Hi all,

I would like to start a discussion for FIP-50:
https://cwiki.apache.org/confluence/spaces/FLUSS/pages/446071500/FIP-50+Native+DataFusion+Query+Engine+for+the+Fluss+Rust+Client

Querying Fluss today means Flink or Spark, and both bring their own
cluster runtime, so FIP-50 proposes a new crate, fluss-datafusion,
that registers a Fluss cluster as a DataFusion catalog so that SELECT
* FROM fluss.db.t runs in the calling process. It is a library and  a
query gateway would sit above it and make it usable for a larger
subset of tools/clients.

Most of what the crate needs from the core client already exists in
the Java client, so the work is porting rather than design: filter
pushdown, KvBatchScanner for the FIP-17 KV scan, LookupBatch,
FlussBucketingFunction.bucketForRowKey, and SortMergeReader for the
primary key merge.

FIP-50 sits directly above FIP-48: the union read kernel supplies the
lake side and this proposal supplies the catalog, planner and pushdown
above it, which means tiered reads in FIP-50 depend on FIP-48.

Looking forward to your feedback.

-- Anton

Reply via email to