Uber Automates Service Mesh Dependency Mapping
The new system replaces slow distributed tracing to identify critical failure points in real-time.
Updated on Oct. 16, 2026 in Data Centers
Live Poll
Do you believe automated monitoring systems make the digital services you use more reliable?
Uber has developed an automated system to map service mesh dependencies, identifying how failures in one service affect others. The system classifies these relationships as fail-close or fail-open using RPC middleware to improve upon traditional distributed tracing.
Why it matters
Traditional distributed tracing proved too slow for Uber's scale, often failing to capture rare dependency errors. This new approach provides faster feedback loops, allowing engineers to identify critical failure points more reliably than previous methods.
The system identifies fail-close dependencies with a failure correlation probability of 0.8 or higher, while fail-open dependencies are marked at 0.2 or lower. It processes 500 requests per second, a performance density required because traditional tracing previously took over five hours to capture a single failed request.
The players
Uber
A ride-sharing and food-delivery technology company that maintains a large-scale service mesh architecture.
The details
The system utilizes inbound and outbound RPC middleware—software that acts as a bridge between different applications—to track request outcomes in real time. Inbound middleware generates request IDs and logs caller data in an in-memory tracker, while outbound middleware records downstream service endpoints. By ignoring retries during this logging process, the tool prevents incorrect dependency identification, mapping how a caller and callee service interact within the Muttley proxy environment.
Timeline
October 16, 2026: Official publication of the mapping system documentation.
The Tech Race
Uber is moving away from traditional distributed tracing to address the visibility gap in its service mesh. This development follows a pattern of large-scale infrastructure providers creating custom middleware to bypass the limitations of standard observability tools.
This tool is an internal engineering solution and does not directly change user-facing ride or delivery applications. It serves as a model for platform engineers at large firms who face similar latency and error-detection bottlenecks in high-volume microservice environments.
The takeaway
Automated dependency mapping is replacing standard tracing as the preferred method for identifying failures in high-traffic service meshes. Engineers should watch for future open-source releases or similar middleware implementations that could replace legacy tracing for cloud-native architectures.
Further reading
For more on how major platforms manage infrastructure scale, see the Data Centers section.
Source note: This article includes information reported by InfoQ.
Live Poll
Do you believe automated monitoring systems make the digital services you use more reliable?
