Automatic DBA

Data refreshes are one of the hardest things to accomplish in an Agile environment: The downtime between sprints is miniscule and most modern systems have quite a bit of metadata that must be contained within specific environment such that ‘unwanted’ interactions do not occur. “Like what?” I can here you say. How about sending account cancellation to active members during a test. Yeah…I can hear the gasps of horror from here.

This is the Redgate Tools Team, a small software company out of Cambridge, England. How small? Here’s the entire group…..

Ahh, the English, their Easter Egg and their dry sense of humor. Anyway, Redgate builds the best SQL management tools I’ve ever seen: Fast, reliable, and easy to use. They have quite a tool selection, but my favorites are SQLCompare and SQLDataCompare.

SQLCompare is a schema level comparison tool that can either report on the differences between two databases or it can alter the target database to match the sourcedatabase…VERY POWERFUL. There are a few gotcha’s here and there (like circular relationships…has a hard time with those), but when you are in a hurry it’s a great time saver.

While SQLCompare is centered on structure, SQLDataCompare is all about what is in those structures, namely data. Given two databases, it can compare every row on every table with a corresponding row on a corresponding table. Quick…real quick. The configuration can be intimidating, but once you have a handle on it you can get a lot of work done quickly and reliably.

And the best part of Redgate products are that once a project is created and configured it can be run from the command line, which in turn is easily integrated into a schedule, SSIS, or FinalBuilder automation.

So here’s the scenario:

• Four environments must have refreshed data from PROD: DEV, TEST, QA, and PROOF.
• Each environment has 5 major databases with a footprint of about 500GB.
• > 100 tables hold environmentally sensitive data spread out over 5 databases.
• There can be NO ENVIRONMENTALLY SENSITIVE DATA LEAKAGE! (e.g., Data from PROD can’t exist in TEST if it could cause unwanted results or damage.)
• Any one environment can be down no longer than 24 hours.
• All the refreshes have to be done in a calendar week.

So…Here’s the game plan:

We created a new environment (What’s that? Number 7?) that will be used as a staging area where the data refresh preparations will be conducted. This environment will hold an extra database that will gather and hold all environmentally sensitive data from all the environments using SQLDataCompare.

1. Each Saturday a FinalBuilder project is scheduled to gather all the environmentally sensitive data from each environment and store it in the hold area.
2. Each Sunday a SQL job restores a production copy of all databases down onto this refresh preparation environment.
3. A FinalBuilder project is scheduled for 8AM each day:
   1. It recovered non-essential space from the databases in the form of audit, deprecated, and staging tables.
   2. Environmental sensitive data was pulled from the holding area and overlaid on the PROD data depending on the day:
      1. Monday = DEV
      2. Tuesday = TEST
      3. Wednesday = QA
      4. Thursday = PROOF
   3. The data is scrubbed of personal/sensitive data.

On Monday, which is the first time the space recovery and cleansing routines run, it takes about 6 to 7 hours. Each subsequent day takes about 5 minutes.

1. The Production DBA’s backup the refresh environment and restore it to the target environment.
2. The Production DBA’s resync the security on the target environment. (So that it works the way it did before the refresh.)
3. Conduct any necessary code promotions (either though automation or a request, depending on your process).

There…All done. A data refresh used to take all week without the ability to do anything else. Now…we could do all environments in a single week and still have time for additional tasks. Gotta love Redgate.

Once I took over from the previous App DBA, it became apparent rather rapidly that builds were an exercise in patience…and they took a lot of time. Developers requested scripts be run in multiple environments in what seemed to at random intervals. The master turn list was informal and updated manually. Some of the scripts ran out of order, sometimes with devastating effect. Some of the scripts were environment specific without a way to manage their operation. In a word…CHAOS!

So I told my boss that I was going to automate the SQL builds…”You can do that?!??” My response: “With the correct tools and a documented process you can automate anything.”

Based on what we discovered doing the Process Inventory we know that we can get a big payoff from automating the SQL Build. And from what we covered in my series on documentation, you can’t automate what you don’t understand. So, it looks like we need to breakdown and document a process that will do the following:

• Keep track of changes to the SQL source code.
• Ensure that a script is written so that it can run in every environment.
• Ensure that a scripts runs is every environment.
• Verify that the scripts after they were run.

So with that in mind, he’s the starting scenario: Two development teams in an Agile framework using Microsoft TFS. After several meetings and some chalk talk with the key developers and managers, we came up with the following initial outline.

So this procedure was documented and put into place (with the exception of the FinalBuilder automation). One of the first things we noticed was that we could actually plan a build without worrying about losing some dependent SQL code. The second thing we noticed was that the process was cumbersome. So we got rid of some of the quality gates (How many gates do you really need?) and started coding some customized tools. In this case FinalBuilder.

Seque: If you are doing builds of any kind, YOU NEED FINALBUILDER! Why?

• It’s quick to setup and use.
• It integrates with most open-source platforms and source-code control systems.
• It’s completely customizable, including the ability to use in-house DLL’s. (I’ve done this and it works great!)
• A license is cheap.

OK….So I get back together with the development teams and we revise the process, which turns into this:

Based on this I write the process documentation, paying particular attention to the detail needed to providing the necessary information for:

• Code Promotion Requests
• Production DBA Support
• Sprint Team Feedback
• Build List Verification

Once we the team felt comfortable with what I had, we started the automation process in earnest. Until the automation process was completed, the existing process would stay in place. Once completed with testing the FinalBuilder SQL Build, we cut over and never looked back.

There is quite a bit of the unsaid in this article: I apologize for that. I, unfortunately, have limited time that I can dedicate to writing and this subject is quite detailed and extensive. If you would like to have a fuller explanation of any items, please contact me.

One of the first steps to initiating a maturity model program is to take an inventory of what processes exist. There are generally more processes than initially thought in any unit and it takes a bit of time to sit down and hash out what they are.

1. proc·ess/ˈpräˌses/

   Verb: Perform a series of mechanical or chemical operations on (something) in order to change or preserve it: “the stages in processing the wool”.

   Noun: A series of actions or steps taken to achieve an end. 

 With this definition in mind, how many processes (noun form of definition above) does your unit have? Probably more than you think. Anything can be automated as long as you understand the process steps and you have the necessary resources to accomplish the automation. Let’s take a look at a simple example; Spending Accounts (SPAA) Database Management. Keep in mind this is just an example and not an all-encompassing, cohesive list. Here are some of the processes that exist for managing the database in my old position:

 This is a short list of all the processes that are required, but I’d like to keep it simple for this exercise. Next, let’s categorize the processes into major and minor effort (i.e., how much resources does it take to accomplish – this could be considered resource savings or the REWARD), the dependency type of process (internal, external, mixed – i.e., will I need outside resources to finish the task?), and automation effort (i.e., how much resources would it take to automate – this could be considered the cost or the RISK). This is a much tougher task with a lot of variability. It is best done with a small group of knowledgeable associates; any group larger than 4 is going to lose focus and the discussion will breakdown pretty fast.

So let’s say this is the result (keep in mind this is hypothetical):

 So with these numbers we can start the evaluation/prioritization step. Remember, we are looking to get the biggest bang for the buck. There are many measures you could you, but the one I like to use is the Breakeven Point; at what point did the effort payoff? I understand that this is highly subjective and imprecise, but you have to start somewhere. Here are the results:

 So I can see that Data Refresh and Database Builds are taking the most of my time. They just happen to also be the fastest payoff for an automation investment, so I’ll target them first for automation. I also notice that while these tasks will require the same effort to automate, one has mixed (external) process dependencies so I might want to lower that tasks priority until I can arrange the necessary outside resources time.

Process improvement is a never ending endeavor: On a regular basis (e.g., once a quarter) this exercise should be conducted again to see what has changed and how it impacts existing maturity efforts. Business processes change, new technology becomes available, associates come up with new ideas for automation…These will impact what is possible and how they can be employed to improve the existing processes.

Next we’ll discuss breaking down a process targeted for automation.

One of the items I ran into when I took over as the App DBA for SPAA was the complete disregard of source code control for SQL objects. At times there could be two, or maybe three, different scripts for the same object in the same build. So when the build occurred, it was possible to overwrite change A with change B, which in turn usually resulting in a defect. It was frustrating and it had to change.

So to put an end to this issue we instituted formal source code control for all SQL objects. The rules were:

• Each SQL object would have a single source code, or SQL script, file.
• Each SQL object script would contain all DDL, including security grants, so that they would be whole and complete.
• Each SQL object script would be written in such a way as to Do No Harm (or in Latin: Primum Non Nocere).

Following these rules allowed us to ensure that all changes to a single object were contained within a single file and that file would have a history that could be tracked. They also ensured that if the file was executed, it contained everything necessary to allow proper operation of the SQL object. Lastly, it would allow the script to be run on any environment, at any time, and insuring that the database would be in the proper state.

Do No Harm

The Do No Harm philosophy had two main paths: table scripts, and everything else. The reason the table scripts have to be treated different is that they are objects with permanent state. All the other objects (stored procedures, functions, views, etc.) are stateless. Regardless of which type of scripts was managed, all scripts followed this simple rule:

All DDL manipulation occurred within the confines of a controlled database transaction.

This ensured that the script worked completely, or returned the object to its original state.

So for the stateless objects the rule was drop and recreate. Doing this provided several advantages:

1. You could query the system tables for when the object what created. This works well if you, say, want to verify a turn the Turn Center just completed.
2. If the object didn’t exist, it would just create it and move on. Before it was possible to get an error when you ALTERed an object that didn’t exist.

For objects with permanent state (tables), you have to maintain the state. Therefore you have to alter in place…And this is where it gets tricky. So for table scripts the following rules were followed:

1. Tables are created once, so you had to check for existence before creating the table.
2. DDL could be added to the script, but existing DDL was never altered.
3. As new DDL is added to the script, the appropriate requirement checks had to be made prior to execution. Say for example you wanted to add a column, that was segregated from the original CREATE statement and you checked to see if the column exists prior to ALTERing the table and adding the column.

It did take a little time to train the developers to follow this philosophy, but the payoffs were well worth it:

• It allowed HEAVY automation, including automated SQL builds to the lower environments.
• It reduced downtime during turns.
• It reduced turn preparation.
• It allowed automation of turn verification.
• It reduced defects.

Now all we need is a starting code base….

Creating the Code Base

To establish the original code base files, we needed a way to create ~12,000 individual SQL scripts. No short order by any imagination. So we used the Do No Harm templates as a base to construct a program that read the system tables of the PROD databases. The program extracted all the necessary information to reverse-engineer the T-SQL to a file, then that file was pushed to the code base structure of TFS. Easy!

The question you have to ask yourself is this: Why isn’t everyone doing this?

As Agile goes into a full court press, many seem to downplay the importance of environmental management, which is also sometimes known as configuration management. This is usually to the detriment of all concerned.

The focus of Agile is generally on the various aspects of development: Defining what to develop, how to develop, when to develop. But ask yourself this: How many times has anyone talked about environmental management? Automated testing? Production support? I heard nothing about these until my co-worker Keron and I started asking very sharp questions about migration times, data refreshes, and build verifications. Even the instructors for the SCRUM class we attended were befuddled by our questions. It was like it never occurred to anyone that configuration management was important to a successful Agile implementation.

So why is it important? Any derivative of an Agile methodology must include a framework of continuous integration into target environments. This puts the product (process, code, intellectual property, etc.) into a place where you get the most benefit from it in the least amount of time. From a graphical point of view, our ideal SCRUM process flow looks something like this….

Bob Johnson called this diagram the Left Brain/Right Brain Scrum explanation. The actual start and end points are not important; what is important is the understanding of the two very different functions that promote fast turn-around times for any development process. Also, that the ‘team’ goes beyond those in the SCRUM.

The Right Brain is responsible for creativity and pattern recognition: This is the domain of the Development Team. They work the backlog at the discretion of the Product Owner. They get feedback from the Production Support team and make changes as assigned.

The Left Brain is responsible for process and sequencing: This is the domain of the Environment Team. They manage the output from the SCRUM team(s) in a proven, repeatable manner that is easily automated. They gather data about the process to:

• Implement continuous process improvement.
• Provide feedback to the Production Owner and SCRUM team(s).

Case in point, one of the repeating tasks that occurred in the environment was a data refresh. The refresh was problematic on several fronts:

1. Data had to drop down from the PROD environment to the lower environment.
2. Code had to be promoted up an environment.
3. The environment specific metadata had to stay in the same environment.

Sounds easy enough, but when you’re talking about 5 databases, >1000 tables, and a ~1.5TB footprint nothing is easy. And, ohh, by the way…you have to do it between Sprints and not slow down development. Without getting in the weeds, we set out a plan (which I’ll write about later in detail) that allowed continuous process improvement to evolve the process naturally and incrementally. The outcome?

A little planning, a little automation and, BAM!, a process that takes a couple of hours. Don’t know about you, but could you imagine the savings of time and resources if you did this with just 10% of your company’s IT processes? (Sniff, sniff) I smell shareholder value!