Quasar “Seneca” 3.14.3 Released

Quasar 3.14.3 is now available.

The release focuses on immediate visibility for asynchronous writes, native Parquet support, and faster analytics.

Query asynchronous writes immediately

Data accepted by an asynchronous pipeline is now available to queries immediately, including deduplicated writes.

Previously, applications had to wait for the pipeline to flush before reading the data they had just written. They can now combine asynchronous ingestion with immediate access to new data.

New API and shell commands can flush a pipeline and report its queue length. Random bulk readers can also read pending asynchronous writes.

Native Parquet import and export

The IMPORT command introduced in 3.14.2 now supports Parquet.

The new COPY SELECT statement exports query results to CSV, TSV, or Parquet. Parquet exports support configurable compression codecs and levels.

CSV import is also faster. The timestamp column can now be selected by name or position.

Faster aggregation and SELECT queries

Aggregation is faster and uses less memory, especially with high-cardinality groups.

SELECT queries can now use the table shard index to reduce the amount of data they scan. This builds on the multithreaded SELECT engine introduced in 3.14.2.

Apache Arrow streams

Bulk readers and batch writers now support Apache Arrow streams for efficient column-oriented data transfer.

Bulk readers can retrieve metadata without selecting data columns, preserve the requested column order, and explicitly select $table and $timestamp. Random bulk readers now support aggregated tables.

Linux packages now include the static C API library. The C API no longer depends on third-party shared libraries at runtime.

Foundations for general tables

The 3.14.3 kernel supports tables without a time-series structure. This is groundwork for the general-table support planned for a future major release.

The release also adds $rowid, configurable limits on rows per shard, and APIs to inspect table shards.