schema of the root partitioned series's
per-partition LiveSeries (kept so the composed factory's
input type is correct).
schema of the current chained output.
partition key type.
partition column name (literal). Threaded
through from the root LivePartitionedSeries so chained-rolling
typing chains (e.g. partitionBy('host').fill(...).rolling({...}))
resolve FusedPartitionedRollingSchema<R, ByCol, FM> correctly.
ReadonlyschemaSchema of the chained output. Captured by running the factory
once on a stub LiveSeries<SBase> at construction.
Apply a further per-partition transform on top of the existing
factory chain. Equivalent to chaining one more sugar method
via a custom function. Returns a unified LiveSeries<R2>.
Optionaloptions: Partial<LiveSeriesOptions<R2>>Same as LivePartitionedSeries.collect, applied through the factory chain.
Optionaloptions: Partial<LiveSeriesOptions<R>>Optionaloptions: { limit?: number }Partition column drops by default. rolling's output
schema only retains columns named in mapping. Include the
partition column with a passthrough reducer (e.g.
host: 'last') to keep it visible in the unified output.
Optionaloptions: LiveRollingOptions & { trigger?: { kind: "event" | "count" } }Partition column drops by default. rolling's output
schema only retains columns named in mapping. Include the
partition column with a passthrough reducer (e.g.
host: 'last') to keep it visible in the unified output.
Partition column drops by default. rolling's output
schema only retains columns named in mapping. Include the
partition column with a passthrough reducer (e.g.
host: 'last') to keep it visible in the unified output.
Keyed-form fused multi-window rolling on a chained
LivePartitionedView. Same shape as the root variant — each
partition's chain output flows into one fused rolling that
maintains N windows in one ingest pass and emits one merged
event per partition per boundary.
Clock trigger required (same constraint as the root partitioned fused — synced cross-partition emission).
Per-partition stream sampling on a chained view. Same semantics as LivePartitionedSeries.sample — stride only, safe by construction (no multi-entity bias); each partition's chain output is thinned independently with its own counter.
Materialize the chained view per-partition as a
Map<K, LiveSource<R>>. Runs the composed factory once per
existing partition; auto-spawn from the root partitioned
series is not propagated into this map (the snapshot
reflects partitions at the time of the call).
Each factory output (a LiveView / LiveRollingAggregation /
etc.) holds an internal subscription to its source. To avoid
accumulating listeners across repeated calls, every factory
output's dispose() is registered on the root's disposer set
— calling partitioned.dispose() on the root cleans up every
toMap-created subscription chain.
For a live-updating per-partition view, subscribe to the root
partitionBy directly with toMap() and call the factory
yourself, or use collect() for a unified buffer.
Chained typed view over a LivePartitionedSeries. Returned by every sugar method on the root partitioned series and on this view, composing the operator factory at each step.
The view is lazy: factories aren't run until a terminal (
collect(),apply(),toMap()) is called. Each terminal delegates back to the root's per-partition state, applying the composed factory chain to each partition'sLiveSeries.Lifecycle. All real state lives on the root
LivePartitionedSeries— chained views are just deferred factories that point back at the root. They don't register their own subscriptions on the source. Disposing the root disposes everything: terminals subscribed to factory outputs are tracked on the root's internal disposers, including outputs created byview.toMap().Example