Choose Stream History

Choose a cached representation that fits the screen:

Need	Builder	Retained data
Every emitted value in order	streamedOptions	Unlimited history by default.
A recent event window	streamedOptions with maxChunks	Up to the specified number after an emission.
Current state	liveOptions	The latest emitted value.

For a long-running event display, choose a finite maxChunks. For current progress or a feed of complete state, choose a live query. Unlimited accumulation grows with each emission. Use it for finite histories or workloads that deliberately retain every emission. If history must survive cache eviction or application restarts, persist it outside the query cache.

These choices apply to both RPC streams and supported HTTP SSE endpoints. Live queries normalize successful undefined to null. Accumulated histories preserve their emitted values.

Set retention and buffering separately

maxChunks limits cached elements after each new emission. Seeded or hydrated history can exceed the bound until another element arrives. One element can contain a large object or array, so the bound does not limit bytes. The cache discards older elements and preserves the remaining order. Use the same retention and refetch policy when building an exact key and its options.

RPC rpcOptions.streamBufferSize controls the ready client’s transport buffer before consumption. It does not set cache history or the rate of cache writes. HTTP transport buffering belongs to the ready HTTP API client. Changing a history bound does not slow the producer.

An initial accumulated or live fetch exposes its first value while the stream remains open. An accumulated replace refetch publishes its replacement only after successful completion. If the replacement stream stays open indefinitely, the previous history remains visible. See Stream Snapshots for finite server captures.

Assess writes and references separately

Accumulated streams build new history arrays without changing previous publications. A history bound limits array length. Each consumed value still contributes to cache updates. Live queries also accept each value, including repeated values. Native structural sharing can preserve equal data references without suppressing those cache writes.

For JSON-compatible values, TanStack’s default structural sharing can reuse unchanged objects and nested values. A moving window changes array positions, so sharing can produce more distinct references. Compare the data shape your application receives before choosing structuralSharing: false or a custom native policy.

Cache notifications also include query creation and changes during fetching. Count data writes, data reference changes, and retained elements separately. Framework observer selection and render scheduling affect rendering. Cache notification counts do not count component renders.

If the producer outruns the interface, set a rate or sampling policy in the producer or an application-owned Effect Stream transformation. Sampling intentionally discards values. Retention, transport buffering, and publication rate have separate purposes.