Scale Daft workflows to distributed Ray clusters. Invoke when optimizing performance or handling large data.
Scale single-node workflows to distributed execution.
| Strategy | API | Use Case | Pros/Cons |
|---|---|---|---|
| Shuffle | repartition(N) | Light data (e.g. file paths), Joins | Global balance. High memory usage (materializes data). |
| Streaming | into_batches(N) | Heavy data (images, tensors) | Low memory (streaming). High scheduling overhead if batches too small. |
Best for distributing file paths before heavy reads.
# Create enough partitions to saturate workers
df = daft.read_parquet("s3://metadata").repartition(100)
df = df.with_column("data", read_heavy_data(df["path"]))
Best for processing large partitions without OOM.
# Stream 1GB partition in 64-row chunks to control memory
df = df.read_parquet("heavy_data").into_batches(64)
df = df.with_column("embed", model.predict(df["img"]))
Target: Keep all actors busy without OOM or scheduling bottlenecks.
Calculate the Max Partition Count to ensure each task has enough data to feed local actors.
Batch Size * (Total Concurrency / Nodes)Total Rows / Min Rows Per PartitionExample:
64 * (16/4) = 256.1,000,000 / 256 ≈ 3906.df = df.repartition(1000) # Balanced fan-out
Avoid creating tiny partitions. Use into_batches to stream data within larger partitions.
Strategy: Keep partitions large (e.g. 1GB+), use into_batches(Batch Size) to control memory.
# Stream batches to control memory usage per actor
df = df.into_batches(64).with_column("preds", model(max_concurrency=16).predict(df["img"]))