Skip to content

O'Reilly Report | What is Data Observability?

Disclaimer: This post is a preview of the O'Reilly Report. If you wish to directly download it, click here



While IT and DevOps have numerous quality control measures (such as development practices like continuous integration and testing) to protect against application downtime, most organizations don’t have similar measures in place to protect against data issues. But just as organizations depend on high levels of reliability from their applications, they also depend on the reliability of their data.

When data issues—such as partial, erroneous, missing, or inaccurate data—occur, the impact can rapidly multiply and escalate in complex data systems. These data incidents can have significant consequences and multiple negative repercussions on the business, including lack of trust and loss of revenue. A lack of control or visibility into the data quality can also produce faulty insights, which can lead the organization to make poor decisions that can result in lost revenue or a poor customer experience.


How can Data Observability help?

In your organization, has someone ever looked at a report and said the numbers were wrong? Likely this has happened more than once. No matter how advanced your data analytics and modeling tools are, if the data you’re ingesting, transforming, and flowing through your pipelines isn’t correct, the results won’t be reliable.

Discovering data failures is much more challenging than typical application or system failures. In the case of an application, when something isn’t working, the symptoms are more evident. For instance, if an application crashes, freezes, or restarts without warning, you know you’ve got a problem. However, data issues are generally hard to notice since the data won’t freeze, crash, restart, or send any other signal that there’s a problem.

This is where data observability comes into play. You need to be able to observe and be aware of these types of silent changes to the data so that you can fix them preemptively—before your CEO is coming to tell you that numbers look wrong. Data observability is a solution that provides an organization with the ability to measure the health and usage of their data within their system, as well as health indicators of the overall system. By using automated logging and tracing information that allows an observer to interpret the health of datasets and pipelines, data observability enables data engineers and data teams to identify and be alerted about data quality issues across the overall system. This makes it much faster and easier to resolve data-induced issues at their root rather than just patching issues as they arise.

Critically, what makes a data observability solution unique from application observability is that the data must be logged and traced from within the data pipelines, where the data is created and activated for use. This allows you to measure the pertinence of the usage of your data within your entire system (all your applications and pipelines) as well as monitor health indicators of the overall system. Why is this important? Because while your data pipelines may look fine—the data isn’t using too much memory or taking up too much storage space—if the data outputs from those pipelines are providing garbage data, then the value of the data is worthless. On the other hand, if you only observe a table within your database, it may tell you what queries were performed, but you won’t know how it’s being used, or by whom.

This matters because you won’t know if the data being used is of value to the end user. Thus, to ensure that the data analytics you're performing are accurate and valuable, you need to observe both the data and the pipelines at the same time.

If you’re unable to have this type of end-to-end visibility, you’re likely to suffer what we call a data incident or a data failure. If you'd like to know more about what this means and why they should be avoided at all costs, download the report.