필사 모드: Open Source Worth Watching Right Now (3) Infrastructure and Databases

• Introduction — licenses redrew this board
• Snapshot
• Analytics — asking questions without moving the data
• Extending Postgres
• Embedded and cache
• Platform and IaC
• Check before you adopt
• Links

Introduction — licenses redrew this board

The change that infrastructure and databases have gone through over the past few years came from licensing rather than from technology. Widely used projects moved to licenses that restrict commercial use, and each time the community answered with a fork.

That is why, in this field, checking the license comes before comparing performance in many cases. The list below is not a ranking but a map grouped by the problem each project solves.

Snapshot

All figures as of 2026-08-12.

Analytics — asking questions without moving the data

duckdb/duckdb is an analytical database that runs inside your process. Because there is no server, you can query Parquet files as they are from a Python notebook or a CI job. Aggregations at a scale that would once have gone onto a cluster now often finish on a laptop. On the other hand, it is not built to take over the seat of a shared warehouse that many users write to at the same time.

-- Example: aggregate a remote Parquet file directly, with no load step
SELECT country, count(*) AS n
FROM read_parquet('s3://bucket/events/*.parquet')
GROUP BY country
ORDER BY n DESC
LIMIT 10;

ClickHouse/ClickHouse, by contrast, is a server that takes large analytical queries around the clock. Columnar storage and vectorized execution let it aggregate big tables quickly. It is not a good fit for workloads with frequent updates and deletes, or for designs where multi-table joins are the core.

risingwavelabs/risingwave is a database that keeps the results of stream processing available like a table. It becomes a candidate when you want to define real-time aggregations in SQL. Because it is a structure that has to keep holding state, it is harder to operate than a batch pipeline.

Extending Postgres

pgvector/pgvector adds a vector type and similarity search indexes to Postgres. If your data already lives in Postgres, this is the option to try first before you bring in a separate vector database. GitHub cannot classify the license automatically, but opening the LICENSE file shows permissive license wording of the same family as Postgres itself.

cloudnative-pg/cloudnative-pg is an operator that runs Postgres clusters on Kubernetes. It handles failover and backup procedures declaratively. Note, though, that if you move here from a managed database, the operational responsibility your cloud provider used to carry on your behalf passes to your team.

electric-sql/pglite builds Postgres to WebAssembly and runs it inside a browser or Node. It is worth using for tests and local-first applications. The repository was created in February 2024, so it is a young project — check the scope and the limits of extension support first.

Embedded and cache

tursodatabase/libsql is a project that branched off from SQLite and added replication and a server mode. It fits designs that want to keep reads close by in edge environments.

valkey-io/valkey is the fork created after the Redis license change, and it keeps BSD 3-Clause. The repository was created in March 2024, but the codebase itself continues a long-proven lineage. If license terms are the reason you are looking for an alternative, this is the most direct candidate.

nats-io/nats-server takes care of messaging and streaming. Having started in 2012 and been operated for a long time, it becomes a candidate when you want to handle request and response together with publish and subscribe in a lighter setup than Kafka. If long-term retention and large-scale reprocessing are the center of your use case, it has a different grain.

Platform and IaC

siderolabs/talos is a Linux distribution designed exclusively for Kubernetes nodes. It removes SSH and the shell and makes the machine manageable only through an API, which reduces the attack surface. It is not a fit if you want to keep the familiar way of operating, where you go into the server and fix things there.

opentofu/opentofu is the IaC tool that split off after the Terraform license change, and it keeps MPL-2.0. When your existing configuration assets are fine and only the license is the problem, it becomes a practical option. That said, commercial add-on features and the details of the provider ecosystem keep diverging, so you have to check that the modules you use work on both sides.

Check before you adopt

Check the full license text yourself, and route commercial adoption through legal review. This post is not legal advice.

What matters in this area is that there is a real precedent for licenses changing. The fact that something carries a permissive license today is no guarantee that it will stay that way. When you choose a database, look at the cost of leaving as well. Whether you can export the data in a standard format, whether the protocol is compatible, and whether a migration path exists in documentation are what actually reduce license risk.

Repository details (stars, license, recent activity) were checked directly on GitHub on 2026-08-12 and are point-in-time values. The numbers and the status change.

Links

Series: Previous post — Build tools, editors, CLIs, and terminals · Next post — Observability and security

Related posts on this blog:

• A deep look inside DuckDB internals
• OpenTofu compared with Terraform
• The SQLite renaissance and embedded replicas
• PostgreSQL internals and pgvector

Tools: SQL Playground