Module spill 

Spill-to-disk infrastructure for blocking operators.

Provides:

• SpillConfig / SpillManager: temp-file lifecycle management
• SpilledRun / RunWriter / RunReader: the single spill file format — columnar v2 run (LRSC): versioned header, schema fingerprint, per-section checksums, optional per-section zstd body (columns are contiguous postcard-encoded value slices, one section per column group)
• HashPartitionSpiller: per-partition run writers
• DiskQuota: separate disk usage tracking for spill operations

Structs

DiskQuota

Tracks disk space used by spill operations for a single query.

HashPartitionConfig

Configuration for hash-based partition spill.

HashPartitionSpiller

A partition spill writer that routes rows by hash into separate files.

RunHeader

Header for a columnar spill run file.

RunReader

Reads a columnar run file back, validating the header (magic, version, checksum) on open and each section frame (checksum) on decode.

RunWriter

Writes a columnar run to disk with a versioned header, schema fingerprint, and per-section checksums.

SpillConfig

Configuration for disk spill behavior.

SpillManager

Manages spill-file creation, cleanup, and tracking for one query execution.

SpilledRun

Metadata for a single spill run file (sorted spill).

Enums

RunCompression

Compression marker stored per section frame.

Constants

COLLECTOR_RUN_ROWS_MAX

Maximum rows per terminal-collector run file. Bounds both the writer-side body buffer and the reader-side full-run load peak.

COLLECTOR_SPILL_ROWS_DEFAULT

Default terminal-collector spill threshold in logical rows.

HASH_JOIN_MAX_DEPTH

Maximum Grace Hash Join repartition depth.

HASH_JOIN_PARTITIONS_DEFAULT

Default Grace Hash Join partition count.

HASH_PARTITION_SEED

HASH_PARTITION_VERSION

Fixed hash algorithm for partition spill operations.

Functions

finalize_partitions_with_runtime

Finalize a partition spiller through the runtime’s spill manager.

hash_bytes_partition

Compute the partition index for pre-serialized key bytes.

hash_column_partition

Compute one partition index per batch row over a column subset.

hash_row_partition

Compute the partition index for a row of values.

schema_fingerprint

Schema fingerprint for a run file: FNV-style hash over column names.