Overview

Unlike traditional streaming processors such as Flink or ksqlDB, Timeplus allows you to store data natively in the platform, enabling multiple scenarios:

1. Ingest raw data into Timeplus for reliable processing or reprocessing.
2. Store intermediate results generated by Materialized Views or Tasks, and perform cascading queries on these results.
3. Store final processed results that are ready to be served directly to applications.

This means Timeplus can function as a standalone, self-contained platform to manage the entire data lifecycle — from ingest to process to serve.

Alternatively, it can integrate seamlessly with your existing infrastructure (e.g., upstream Kafka, Pulsar, PostgreSQL; downstream ClickHouse, S3, etc.).

This flexibility gives you full control to optimize for performance, cost-efficiency, and latency based on your use case.

Storage Engines

There is no single storage engine that performs best across all scenarios. To address this, Timeplus provides several built-in storage engines, each optimized for different use cases. The architecture is extensible, making it possible to add new engines as needed.

At a high level, Timeplus supports three categories of storage engines:

1. Columnar Store

   • Different Flavors: Changelog KV, Versioned KV, and Append Stream.
   • Optimized for analytical queries and high-throughput streaming persistence.

2. Row Store

   • Implementation: Mutable Stream.
   • Optimized for high concurrent mutable, row-oriented workloads.

3. Dictionary

   • A special key/value storage engine.
   • Acts as a cached lookup table, highly effective when combined with joins or remote lookups.

Each engine involves trade-offs in terms of latency, throughput, mutability, and query patterns. You can refer to their dedicated documentation for deeper details.

Data Serving

Once data is persisted in Timeplus, it can be served in two modes:

• Streaming queries for real-time, continuous insights.
• Historical queries for ad-hoc or backfill analysis.

Applications can access the data via multiple interfaces:

• SDKs / Drivers (language-specific)
• WebSocket API
• Timeplus Native Client
• Timeplus Web Console

Data delivery is flexible: it can be pushed to clients (pub/sub style) or pulled on demand by applications. The choice depends on your integration and latency requirements.