What is a delta store in HANA's columnar storage architecture?

HANA's columnar format is highly optimized for reads but comparatively expensive to update in place — inserting a single row into a compressed, dictionary-encoded column would require re-encoding potentially the whole column. To avoid that cost on every write, HANA maintains a separate, smaller delta store that accepts new writes quickly (in a simpler, less heavily compressed row-oriented structure), while the bulk of the table's data remains in the heavily optimized, read-efficient main store.

flowchart LR
    A[New INSERT/UPDATE] --> B[Delta store - fast writes, less compressed]
    C[Main store - heavily compressed, read-optimized] --> D[Query reads both stores together]
    B --> D

Queries transparently read from both the main store and the delta store together to produce a complete, current result, so this split is invisible to application code — it's purely an internal performance optimization that lets HANA offer both fast writes and fast, heavily-compressed reads without those two goals directly conflicting on the same physical structure.