Node
Interfaces
- BucketingAlgorithmBuilder
- JoinAlgorithmBuilder
- JoinsFrame
- SortAlgorithmBuilder
Classes
- Aggregate
- Every group becomes one row, and the rows of a group can arrive in any batch, so the input is drained before the first row is emitted.
- Batch
- BatchBy
- Cache
- A write-through: every batch is yielded onward as it is cached, so there is no materialisation barrier, but the cache is an effect outside the stream.
- Collect
- CollectRefs
- Records the references it sees - an effect outside the stream - and passes every row.
- Constrain
- Throws on the first violating row - an effect outside the stream - and passes every row otherwise.
- Count
- CrossJoin
- Every left row is paired with every right row; the right side is pulled on the first batch.
- Discard
- Distinct
- Drops a row whose key was already seen; the decision depends on the row alone and the seen-set is internal to the stream.
- Drop
- DuplicateRow
- ExpandColumn
- Filter
- Join
- Row count depends on the join type and on both sides' data; the hash join buckets the whole input before emitting.
- JoinEach
- The right side is built from each left batch's row values, so no static edge exists and the row count is unknown.
- Limit
- Offset
- Outputs
- The root of a plan with several consumers of one prefix
- Read
- Rename
- RenameEach
- Repartition
- Buckets the whole input by key before emitting; every row passes, in a different order.
- Result
- The plan root (or first child of Outputs) that fetch/get/run pull; adds no steps.
- Select
- Sort
- Every row passes, but the output order is a function of the whole input, so the input is drained first.
- TopN
- Opaque and blocking: which rows survive depends on the whole input.
- Transaction
- A ROOT grouping write roots that commit together; never on a spine.
- Transform
- Effect unknown but never blocking; a Stateful one runs as its fresh() instance.
- Until
- Validate
- Throws when a batch does not match the expected schema - an effect outside the stream - and passes every row otherwise.
- WindowColumn
- A window value is a function of the whole partition, so the input is drained (or repartitioned) before the column is computed.
- WithColumn
- Write
- Hands every batch to the loader - an effect outside the stream - and passes it on unchanged.