Multi-Node Cluster Guide

ZeptoDB multi-node cluster configuration and operations guide.

---

Architecture

┌─────────────────────┐       TCP RPC        ┌──────────────────┐
│  Coordinator (HTTP)  │◄───────────────────►│  Data Node 1     │
│  zepto_http_server   │                      │  zepto_data_node │
│  port 8123           │       TCP RPC        │  port 9001       │
│  Web UI + API        │◄───────────────────►├──────────────────┤
│  node_id=0           │                      │  Data Node 2     │
│                      │                      │  zepto_data_node │
│                      │                      │  port 9002       │
└─────────────────────┘                      └──────────────────┘

• Coordinator (zepto_http_server): Provides HTTP API + Web UI, distributes queries via scatter-gather
• Data Node (zepto_data_node): Executes queries via TCP RPC, stores data

---

Binaries

zepto_http_server (Coordinator)

./zepto_http_server [options]

Option	Default	Description
--port, -p	8123	HTTP listening port
--ticks, -n	10000	Number of sample ticks to generate at startup
--no-auth	false	Disable authentication
--node-id	0	Cluster ID for this node
--add-node id:host:port	—	Register a remote data node (can be used multiple times)
--log-level	info	Log level (debug/info/warn/error)

The Coordinator is always active. If there are no data nodes, it operates in standalone mode. When nodes are added via --add-node or POST /admin/nodes, it automatically switches to cluster mode.

zepto_data_node (Data Node)

./zepto_data_node <port> [num_ticks] [options]

Argument	Default	Description
port	(required)	TCP RPC listening port
num_ticks	0	Number of sample ticks to generate at startup
--node-id	port number	Node ID within the cluster
--symbol	1	symbol_id for sample data
--coordinator host:port	—	Auto-register with the Coordinator
--api-key	—	Admin API key to use for registration
--advertise-host	localhost	Host address to advertise to the Coordinator

---

Quick Start

1. Build

cd build
ninja zepto_http_server zepto_data_node

2. Method A: Start Coordinator first → Data Nodes auto-register

# Terminal 1 — Coordinator
./zepto_http_server --port 8123
# The admin API key will be printed → copy it

# Terminal 2 — Data Node (auto-register)
./zepto_data_node 9001 --coordinator localhost:8123 --api-key $ADMIN_KEY

# Terminal 3 — Data Node (auto-register)
./zepto_data_node 9002 --coordinator localhost:8123 --api-key $ADMIN_KEY

3. Method B: Specify all nodes at startup

# Terminals 1~2 — Start Data Nodes first
./zepto_data_node 9001
./zepto_data_node 9002

# Terminal 3 — Coordinator (connect to nodes)
./zepto_http_server --port 8123 \
  --add-node 9001:localhost:9001 \
  --add-node 9002:localhost:9002

4. Verify via Web UI

Open http://localhost:3000/cluster in a browser → 3 nodes should be displayed (coordinator + 2 data nodes)

---

Runtime Node Management

Even after the cluster has started, you can dynamically add/remove nodes via the REST API.

Add a node

# Start a new data node
./zepto_data_node 9003

# Register with the Coordinator
curl -X POST http://localhost:8123/admin/nodes \
  -H "Authorization: Bearer $ADMIN_KEY" \
  -d '{"id":9003,"host":"localhost","port":9003}'

Remove a node

curl -X DELETE http://localhost:8123/admin/nodes/9003 \
  -H "Authorization: Bearer $ADMIN_KEY"

List nodes

curl http://localhost:8123/admin/nodes \
  -H "Authorization: Bearer $ADMIN_KEY"

Example response:

{
  "nodes": [
    {"id": 0, "host": "localhost", "port": 8123, "state": "ACTIVE", "ticks_ingested": 10000, "ticks_stored": 10000, "queries_executed": 5},
    {"id": 9001, "host": "localhost", "port": 9001, "state": "ACTIVE", "ticks_ingested": 0, "ticks_stored": 0, "queries_executed": 0},
    {"id": 9002, "host": "localhost", "port": 9002, "state": "ACTIVE", "ticks_ingested": 0, "ticks_stored": 0, "queries_executed": 0}
  ]
}

Check cluster status

curl http://localhost:8123/admin/cluster \
  -H "Authorization: Bearer $ADMIN_KEY"

Example response:

{
  "mode": "cluster",
  "node_count": 3,
  "ticks_ingested": 10000,
  "ticks_stored": 10000,
  "queries_executed": 5
}

---

Examples

Single Node (Standalone)

./zepto_http_server --port 8123

The cluster tab shows 1 node. mode: standalone. When a data node is added, it automatically switches to mode: cluster.

2-Node Cluster

# Terminal 1
./zepto_data_node 9001

# Terminal 2
./zepto_http_server --port 8123 --add-node 9001:localhost:9001

4-Node Cluster

# Terminals 1~3: data nodes
./zepto_data_node 9001
./zepto_data_node 9002
./zepto_data_node 9003

# Terminal 4: coordinator
./zepto_http_server --port 8123 \
  --add-node 9001:localhost:9001 \
  --add-node 9002:localhost:9002 \
  --add-node 9003:localhost:9003

Remote Host

# Coordinator (10.0.1.1)
./zepto_http_server --port 8123

# Machine A (10.0.1.2) — auto-register
./zepto_data_node 9001 \
  --coordinator 10.0.1.1:8123 \
  --api-key $ADMIN_KEY \
  --advertise-host 10.0.1.2

# Machine B (10.0.1.3) — auto-register
./zepto_data_node 9001 \
  --coordinator 10.0.1.1:8123 \
  --api-key $ADMIN_KEY \
  --advertise-host 10.0.1.3

---

Web UI Cluster Tab

Information available on the /cluster page:

Section	Description
Summary Cards	Mode (standalone/cluster), number of active nodes, total ticks, total queries
Node Status Table	Per-node ID, Host, Port, State, Ticks, Queries
Ticks Stored Bar Chart	Bar chart of ticks stored per node
Ingestion History	Per-node time-series ingestion trend
Queries History	Per-node time-series query execution trend
Latency History	Per-node ingest latency trend

Node state colors:

• 🟢 ACTIVE — Normal
• 🟡 SUSPECT — Response delayed
• 🔴 DEAD — Unreachable
• 🔵 JOINING — Joining the cluster
• ⚪ LEAVING — Leaving the cluster

---

High Availability (HA)

Active-Standby architecture with automatic failover when the coordinator fails.

Coordinator (ACTIVE)  ←── ping (500ms) ──  Coordinator (STANDBY)
  port 8123, rpc 9100                        port 8124, rpc 9101
     │                                            │
     ├── Data Node 1 (9001)                       │
     └── Data Node 2 (9002)                       │
                                                  │
  If ACTIVE dies → after 2 seconds STANDBY is promoted to ACTIVE
                   Node list is automatically re-registered

HA Options

Option	Description
--ha active|standby	HA role
--peer host:port	RPC address of the peer coordinator
--rpc-port PORT	RPC port for this node (default: HTTP port + 1000)

Starting HA

# Terminal 1 — Active Coordinator
./zepto_http_server --port 8123 --ha active --peer localhost:9101 --rpc-port 9100

# Terminal 2 — Standby Coordinator
./zepto_http_server --port 8124 --ha standby --peer localhost:9100 --rpc-port 9101

# Terminals 3~4 — Data Nodes (register with Active)
./zepto_data_node 9001 --coordinator localhost:8123 --api-key $ADMIN_KEY
./zepto_data_node 9002 --coordinator localhost:8123 --api-key $ADMIN_KEY

Failover Behavior

1. Standby pings Active every 500ms
2. If no response for 2 seconds, Standby is automatically promoted to Active
3. The registered data node list is automatically re-registered with the new Active
4. Data node processes are unaffected (independent processes)
5. Clients connect to the new Active’s HTTP port

Client Switchover After Failover

When Active dies, Standby (8124) becomes the new Active:

# Before: http://localhost:8123
# After:  http://localhost:8124

In production, place a load balancer (ALB/NLB) in front and use health checks for automatic switchover.

---

Troubleshooting

Symptom	Cause	Solution
Not in cluster mode	Running an older version of the server	Update to the latest build (coordinator is always active)
Node shows as DEAD	Data node is down or port mismatch	Check the data node process and verify the port
Web UI shows only 1 node	Data node is not registered	Add via --add-node or POST /admin/nodes
Connection refused	Data node has not started yet	Start data nodes first, then start the coordinator