Part 2 of this article series on machine learning (ML) and autonomous databases concluded with questions about the impact of automation on the DBA role and whether we need to start coming to terms with ML and autonomous databases.
Starting with the first question, automation is not cloud (although database as a service does have some key advantages). Automation entails fundamental changes in how we interact with machines (or the lack of interaction).
An Autonomous Analogy
There have been many articles expressing the view that autonomous vehicles will be mainstream in about 5 years for highways and urban areas. Without going into too much detail, there are various levels of autonomy—level four being highway and urban (well-mapped environments) and level five being fully autonomous (no human interaction needed anywhere). At level four or five, there could be a fleet of vehicles at the ready to swing by to transport you to your final destination with a click.
Are there similarities with autonomous databases? Yes. There are many typical/traditional tasks and activities that can and will be automated. When autonomous databases and cloud deployment become mainstream, we will have a platform that is secure, self-administering, self-patching/upgrading, self-tuning, self-healing (more to come on this), and much more. ML will be doing the driving for us.
The cloud is an important part of the equation for database transformation. When software and infrastructure are deployed in the cloud, vendors can push continuous deployment of fixes, new features, and new functionality. That means that if vendors get changes wrong, it will be quick and relatively easy to fix when they host it, with the ability to minimize risk by pushing the changes out to a subset of customers. With this ability comes a tendency toward additional risk-taking, as well as faster development and deployment. Release cycles can become weeks, not months or years, and this will speed up the transformation to automation. It is happening now.
Keeping an Eye on Artificial Intelligence
Moving to automation means many decisions and modifications will occur without human involvement. Alterations are likely to occur that are not to the benefit of the end user (we will not get this 100% right in its nascence). Data “owners” will need very good information about what was adjusted, when it changed, and whether it was ultimately good for the user experience. The right tools will be needed as uptime, resource consumption/capacity, and relative health become basic “pay-to-play” propositions managed by automation. As end user experience continues to trump all other concerns in the application world, taking the end user vantage point will be central to monitoring autonomous changes.
Will automation replace many tasks a DBA currently performs? Undoubtedly. Do we need to start coming to terms with the changing winds surrounding ML and automation? We would be wise to do so. (For more on this, read part one of this series.)
Anomaly detection will play a significant part in ML and automation. Anomaly detection is statistical modeling and probability theory dusted off. Take the set of datapoints, find the right way to structure the statistical model for the distribution of the data, and look for outliers—things you want to know about. If an outlier is observed, do X (notify, kill a process, run an OS command, etc.).
Take this one step further. When an incident occurs that impacts end-user experience in a negative way, it is flagged. ML will gather a profile about the occurrence—it will note contributing factors such as the fact that connections increased by 10%, active memory usage went above 80%, a specific wait state became an outlier, or an anomalous query was running. Having this profile, ML will become “aware” of when another occurrence is likely to happen (probability). The anomalous query is running again, active memory is starting to increase, connections are starting to rise, or the same wait state is gaining momentum. Now the ML has the descriptive, predictive, and prescriptive analytics to manage without human involvement within defined parameters. And, it gets smarter with time.
This transformative process will not be limited to databases or IT. Humans think in linear terms while computers act/learn exponentially—the rate of change could surprise us. Is automation to your benefit? It will depend on your role. At a recent major conference, a CEO stated that autonomous databases will be here within a couple of years, and customers are going to have to be willing to pay less for them. Where do you weigh in?