Using Distributed Tracing to Identify Inefficient Resources Composition in Cloud Applications

Abstract

Cloud-Applications are the new industry standard way of designing Web-Applications. With Cloud Computing, Applications are usually designed as microservices, and developers can take advantage of thousands of such existing microservices, involving several hundred of cross-component communications on different physical resources.Microservices orchestration (as Kubernetes) is an automatic process, which manages each component lifecycle, and notably their allocation on the different resources of the cloud infrastructure. Whereas such automatic cloud technologies ease development and deployment, they nevertheless obscure debugging and performance analysis. In order to gain insight on the composition of services, distributed tracing recently emerged as a way to get the decomposition of the activity of each component within a cloud infrastructure. This paper aims at providing methodologies and tools (leveraging state-of-the-art tracing) for getting a wider view of application behaviours, especially focusing on application performance assessment.In this paper, we focus on using distributed traces and allocation information from microservices to model their dependencies as a hierarchical property graph. By applying graph rewriting operations, we managed to project and filter communications observed between microservices at higher abstraction layers like the machine nodes, the zones or regions. Finally, in this paper we propose an implementation of the model running on a microservices shopping application deployed on a Zonal Kubernetes cluster monitored by OpenTelemetry traces. We propose using the flow hierarchy metric on the graph model to pinpoint cycles that reveal inefficient resource composition inducing possible performance issues and economic waste.

Data Fusion of Observability Signals for Assisting Orchestration of Distributed Applications

“…In [18], Pythia is presented as an automated framework that suggests the activation of instrumentation tools to manage a newly-observed performance problem. In [19], the authors propose a solution for correlating information that is present in multiple traces, considering all traces as a single graph and decomposing tracing data into multiple vertices and edges. In [20], OpenTelemetry traces are processed to identify bottlenecks in the performance of a distributed application that is deployed across the cloud continuum.…”

A distributed tracing pipeline for improving locality awareness of microservices applications

“…‡ The collected traces can be processed to obtain a graph model for representing the relationships and dependencies between the constituent microservices of an application and to identify potential areas of complexity or risk, for example, tightly coupled microservices or potential single points of failure. Up to now, distributed tracing has been applied to microservices-based systems for different goals: visualization, 7 performance analysis, 8,9 fault diagnosis, 10 anomaly or smell detection. [11][12][13] In this paper, we exploit distributed tracing to define a big data pipeline to dynamically collect tracing data from running applications that are used to identify a given number k of microservices groups that allow for keeping low the response times of the most critical operations under a defined workload, when they are deployed onto different virtual or physical machines, for example, during DevOps stages.…”

“…The collected traces can be processed to obtain a graph model for representing the relationships and dependencies between the constituent microservices of an application and to identify potential areas of complexity or risk, for example, tightly coupled microservices or potential single points of failure. Up to now, distributed tracing has been applied to microservices‐based systems for different goals: visualization , 7 performance analysis , 8,9 fault diagnosis , 10 anomaly or smell detection 11–13 …”