> For the complete documentation index, see [llms.txt](https://ciusji.gitbook.io/luden/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://ciusji.gitbook.io/luden/features/stream-mode.md).

# Stream Mode

LUden supports streaming of incoming events for data warehouse. In Stream mode all events are loaded to destinations in real-time one by one. That is the fastest way to get data to destinations, but it may not be as performant and cost-effective as [Batch mode](/luden/features/match-mode.md).

Luden uses intermediate queue that helps to survive temporary destination unavailability or slowdowns.

Currently, Luden supports only one streaming thread for destination. That may be a bottleneck in cases of high event number. In that case [Batch mode](/luden/features/match-mode.md) is recommended.

## Pipeline

* Apply multiplexing, put each multiplexed event to the destination queue. Queue items are persisted in `events/queue.dst=${destination_id}` dir.
* Separate thread processes each queue. For each event:
  * Run `table_name_template` expression. If the result is an empty string - skip. If evaluation failed, the event is written to `events/failed`
  * Apply LookupEnrichment step
  * Apply Transformation and MappingStep (get BatchHeader)
  * Get Table structure from memory (if memory cache is missing, get the schema from DWH)
  * Do **table patching** (see above in batch step)
  * Insert object with explicit typecast (if it is configured in the [JavaScript Transformation](/luden/features/javascript-transform.md)) using INSERT statement.
  * If INSERT failed, refresh schema from DWH and repeat the step
  * If failed, write the record to `events/failed`
  * If success, write the event to `events/archive`