Skip to main content

​
Architecture Overview

KOSTRA is a Kubernetes-native platform offered two ways: as a fully managed SaaS that KOSTRA operates for you, or self-hosted in your own environment when you need it. Both run the same software; only who operates the cluster differs. This page describes how the system is structured and how data moves through it.

​
Deployment options

  • Managed SaaS. KOSTRA runs and maintains the platform for you. You connect your cloud accounts and start using it — no cluster to operate, upgrades handled for you. This is the fastest way to get value.
  • Self-hosted. For customers with data-residency, compliance, or policy requirements, KOSTRA can run inside a Kubernetes cluster you control — including in your own cloud region. You operate the cluster; the platform is identical to the SaaS.
Both options are containerized microservices orchestrated by Kubernetes and installed with Helm charts. Ingress is handled by an NGINX ingress controller, and TLS is provisioned and auto-renewed through cert-manager and a public certificate authority.

​
Core services

KOSTRA is composed of focused services, each with a single responsibility:
  • rest_api — the primary API, backed by MySQL. It serves organizations, pools, resources, expenses, recommendations, allocation rules, and chargeback.
  • auth — authentication and authorization, issuing tokens used across the platform.
  • ngui — the KOSTRA web interface (React), which talks to a GraphQL server that in turn calls rest_api and auth.
  • diworker and diproxy — billing and resource import. These connect to each provider, pull billing and discovery data, and write it to the data stores.
  • bumiworker — the recommendation engine. It runs optimization modules on a schedule and records dated, evidence-backed findings.
  • metroculus — collection and storage of resource utilization metrics used for rightsizing.
  • insider — cloud pricing data used to quantify savings.
  • herald — notifications and email delivery for alerts, reports, and account events.
  • risp — reserved-instance and savings-plan analysis.
Additional workers handle scheduled tasks, integrations, and data export. Together they communicate synchronously over HTTP and asynchronously over a message queue.

​
Data stores

KOSTRA uses purpose-fit databases:
  • MySQL — transactional data for the API (organizations, users, resources, rules).
  • MongoDB — raw billing records and recommendation output.
  • ClickHouse — high-volume expense analytics for fast aggregation over time.
  • Redis — caching and coordination.
  • InfluxDB — time-series data used by supporting services.

​
How data flows

The path from a provider bill to a dashboard is straightforward:
  1. You connect a cloud account by supplying read-oriented credentials.
  2. diworker imports billing and discovers resources through the provider SDKs.
  3. Raw records land in MongoDB; normalized expenses are written to ClickHouse for analytics.
  4. bumiworker evaluates recommendation modules against resources, metrics, and expenses, and writes findings back.
  5. rest_api exposes the results, and ngui presents dashboards, recommendations, chargeback, and modernization views.
Provider utilization metrics are collected so that rightsizing decisions are grounded in real usage across every connected cloud.

​
Security posture

KOSTRA encrypts cloud credentials at rest, forces traffic over TLS, and governs access with role-based permissions. On the managed SaaS, KOSTRA operates this posture for you; when self-hosted, the same controls run inside your own infrastructure so data never leaves it. See Security & Data Protection for detail.

​
System requirements (self-hosted)

If you self-host, KOSTRA targets a single-node or small multi-node Kubernetes cluster for evaluation, scaling out for production estates. As a practical starting point, provision a cluster with several CPU cores, enough memory to run the databases comfortably, and durable storage for MySQL, MongoDB, and ClickHouse. Exact sizing depends on the number of accounts, resources, and history you retain; KOSTRA’s team can help you size for your estate. On the managed SaaS, sizing and scaling are handled for you.