Introducing SLI/SLO to Improve Service Reliability - Part 3: Service Application Cases

LINE's Service Reliability team operates SLI and SLO as quality standards for its messaging core services, while also pursuing demand forecasting, performance testing, automation, and stronger observability together. This installment focuses on messaging, channel, and authentication services to summarize how SLI/SLO is actually applied to real services.

Introducing SLI/SLO is not just about picking metrics—it's a process of redefining the service's core journeys. First, you need to identify the CUJ (Critical User Journey) based on user experience and business goals, and decide within it what should be considered the most important functionality.

• The organization with the deepest service understanding defines the CUJ, the infrastructure team builds the large-scale measurement environment, and SRE implements the SLI measurement tools.
• Example CUJs include the service sign-up flow, sending and receiving messages, LINE Login, profile information, and authentication/encryption functions.
• All stakeholders must share the same standard for SLO to serve as the criterion for operations and launch decisions.

At the SLI stage, for each CUJ you choose the measurement point among gateway, frontend, and backend, and designate representative APIs to distinguish success from failure. For example, a message send request might set a standard of 99.9th percentile response time under 500ms, while overall requests might target a 99.999% success rate. If measurement is too complex or success/failure cannot be clearly separated, that CUJ is excluded, and if necessary, a dedicated SLI-specific metric is created instead.

SLO is the stage where you decide how well the defined SLI must be met over a given period. Based on 28 days (40,320 minutes), a 99.9% SLO allows for about 40 minutes of error budget. For example, if you exceed the 99.9% SLO and achieve 99.98%, with error budget remaining at 80% or more, the service can be considered stable and more investment can be directed toward new features. Conversely, if the remaining budget is around 10%, resources should be redirected toward stabilization and reliability improvements.

At the visualization stage, you build dashboards that show SLO and error budget at a glance. The overall status uses a simple color scheme for quick judgment, while detailed per-CUJ dashboards are separated out for in-depth analysis. Items that meet their targets well should be shown in green, items needing attention in orange, and failing items in red.

In operations, SLI quantifies service health, and error budget determines resource allocation. On-call response and regular meetings also make use of error budget alerts to connect incident response with preventive activities. Ultimately, SLI/SLO is a tool for managing reliability and business goals at the same time.