What is an error budget in SRE, and how does it help balance reliability with feature development?

Question

Accepted Answer

An error budget is the maximum amount of unreliability that a service can tolerate while still meeting its Service Level Objective. It represents the difference between perfect reliability and your target reliability, expressed as a quantifiable amount of acceptable failures or downtime.

The calculation is straightforward. If your SLO is 99.9 percent availability, your error budget is 0.1 percent. Over a 30-day month, this translates to approximately 43.2 minutes of acceptable downtime. If your service processes one million requests daily, the error budget allows for 1,000 failed requests per day.

Error budgets help balance reliability with feature development by creating a shared framework for decision-making between development and operations teams.

When the error budget is healthy, meaning you have plenty of budget remaining, teams can take more risks. This might include deploying new features more frequently, running experiments, or performing maintenance that could cause brief disruptions. The budget provides room for innovation and change.

When the error budget is depleted or nearly exhausted, teams must prioritize reliability work. This might mean slowing down feature releases, investing in stability improvements, or rolling back recent changes that increased error rates. The depleted budget signals that the service needs attention before more changes are made.

This approach eliminates the traditional conflict between "move fast" and "keep things stable." Both goals are valid, and the error budget provides an objective way to determine which should take priority at any given time.

Error budgets also provide accountability. Development teams own the budget and are responsible for how it is spent. If a bad deployment consumes half the monthly error budget, the team feels the impact directly through reduced ability to ship new features.

The key principle is that 100 percent reliability is neither achievable nor desirable. Users cannot perceive the difference between 99.99 percent and 100 percent, but the engineering cost to achieve that last 0.01 percent is enormous. Error budgets acknowledge this reality and turn it into a practical management tool.