CeresDB 1.0 Release: Cloud‑Native Time‑Series Database Design, Features, and Performance Evaluation

After nearly a year of open‑source development, Ant Group officially released CeresDB 1.0, a production‑ready cloud‑native time‑series database. The source code is available at GitHub and documentation at docs.ceresdb.io .

Design Goals

The CeresDB team, with five years of experience in time‑series data, identified challenges such as high‑cardinality tag handling, modern distributed solutions, and compute‑storage separation. The project aims to be a next‑generation cloud‑native time‑series database that efficiently supports both traditional timeseries and analytic workloads across single‑node and large‑scale cluster deployments.

Core Features

Storage Engine : Column‑mixed storage with an efficient XOR filter.

Cloud‑Native Distributed Architecture : Compute‑storage separation (OSS for data storage, WAL supporting OBKV and Kafka) and HASH‑based partition tables.

Deployment & Operations : Supports both standalone and distributed cluster deployments, with Prometheus + Grafana self‑monitoring.

Read/Write Protocol : SQL query and write support, built‑in high‑performance protocol with multi‑language SDKs, and Prometheus remote‑storage compatibility.

Multi‑Language SDKs : Client SDKs for Java, Python, Go, and Rust.

Technical Solutions

Storage Engine Exploration : Traditional inverted‑index models suffer from high‑cardinality tag explosion. CeresDB adopts column‑mixed storage, partition scanning, pruning, and efficient filters to mitigate this.

Data is organized using columnar storage and pruning techniques, as illustrated in the following diagram.

Distributed Solution : CeresDB separates compute from storage, enabling elastic scaling, high availability, and load balancing. The cluster consists of CeresMeta (metadata and scheduling), CeresDB instances (data storage), external WAL services, and object storage for SST files.

Performance Optimizations & Experiments

CeresDB combines column‑mixed storage, partition pruning, and efficient scanning to maintain high write and query performance under high‑cardinality workloads.

Write Optimizations : A LSM‑style write path avoids complex inverted indexes, delivering superior ingestion speed.

Query Optimizations : Includes min/max pruning, XOR filters for Parquet row‑group selection, concurrent scanning across multiple SST files, parallel row‑group fetching, IO merging for OSS objects, and local caching.

Benchmark Results

Using TSBS with 10 tags, 10 fields, and 1 million time series, CeresDB achieved over 1.5× the write throughput of InfluxDB 1.8.5, reaching up to 2× in steady state.

For low‑selectivity queries (e.g., os=Ubuntu15.10), CeresDB was 26× faster than InfluxDB (15 s vs 6 m 43 s). For high‑selectivity queries, InfluxDB was faster, with CeresDB being about 5× slower after warm‑up.

2023 Roadmap

Enhance performance, storage formats, and indexing.

Advance distributed features such as automatic load balancing, higher availability, and reliability.

Broaden ecosystem integration: support PromQL, InfluxQL, OpenTSDB protocols, Kubernetes deployment, monitoring tools, and data import/export utilities.

Community Invitation

The CeresDB team welcomes contributions, questions, and discussions. Contact via email [email protected] or join the DingTalk group (QR code shown below).