Thursday, March 21, 2024

Mixing Db2 Database Administration with DevOps - Part 4: Database Schema Change and DevOps

Traditionally, the DBA is the custodian of database changes. But the DBA typically is not the one requesting the change. Usually, a programmer does that. There are times when the DBA requests changes, such as to improve performance or to utilize new features, but this is not as common as development changes. Regardless of who requests the change, the DBA must be involved to ensure that each change is performed successfully and with no impact on the rest of the database. 

After moving to a DevOps approach, there is a shift that places more of the responsibility for database change on the developer. But to succeed, the DBA still must be involved to oversee, analyze, and approve any changes. As is common with DevOps practices, it is desirable to automate as much of the process as possible in order to remove manual, error-prone tasks and increase the speed of delivery. 

This requires a tool that automates complex database changes and integrates into the DevOps toolchain. Without such a tool incorporating database changes into application delivery and deployment remains a slow, mostly manual process. 

To effectively make database changes, the DBA needs to consider multiple issues, the most important of which are the appropriateness of the change in terms of the database design and the impact of the change on all other database objects and applications. Additionally, the DBA must determine if the change conforms to standards, how best to implement the change, and the timing of the change. 

The ideal arrangement is for database schema changes to be incorporated into the DevOps toolchain using a tool that allows developers to request changes. Those changes should be analyzed and compared against standards and rules for conformance. Non-compliant changes should automatically be referred back to the developer for modification and resubmission. Compliant changes should be accepted and cause a script to be generated using the most appropriate mechanisms to implement the change. This is a non-trivial activity which if done properly can eliminate a lot of manual downtime. The generated script should be presented to the DBA for review and upon acceptance, be implemented.

Db2 for z/OS, like all of today’s major DBMS products, do not support fast and efficient database structure changes for all types of change. A quick example: try to add a column to the middle of an existing row. This requires a complex series of metadata capture, data unloading, DROP, and CREATE statements. And don’t forget about all of the dependent objects and structures like indexes, referential integrity, authorizations, and so on.

Adding to the difficulty of making schema changes is the fact that most organizations have at least two, and sometime more, copies of each database. There may be copies of the database at different locations or for different divisions of the company. And at the very least, a test and a production version will exist. But there may be multiple testing environments—for example, to support simultaneous development, quality assurance, unit testing, and integration testing. Each database change will need to be made to each of these copies, as well as, eventually, to the production copy. So, you can see how database change can quickly monopolize a DBA’s time.

You can see how a robust, time-tested process that is designed to effect database changes is required. BMC, Liquibase, and IBM all offer DevOps-integrated database change management solutions for Db2 for z/OS.

As you review their capabilities, be sure that the tooling supports the type of changes your organization requires. For example, be sure that the tool is aware of all the different requirements for making any change you may need.

From my experience, vendors can focus on the development side of the DevOps experience and minimize the complexity of database change. All too often the tool demo shows a request to add a column... how boring is that? How about changing a data type from numeric to text? That would be a bit more challenging... or requiring a tablespace be converted to Universal from segmented as part of the change (perhaps to support larger sizes)?

No comments:

Post a Comment