Kubernetes Deployment Overview

Cardano Rosetta Java ships official Helm charts for Kubernetes deployment. The charts support both single-host K3s deployments and any managed Kubernetes cluster (EKS, GKE, AKS).

Architecture

The deployment consists of 4 Helm subcharts plus orchestration resources in the parent chart:

Parent chart: cardano-rosetta-java
├── cardano-node          StatefulSet — Cardano blockchain node + socat sidecar
├── postgresql            StatefulSet — PostgreSQL with blockchain-tuned configuration
├── yaci-indexer          Deployment  — Blockchain data indexer (Yaci Store)
└── rosetta-api           Deployment  — Rosetta HTTP API

Monitoring (Prometheus, Grafana) is not included — production clusters should bring their own monitoring solution.

Key Architectural Differences from Docker Compose

UNIX socket bridging via socat

Docker Compose shares the Cardano node's UNIX socket (/node/node.socket) via volume mounts. In Kubernetes, pods cannot share UNIX sockets. A socat sidecar inside the cardano-node pod forwards TCP connections on port 3002 to the UNIX socket. The yaci-indexer connects using the n2c-socat Spring profile.

Startup ordering via init containers

Docker Compose depends_on chains are replaced by Kubernetes init containers:

cardano-node (init: mithril-download → cardano-node starts)
    │
    ├─ postgresql starts immediately (no node dependency — it's just a database)
    │
    └─ yaci-indexer (wait-for-postgres: pg_isready
                     wait-for-node-tcp: nc cardano-node:3002
                     copy-node-config: copies configs from node image)
         rosetta-api (wait-for-postgres
                      wait-for-indexer: /actuator/health
                      copy-node-config: copies configs from node image)

The cardano-node pod runs three containers: the node itself, a socat sidecar (bridges the UNIX socket to TCP port 3002), and cardano-submit-api — so READY shows 3/3 when fully up.

Index creation is handled natively by yaci-indexer after sync reaches tip. Monitor progress at http://<yaci-indexer>:9095/actuator/rosetta-indexes.

Three sync stages

Stage	What's happening	Pods ready
SYNCING	yaci-indexer catching up to chain tip	All pods up, API responding
APPLYING_INDEXES	Indexer reached tip, DB indexes being built	All pods up, API responding
LIVE	Fully synced, all indexes valid	All pods ready, API fully functional

Hardware Profiles

Three built-in profiles scale resources across all components:

Profile	Total RAM	Total vCPU	Use case
entry	32 GB	4 cores	Preprod, single host, K3s
mid	48 GB	8 cores	Mainnet production
advanced	94 GB	16 cores	High-throughput production

See Helm Values Reference for per-component breakdown.

Prerequisites

Software

• Helm v3.14+ (helm version)
• kubectl configured for your cluster
• K3s v1.28+ or any managed Kubernetes cluster (EKS, GKE, AKS)

Access

• A DB_PASSWORD for PostgreSQL (min 16 characters recommended for production)
• Outbound internet access (Mithril snapshot download, ~20 GB for mainnet)

Hardware (minimum for preprod / entry profile)

• 8 vCPU
• 32 GB RAM
• 150 GB fast SSD (for node data + PostgreSQL)

Hardware (recommended for mainnet / mid profile)

• 16 vCPU
• 64 GB RAM
• 700 GB fast SSD (500 GB node + 200 GB PostgreSQL)

Next Steps

• Deployment Runbook — step-by-step deploy for preprod and mainnet
• Helm Values Reference — full configuration reference