A simple, high-performance timeseries database for IoT and edge computing. HTTP + TCP, sub-millisecond reads, 233× faster than InfluxDB on writes, Gorilla compression.

Traditional databases write to WAL, then copy to disk blocks, then update indexes — triple the IO. GTSDB takes a different approach:

Traditional:   WAL → Memory → Disk Blocks → Index
GTSDB:         WAL → Index (if needed) ─── read directly from WAL

• WAL is the database. No separate block storage. No double-write penalty.
• Indexes are optional. Created on-demand for fast time-range queries.
• Result: Less IO, less memory, simpler code, crazy fast.

10 sensors × 1,000 points each. GTSDB over TCP, InfluxDB over HTTP. See benchmark repo.

# 1. Run the server
go run .

# 2. The server starts on TCP :5555 and HTTP :5556.
#    Check the logs for the auto-generated root token:
#    "Created default root user with token: abc123..."

# 3. Write data
curl -X POST http://localhost:5556/ \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <token>" \
  -d '{"operation":"write","key":"sensor1","write":{"value":42.5}}'

# 4. Read latest 10 records
curl -X POST http://localhost:5556/ \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <token>" \
  -d '{"operation":"read","key":"sensor1","read":{"lastx":10}}'

All operations via POST / with Authorization: Bearer <token> header.

Write a data point:

{
    "operation": "write",
    "key": "sensor1",
    "write": { "value": 42.5 }
}

Read latest data:

{
    "operation": "read",
    "key": "sensor1",
    "read": { "lastx": 10 }
}

Read with time range & downsampling:

{
    "operation": "read",
    "key": "sensor1",
    "read": {
        "start_timestamp": 1717965210,
        "end_timestamp": 1717965211,
        "downsampling": 3
    }
}

Batch write (up to 10,000 points):

{
    "operation": "batch-write",
    "points": [
        { "key": "sensor1", "value": 42.5, "timestamp": 1717965210 },
        { "key": "sensor2", "value": 99.9, "timestamp": 1717965210 }
    ]
}

Export as CSV:

{
    "operation": "export",
    "key": "sensor1",
    "export": { "format": "csv", "lastx": 100 }
}

Health & Monitoring (no auth required):

• GET /health — JSON health status
• GET /metrics — Prometheus metrics

JSON-line protocol. Same operations as HTTP. See TCP Protocol.

graph TD
    TC[TCP Client<br/>port 5555] --> TS[TCP Server<br/>goroutine]
    HC[HTTP Client<br/>port 5556] --> HS[HTTP Server<br/>goroutine]
    
    TS --> HL[Handler Layer]
    HS --> HL
    HL --> FO[Fanout Pub/Sub<br/>SSE push]
    
    HL --> BL[Buffer Layer]
    BL --> WAL[.aof<br/>Write-Ahead Log]
    BL --> IDX[.idx<br/>Index]
    BL --> GOR[.aof.gor<br/>Gorilla Compressed]
    
    WAL --> FS[File System]
    IDX --> FS
    GOR --> FS

Default config: gtsdb.ini. Override with command line argument:

./gtsdb myconfig.ini

Example gtsdb.ini:

[listens]
tcp = :5555
http = :5556

[paths]
data_directory = data

[buffer]
buffer_size = 700
compaction_compression = true

See Configuration Reference for all options.

go test -benchmem -run=^$ -bench ^BenchmarkMain$ -benchtime=5s

cpu: 13th Gen Intel(R) Core(TM) i7-13700KF
BenchmarkMain-24    26396    135241 ns/op    4249 B/op    5 allocs/op

This benchmark runs 50% read and 50% write operations to 100 different keys over a single TCP connection.

make Benchmark

See gtsdb-benchmark for GTSDB vs InfluxDB comparison benchmarks.

# Build & run
go run .
go build .

# Tests
go test ./...
go test ./... -skip=TestMain -coverprofile=docs/coverage -p 1
make GenerateTest          # Coverage report (HTML + SVG badge)

# Code quality
golangci-lint run
make lint-fix

# Benchmarks
make Benchmark             # Concurrent data structures

MIT