The Case for More Flexible Distributed Tracing

Distributed tracing has been widely adopted as a means to troubleshoot distributed systems in which traditional, device-centric debugging loses traction. However, the debugging capabilities of tracing are underused today because of the immutable nature of tracing instrumentation. Additionally, current tracing solutions do not address in-network compute, and hence cannot be considered end-to-end.We argue for a new tracing workflow in which tracing instrumentation is made more flexible, allowing DevOps to use a query language to specify tracing tasks for computation happening both on x86 servers and network devices.