DigitalOcean Overhauls Availability Metrics for Improved Customer Experience
DigitalOcean has announced a significant overhaul of its availability measurement framework, aimed at providing a more accurate reflection of customer experience. This change, which began in early 2025, addresses discrepancies between reported uptime metrics and actual user experiences, ensuring that the platform’s performance aligns more closely with customer expectations.
Identifying the Problem: Inaccurate Availability Metrics
At the start of 2025, DigitalOcean’s internal analysis revealed a troubling inconsistency in its availability numbers. The platform’s monthly availability fluctuated between 99.5% and 99.9%, often influenced by declared high-severity incidents rather than actual performance. Despite these seemingly high numbers, customers continued to experience issues that were not reflected in the metrics, leading to escalations and dissatisfaction.
The previous methodology for measuring availability had served DigitalOcean well during its early growth stages but became increasingly inadequate as the company expanded its offerings. The incident-based approach treated all declared incidents as total outages, while lower-severity issues were effectively ignored. This created a structural dilemma: expanding coverage to include minor issues would artificially degrade the overall availability metric.
Introducing a New Framework: Control Plane and Data Plane
To address these challenges, DigitalOcean implemented a new operational framework that separates availability measurements into two distinct categories: Control Plane and Data Plane. Each category utilizes different Service Level Indicators (SLIs) tailored to their specific functionalities.
The Control Plane encompasses orchestration tasks such as API calls and Cloud Panel operations. Here, success is measured by the rate of valid requests while excluding client errors (4XX) from failure counts. This approach allows for a clearer understanding of how users interact with the platform without conflating user errors with system failures.
The Data Plane focuses on live product instances like CPU and GPU Droplets, DOKS Clusters, and other resources. Measuring availability in this area is more complex due to varying failure modes across different products. For instance, some products are evaluated based on resource minutes—indicating whether they are available and healthy—while others use request-based metrics similar to those in the Control Plane.
A Comprehensive Approach to Aggregation and Alerting
With the new framework in place, DigitalOcean faced additional challenges related to data aggregation across multiple regions with varying traffic volumes. To ensure accurate representation of overall performance, the company adopted a weighted request average approach that accounts for traffic distribution across data centers (DCs). This method prevents smaller DCs from disproportionately affecting global metrics.
For example, during an incident where one DC handles only 20% of total traffic, its impact on overall availability is appropriately minimized through this weighting system.
In terms of alerting mechanisms, DigitalOcean transitioned from single burn rate alerts to a multi-window alerting system that combines short-term and long-term metrics. This dual approach allows for more nuanced detection of issues while reducing false alarms triggered by brief spikes in error rates.
Error Budgets: A New Decision-Making Tool
The introduction of error budgets marks another significant advancement in DigitalOcean’s operational strategy. An error budget represents the allowable percentage of downtime within a specified window—in this case, a rolling 30-day period rather than a fixed calendar month. This shift acknowledges that customer trust accumulates over time and does not reset monthly.
- Green Zone (0-60% usage): Normal operations can continue.
- Yellow Zone (61-80% usage): Caution advised; verify no impact on dependencies.
- Orange Zone (81-100% usage): Increased risk; large rollouts paused; focus on reliability work.
- Red Zone (>100% usage): Critical risk; all rollouts halted except for essential maintenance.
This structured approach enables teams to make informed decisions about product releases based on current reliability metrics while prioritizing customer experience during periods of instability.
Extending Frameworks to New Product Lines
The principles established within this framework are not limited to core infrastructure products like CPU Droplets or Managed Databases; they also extend seamlessly to newer offerings such as GPU Droplets and AI agents. By applying consistent SLIs across all products, DigitalOcean ensures that all services adhere to the same rigorous standards for availability measurement and customer satisfaction.
What This Means for Customers
This overhaul signifies DigitalOcean’s commitment to transparency and reliability in its service delivery. By implementing a more accurate measurement framework that reflects actual user experiences rather than arbitrary incident counts, customers can expect improved service quality and responsiveness from the platform. As DigitalOcean continues to refine its operational strategies, users can have greater confidence in the stability and performance of their cloud services.
For more information, read the original report here.
