The primary purpose of the virtual warehouse local disk is to accelerate remote data read performance under the architecture of storage and compute separation. In ByteHouse Cloud Data Warehouse, the default write behaviour is to write through to remote persistent distributed storage. Subsequent SQL reads trigger a return to the local disk.
To proactively trigger extraction and loading of cached content in the local disk, refer to Cache preload.
The files stored on the virtual warehouse local disk are mainly divided into data blocks, metadata, unique index files, BSP temporary landing files, temporary files for insert foreground writes and background data merges, and others. Data blocks occupy approximately 70–80% of the space, while index and temporary files occupy around 10%.
For more specific information about the proportions or to make targeted adjustments, submit a ticket or contact the ByteHouse team.
When an SQL error displays Cannot reserve xx MB, not enough space, it is generally due to insufficient reserved space for temporary files on the local disk. The reason for this problem may be increased or more frequent foreground insert operations, larger batch row counts, more frequent background data merges, or excessively large merged files.
Additionally, ByteHouse Cloud Data Warehouse virtual warehouse supports separate settings for read and write virtual warehouses. If the read and write virtual warehouse are set as the same virtual warehouse, this problem is more likely to occur.
You can resolve this issue using the following methods: