Microservices’ inherent complexity creates significant troubleshooting challenges, particularly in user request analysis. While nonintrusive distributed tracing addresses this by monitoring cross-service execution without code instrumentation, current frameworks still exhibit critical deficiencies: Current approaches suffer from language-dependent instrumentation that restricts multienvironment adaptability and kernel-to-user space data transfers, introducing prohibitive overhead. Furthermore, their inability to model request causality in concurrent environments frequently produces distorted traces, particularly under asynchronous execution patterns. To address these limitations, we propose DeepTrace, a non-intrusive distributed tracing framework for microservices that leverages method-level delay distribution estimation to optimize trace reconstruction. DeepTrace employs a hybrid architecture combining kernel and user-space processing: low-complexity protocol parsing is offloaded to the kernel via eBPF to minimize overhead, while user-space components handle computationally intensive protocol analysis. The framework integrates cross-correlation method delay distributions with method relationship constraints, enabling precise causal inference for trace reconstruction. This approach ensures both efficiency in data collection and accuracy in modeling request causality across concurrent execution paths. Our testbed results indicate that, compared to state-of-the-art frameworks (e.g., DeepFlow, TreaceWeaver), DeepTrace reduces collection overhead by 89% with a 51% accuracy improvement.