HI version is available. Content is displayed in original English for accuracy.
The pitch: keep Databricks or Snowflake. Bring Rocky for the DAG. Rocky is a Rust-based control plane for warehouse pipelines. Storage and compute stay with your warehouse. Rocky owns the graph — dependencies, compile-time types, drift, incremental logic, cost, lineage, governance. The things your current stack can't give you because it doesn't own the DAG.
A few things I think are interesting:
- Branches + replay. `rocky branch create stg` gives you a logical copy of a pipeline's tables (schema-prefix today; native Delta SHALLOW CLONE and Snowflake zero-copy are next). `rocky replay <run_id>` reconstructs which SQL ran against which inputs. Git-grade workflow on a warehouse.
- Column-level lineage from the compiler, not a post-hoc graph crawl. The type checker traces columns through joins, CTEs, and windows. VS Code surfaces it inline via LSP.
- Governance as a first-class surface. Column classification tags plus per-env masking policies, applied to the warehouse via Unity Catalog (Databricks) or masking policies (Snowflake). 8-field audit trail on every run. `rocky compliance` rollup that CI can gate on. Role-graph reconciliation via SCIM + per-catalog GRANT. Retention policies with a warehouse-side drift probe.
- Cost attribution. Every run produces per-model cost (bytes, duration). `[budget]` blocks in `rocky.toml`; breaches fire a `budget_breach` hook event.
- Compile-time portability + blast radius. Dialect-divergence lint across Databricks / Snowflake / BigQuery / DuckDB (12 constructs). `SELECT *` downstream-impact lint.
- Schema-grounded AI. Generated SQL goes through the compiler — AI suggestions type-check before they can land.
What Rocky isn't:
- Not a warehouse — it's the control plane on top.
- Not a Fivetran replacement. `rocky load` handles files (CSV/Parquet/JSONL); for SaaS sources use Fivetran, Airbyte, or warehouse-native CDC.
- Not dbt Cloud — no hosted UI, no managed scheduler. First-class Dagster integration if you need orchestration.
Adapters: Databricks (GA), Snowflake (Beta), BigQuery (Beta), DuckDB (local dev / playground). Apache 2.0.
I'd love feedback on the trust-system framing, the governance surface (particularly classification-to-masking resolution in `rocky compile` and the `rocky compliance` CI gate), the branches/replay design, the cost-attribution primitives, or anything else that catches your eye. Happy to go deep in the thread.
Discussion (20 Comments)Read Original on HackerNews
Having the compiler know “this column flows into these downstream models” before execution changes the workflow quite a bit. It makes refactors and masking policies much less scary.
Do you expose any kind of “lineage diff” between branches? For example: this PR changes the downstream impact of `customer.email` from A/B/C to A/B/D. That would be useful in code review.
The code is what it is. `cargo test --workspace` runs across 19 crates. CI on 5 platforms (macOS ARM/Intel, Linux x86/ARM, Windows). JSON output schemas are codegen-checked in CI so docs can't drift from the binary.
If you want to skip the marketing copy and look at engine reasoning instead: PR #240 (audit trail), #241 (column classification + masking), #270 (failed-source surfacing in discover).
I'd rather hear "the code is bad" than "the post sounds AI-written".
Of course you would. Reading through and judging the quality of AI output is the largest amount of effort in a world where you can get everything else by prompting. Please internalize this: If you want to be respected you will have to put in effort yourself. There is no way around this.
'A-Lot' of side projects, hobby projects, etc.. are all using AI tools now. Also for marketing, every sales/marketing firm is using AI. So why critisize this guy inparticular.
AI is pervasive, the train has left the station. So that is not a reason to criticize this project. There might be other reasons, I'm not sure, but not that an AI was used.
I do think the author is doing a disservice to themselves by writing the post and comments using LLM, even if the code is mostly agent built. People can tell right away, all the LLM shibboleths are there... it feels cheap. Just write naturally and then Google translate, don't let the LLM speak on your behalf.
What's going to distinguish projects that are built this way is the ability to explain, document, support, and maintain said projects over the long term. That will be the crucible. Gone are the days of "build it and they will come", and I feel a bit sad about that.
It's so easy to let the code grow under you beyond what you have the capacity to do the above for.
I've got the same thing going on. Eschewing paid work and grinding 16, 17 hours a day boiling the sea to build the whole universe from scratch (also a database, but of a different sort than this project) integrating all my favourite DB research papers and ideas that I've accumulated over the last 30 years. Outperforms postgres 2-4x or more, has a battery of correctness tests, Lean proofs, benchmarks, etc. etc.
But frankly I'd be nervous to share. Especially here. I don't even know where it ends up. Not least because if I'm doing it, so are 50 other people, probably.
Workflows knows task A runs before task B. Rocky knows `dim_customer.email` flows from `raw_users.email_address` through three CTEs in `stg_customers`. Different layer, same word.
I'll be more careful with that framing.
- dbt-style semantic-layer versions (v1/v2 of a model) - schema migration history - branch-based (Rocky already has branches + replay)
Different design choice for each, so it helps to know which problem you're trying to solve.
ClickHouse is tractable through the Adapter SDK without engine patching. If you can share roughly your model count and workload shape, I can put a real timeline on it. Open to community PRs too.
https://news.ycombinator.com/item?id=47340079
Not saying yours are, but them -- dashes certainly looks like it ;)
On the schema-grounded AI angle: agreed. The failure mode you describe — structurally valid SQL that joins on the wrong key or aggregates at the wrong grain because the model hallucinated a relationship — is exactly what the compiler is positioned to catch. AI-generated SQL runs through the type checker before it can land, so suggestions that don't validate against the actual DAG never reach the user. NL-to-SQL tools that integrate a compile step would close exactly the gap you're pointing at.
On your two questions:
1. Branch isolation for stateful models — mixed answer, and worth being honest about:
2. Cost attribution. Both bytes scanned and duration are captured per-model in the run record (`bytes_scanned` and duration on `RunRecord`). Budget gating today is on cost (USD) and duration — `max_usd` and `max_duration_ms` in `[budget]` blocks in `rocky.toml`, as independent thresholds. A direct bytes-scanned budget threshold isn't gateable today; the bytes are in the run record for analysis but you can't currently fail CI on "this run scanned more than N TB". Reasonable extension if there's demand.