolina 30 Oct 2018, where I will present Faster SSIS!
Ever wonder why SSIS runs so slowly? Watch SSIS author Andy Leonard as he runs test loads using sample and real-world data, and shows you how to tune SQL Server 2016 Integration Services (SSIS 2016) packages. We’ll start by experimenting with SSIS design patterns to improve performance loading AdventureWorks data. We will implement different change detection patterns and compare execution performance for each. Then, we’ll explain a Data Flow Task’s bottleneck when loading binary large objects – or Blobs. Finally, we’ll demonstrate a design pattern that uses a Script Component in a Data Flow to boost load performance to MySql, whether on-premises or in the cloud. Prerequisites: Familiarity with SQL Server Integration Services (SSIS).
The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to this.