Expand description
Pluggable storage for completed, serialized page blobs.
While a row group is being written the ArrowWriter must buffer every
column’s encoded pages, because Parquet requires each column chunk to be
contiguous in the file while record batches arrive with all columns interleaved.
By default that buffer lives on the heap, so the writer’s peak memory grows
with the row group size. A PageStore lets the buffer live somewhere else
— a local temp file, object storage, etc. — bounding peak write memory
independently of the row group size.
Structs§
- InMemory
Page Store - The default
PageStore, holding blobs on the heap in aVec<Bytes>. - InMemory
Page Store Factory - Factory for
InMemoryPageStore— the default used byArrowWriter. - PageKey
- An opaque, store-allocated handle to a blob held by a
PageStore. - Page
Store Args - Context for a single
PageStoreFactory::createcall.
Traits§
- Page
Store - A pluggable store for completed, serialized page blobs.
- Page
Store Factory - Creates a fresh
PageStorefor each column chunk.