Pixie's magical developer experience is enabled by the Pixie Platform, an edge machine intelligence system designed for secure and scalable auto-telemetry.
The platforms key primitives are:
The system-level design is shown below:
The connection mode between the Vizier Module and Control Cloud is dependent on how Pixie is deployed. To configure Pixie's data transfer mode, see the instructions here.
In this scheme, the browser directly proxies into the Pixie Vizier Module and no customer data is transferred to Pixie's Control Cloud. Communication to Pixie's Control Cloud is limited to account and Kubernetes control data.
In this scheme, data flows through the Control Cloud via a reverse proxy as encrypted traffic without any persistence. This allows users to access data without being in the same VPC/network and avoids connectivity issues between the browser and the cluster. This is set as the default scheme in Pixie Community.
The Pixie Platform collects data with less than 5% CPU overhead and latency degradation. As shown here, the effective overhead attains steady state ~2% in environments running any substantial workloads. This is dramatically more efficient than legacy monitoring systems.
Pixie Platform's distributed architecture allows deployment spanning multiple clusters, clouds and deployment platforms.
As shown in the architecture, this is achieved by deploying PEM's in Linux nodes in both K8s or non-K8s clusters which are connected to Pixie Vizier Modules.
Note: Support for central Pixie Vizier Module and PEM deployments in non-K8s linux nodes have not yet been launched