vira28

vira28

vor 27 Tagendiscuss

Building https://github.com/viggy28/streambed - a full Postgres running on S3 using DuckDB as the query engine.

vira28

vor 28 Tagendiscuss

Certainly interested.

vira28

letzten Monatdiscuss

[dead]

vira28

letzten Monatdiscuss

Agree. Appreciate the reminder.

vira28

letzten Monatdiscuss

Tomas Vondra, a major Postgres contributor recently revived a thread on using Bloom filters - https://www.postgresql.org/message-id/flat/5cd8c20c-14b5-4b0...

So there is more core work happening on support OLAP but I do think it will take some time.

In the meantime, I think we have all the pieces (storage, query engine, table format) to set up a true OLAP. For instance, I created https://github.com/viggy28/streambed to pressure test this idea.

vira28

letzten Monatdiscuss

The idea is pretty similar. As per their README, Iceberg support is deprecated.

vira28

letzten Monatdiscuss

From what I understand Ducklake needs a dedicated metadata database and it also ties to DuckDB land wherease with Iceberg many engines can query directly.

vira28

letzten Monatdiscuss

The challenge with any CDC is making it reliable. Curious, how are you exporting to S3? - Debezium or some service in AWS or home grown tool?

vira28

letzten Monatdiscuss

Currently, Strembed expects REPLICA IDENTITY FULL for getting the before and after value of TOAST column. Since we have the data in object storage, we could populate it without the need for REPLICA IDENTITY FULL. Created an issue https://github.com/viggy28/streambed/issues/25 to track this feature.

vira28

letzten Monatdiscuss

Welcome. To avoid primary running out of disk space, you can configure max_slot_wal_keep_size https://www.postgresql.org/docs/17/runtime-config-replicatio...

Since Supabase is vanilla Postgres, streambed should work with replica as the source.

reg, Planetscale, I haven't looked at their offerings yet.

Where do you host your DB currently? Happy to try out with that provider as the source.

vira28

letzten Monatdiscuss

Aside from the cost, my major motivation is to keep the infrastructure simple. The data is already there in Postgres, so I didn't want to add another data warehouse. I have also shared my thoughts on where this is heading https://viggy28.dev/article/postgres-gateway-drug/

vira28

letzten Monatdiscuss

Thanks for the kind words!

Short answer: yes, column-level schema changes sync to Iceberg automatically[0].

Logical replication (pgoutput in v1) doesn't actually stream DDL statements. Instead, Postgres emits a fresh Relation message describing the table's current column layout right before the next change to that table. So we diff that against the last layout we knew and infer what changed.

From there we evolve the Iceberg schema in place: flush any buffered rows under the old schema first, then write a new metadata version with the change. What's handled today:

  - ADD COLUMN — new field ID allocated; the column's Postgres DEFAULT is carried into Iceberg's initial-default/write-default, so existing rows read back correctly
  - DROP COLUMN — removed from the current schema, existing data files untouched
  - Type widening — int4→int8, float4→float8 (the changes Iceberg considers compatible)
  - REPLICA IDENTITY changes

[0] https://github.com/viggy28/streambed/pull/21

vira28

letzten Monatdiscuss

[flagged]

vira28

letzten Monatdiscuss

Welcome. Would love to hear your experience. Feel free to share here or in the repo. Fully open source.

vira28

letzten Monatdiscuss

Should be fairly doable using binlog-based producer https://github.com/go-mysql-org/go-mysql.

vira28

letzten Monatdiscuss

Both projects are relevant. Curious, what kinda pushdown capabilities that you were looking for?

vira28

letzten Monatdiscuss

Hello, I checked ingestr repo, and it is in my bookmark. Small world.

Agree, CDC is like Death by a thousand cuts. I believe Debezium has a Java library.

My initial need was Postgres compatibilty. Wanted to give an endpoint that BI and dashboard teams can use to query as if they are querying a Postgres replica. Added more context here https://news.ycombinator.com/item?id=48350820

vira28

letzten Monatdiscuss

[flagged]

vira28

letzten Monatdiscuss

Author here. For context, I was the tech lead for the Postgres team at Cloudflare, and this came directly out of a challenge I kept hitting there: BI and dashboard teams needed to run long-running analytical queries, and the answer was always to spin up another bespoke read replica or stand up an ETL dump into an analytical database and query that.

So the question I started with was: what's the fewest components I could get away with? That led to the architecture here — Streambed connects to Postgres as a logical replication subscriber (same mechanism as a read replica) and streams WAL changes straight into Apache Iceberg on S3, queryable from psql via an embedded DuckDB. There are a lot of edge cases to handle, and it's very much early days.

Welcome any feedback.

vira28

vor 2 Monatendiscuss

If I have to guess on why they are providing voluntary exit option, because they have a lot more folks from Europe.

Submissions

Ask HN: Do you use Apache Iceberg in Production?

Ask HN: Want to build something open source on nights and weekends together?

Show HN: Streambed – Stream Postgres to Iceberg on S3, Supports Postgres Wire

Show HN: pg_savior: a seatbelt for Postgres – blocks accidental DELETE/UPDATE

I asked my local LLM to add 23 numbers and got seven wrong answers

Ask HN: Is giving AI agents DB access the new BI-tool problem?

Postgres Is the Gateway Drug

Show HN: Prompt template for solo and small law firms

Show HN: AI that replaces the first 15 minutes of every client call

Tell HN: Even LinkedIn is running on Cloudflare and not Azure

Show HN: We cut RAG latency ~2× by switching embedding model

Root Ventures

comments