Cold partitions in #QuestDB land in object storage as Parquet. Point Apache Iceberg at them and Spark, Trino, DuckDB and PyIceberg can query your history in place.
You write an Iceberg manifest over Parquet you already have and register incrementally new partitions.
Your producer should not lose data to a network hiccup or a #QuestDB restart. Every QWP client in 10.0 buffers unacked writes, in memory by default, on disk if you give it a directory, and replays them when the server comes back. Publishing never blocks either way.
We ran TSBS against #QuestDB's new binary protocol and got 19M rows per second over a network, where line protocol gets 5.3M.
The gain is the wire: 97 bytes per row against 347. ILP fills the link first.
48M rows/s on a single machine, 33M at a million series.
QWP, the new #QuestDB wire protocol, does egress as well as ingestion. Query results stream straight into Apache Arrow.
500 million rows in 2.3 seconds. 220 million rows per second. First batch after 32ms. 18.8 bytes per row on the wire.
Two billion rows at 291M rows/sec.
We just shipped #QuestDB 10.0.
QWP: one binary protocol for writes and Arrow reads, 3.6x faster ingestion, 220M rows/sec out.
Live views in beta.
Web Console notebooks, drivable by coding agents.
Parquet tables with schema evolution.
Enterprise 4.0 cold storage ships next.
