Class Host

constructor

• new Host(
      options: {
          budgetBytes?: number;
          maxSources?: number;
          registry: Registry;
          sources?: SourceRegistry;
          units?: Readonly<Record<string, string>>;
      },
  ): Host

  Parameters

  • options: {
        budgetBytes?: number;
        maxSources?: number;
        registry: Registry;
        sources?: SourceRegistry;
        units?: Readonly<Record<string, string>>;
    }

    • OptionalbudgetBytes?: number

      Cap on retained node values per bound graph, in bytes — see bind. Without one, a host accepting runtime plans retains every distinct spec ever asked of every dataset, so memory scales with questions asked rather than with anything bounded. A host whose plans arrive from callers it does not control should set this.

      What bounds the number of graphs is a separate question with two answers. Datasets registered with add are the host author's own, bounded by what the author adds and released with remove. Sources resolved through a source registry by runAsync are request-driven — every distinct SourceRef a caller supplies binds another graph — so a host exposed to untrusted callers must also set maxSources. The retained total is then at most budgetBytes × (author datasets + maxSources).

    • OptionalmaxSources?: number

      Cap on sources resolved through the source registry, LRU — the request-driven half of the host's footprint. runAsync binds a graph per distinct SourceRef, and a caller chooses the refs, so without a cap an untrusted caller grows the host without bound. Author-added datasets (add) are never evicted by this. Unbounded when omitted — acceptable only when every SourceRef the host will see comes from code the author controls.

    • registry: Registry

    • Optionalsources?: SourceRegistry

    • Optionalunits?: Readonly<Record<string, string>>

  Returns Host

Readonlyregistry

datasets

• get datasets(): DatasetInfo[]

  Returns DatasetInfo[]

add

• add(id: string, series: TimeSeries<SeriesSchema>): this

  Registers a dataset. The graph is built lazily on first use, so seeding many datasets is cheap.

  Parameters

  • id: string
  • series: TimeSeries<SeriesSchema>

  Returns this

graphFor

• graphFor(id: string): BoundGraph

  The bound graph for a dataset, built on first use and kept.

  Parameters

  • id: string

  Returns BoundGraph

has

• has(id: string): boolean

  Parameters

  • id: string

  Returns boolean

remove

• remove(id: string): boolean

  Forgets a dataset: its source, its graph, and every cached node value with them. The explicit end of a binding's lifecycle — the counterpart to add for a host that cycles datasets, where "kept until the process dies" is not a policy.

  Returns whether the dataset was known. A load in flight for the same source id discards its result rather than resurrecting the removed dataset — its runAsync callers get UnknownDatasetError, which is what asking for a concurrently-removed dataset means.

  Parameters

  • id: string

  Returns boolean

run

• run(envelope: Envelope): RunResult

  Resolves an envelope against its dataset's long-lived graph.

  Parameters

  • envelope: Envelope

  Returns RunResult

runAsync

• runAsync(envelope: AsyncEnvelope): Promise<RunResult>

  Resolves either a preloaded dataset or an opaque asynchronous source.

  Equal revisions preserve the graph untouched. A changed revision updates its source in place, retaining compiled nodes while normal invalidation propagates from the new value.

  Parameters

  • envelope: AsyncEnvelope

  Returns Promise<RunResult>