Learning Data Stores

Relational, analytical, time-series, vector, graph, and embedded database experiments for learning.

Last verified: 2026-06-21

Development Environment

If DuckDB or the Python DuckDB module is missing locally, enter the Nix shell:

nix develop

Runnable Starter Project

Start with a Dockerless SQLite workload lab before moving the same workload to DuckDB, Kuzu, LanceDB, InfluxDB, or Neo4j:

python3 projects/sqlite-workload-lab/app.py
python3 projects/sqlite-workload-lab/test_app.py

Target Hands-On Projects

DuckDB analytics:

python3 projects/duckdb-analytics-lab/check_duckdb.py
uv run --with duckdb python projects/duckdb-analytics-lab/app.py

SQLite stays as the always-available baseline; DuckDB is the first named analytical database hands-on.

Time-series, vector, and graph concepts without Docker:

python3 projects/multimodel-db-lab/app.py
python3 projects/multimodel-db-lab/test_app.py

Use this before moving the same ideas to InfluxDB, LanceDB/Chroma, Kuzu, or Neo4j.

Why This Repo Exists

Application repos should not become database catalogs. This repo keeps database experiments small, comparable, and runnable.

The rule is:

• prefer no-Docker embedded examples first
• use local files and generated sample data
• add server databases only when their real behavior matters
• document when a lightweight substitute is acceptable and when it is misleading

What This Repo Teaches

This repo is for choosing and understanding data stores by workload.

Every example should identify:

• write pattern
• read/query pattern
• indexing strategy
• data volume assumption
• local development setup
• backup/export story
• where the database is a good fit
• where it becomes the wrong tool

Learning Path

1. SQL basics with H2 and SQLite-style embedded databases
2. Analytical SQL with DuckDB
3. Time-series modeling with DuckDB first, then InfluxDB 3 Core
4. Vector search with LanceDB and Chroma
5. Graph modeling with Kuzu first, then Neo4j
6. Search and hybrid retrieval
7. Choosing the right database for an app
8. Backup, migration, indexing, and operational tradeoffs

Planned Structure

examples/
  relational-h2/
  analytics-duckdb/
  timeseries-duckdb/
  timeseries-influxdb3/
  vector-lancedb/
  vector-chroma/
  graph-kuzu/
  graph-neo4j/
  search-sqlite-fts/
docs/
  2026-learning-items.md
  database-selection.md
  dockerless-strategy.md
  repository-profile.md
  timeseries-vector-graph-notes.md

Dockerless First

Good default experiments:

• H2 for embedded relational Java tests
• DuckDB for local analytical SQL and time-series-shaped data
• LanceDB for embedded vector search
• Chroma for local vector retrieval experiments
• Kuzu for embedded graph/Cypher experiments, with a maintenance warning

Optional server experiments:

• InfluxDB 3 Core for real time-series ingest/query behavior
• Neo4j for production-style graph database behavior, Cypher tooling, vector indexes, and graph data science

Study Loop

1. model the same small dataset in two stores
2. write equivalent ingest and query examples
3. measure readability before measuring speed
4. document what each store makes easy
5. document what each store hides or makes operationally expensive

What Belongs Elsewhere

• app-specific persistence code belongs in learning-backend-ddd
• retrieval application logic belongs in learning-ai-python
• MQTT/device ingestion examples can start in learning-embedded-iot and land here when the storage comparison is the point
• deployment and backup runbooks belong in learning-platform-engineering

First Milestones

1. Add DuckDB analytics and time-series-shaped examples with generated data.
2. Add H2 as a Java integration-test fixture.
3. Add LanceDB and Chroma examples with the same toy document set.
4. Add Kuzu first, then Neo4j, for graph comparison.
5. Add InfluxDB 3 Core only where server time-series behavior matters.

References

• DuckDB documentation: https://duckdb.org/docs/current/
• InfluxDB 3 Core documentation: https://docs.influxdata.com/influxdb3/core/
• LanceDB documentation: https://docs.lancedb.com/
• Chroma documentation: https://docs.trychroma.com/
• Neo4j documentation: https://neo4j.com/docs/
• Kuzu repository: https://github.com/kuzudb/kuzu
• H2 Database: https://www.h2database.com/html/main.html

License

0BSD. You can use, copy, modify, and distribute this project for almost any purpose.