Timeline for production readiness

[wstrausser]

The FAQ on ducklake.select is currently still stating that Ducklake is not ready for production use but that the hope is for it to mature over the course of 2025.

With 2025 coming to a close, I'm curious what the current timeline is looking like for Ducklake to be suitable for production workloads. I ask because my team currently has about 6-8 months to migrate our data systems from on-premises workloads into the cloud and Ducklake is well-suited for our needs. However, given the tight timeline my only question at this stage is if Ducklake will be in a stable state by this point. If not, we may need to go with Iceberg instead.

[guillesd]

Hi, we are aiming to release 1.0 early 2026 (march/april). We will update our release page once we have more clarity on this!

[LoicRaillon]

Do you have a list of features / fixes that you want to include in the first stable release ?

[guillesd]

https://ducklake.select/roadmap this is planned for Feb

[djouallah]

Just a clarification question: is there no commitment at this stage to always keep the data layer (Parquet files and deletion files) compatible with Delta/Iceberg?

From my understanding, having multiple snapshots within the same Parquet files would currently break that compatibility. I just want to make sure I’m interpreting this correctly.

I’m asking because I genuinely want DuckLake to succeed, and given how much effort Delta and Iceberg are investing in converging toward a more unified data layer ( not metadata, just data), it feels like adoption could be challenging if the data layer itself is not standardized or interoperable.

[guillesd]

We do aim for full compatibility. If this is not the case we can look into it. Snapshots cannot be "within" a parquet file. If what you mean is multiple snapshots can reference the same file, then this also happens in Iceberg. What should happen in compaction (unsure if you are referencing to a potential bug) is that we create a new snapshot that points to new compacted files. I don't think we do anything that is not Iceberg-esk. But if you have a reproducer where you think the behaviour defers please share!

[djouallah]

good to know, but i am still confused, ducklake has the concept of partial file, https://ducklake.select/docs/stable/specification/tables/ducklake_data_file , iceberg smallest unit is a full parquet file ? you can not have an iceberg snapshot that refer to a partial file ?