LoginSign Up
← Back to all sparks
Q

QuestDB

DEVOPS
Velocity6.3

High-performance time-series database for real-time analytics

questdb.io ↗

QuestDB 10.0 collapses ingest and egress into one binary protocol, then aims at agent-run notebooks.

time-series-dbwire-protocolarrowingestion-throughputagent-nativecapital-markets
◆Current state
QuestDB just shipped 10.0, its first major version bump in the visible window, built around QWP — a binary columnar wire protocol that handles writes going in and streams Arrow back out through a single client. The release also carries live views in beta, notebooks driven by coding agents, and the storage groundwork that QuestDB Enterprise 4.0 builds cold storage on. The days since have been spent substantiating it: a parallel-reader benchmark puts 500M rows into Arrow in 2.3 seconds against ClickHouse and TimescaleDB. Around the releases, the feed is mostly engineering deep-dives, capital-markets case studies, and comparison guides.
◆Where it's heading
The protocol work is the thread that matters. QuestDB has been positioning against InfluxDB Line Protocol on ingestion throughput for a while — 33M rows/s on one machine at a million series, roughly 3.6x ILP over a network — and 10.0 turns that from a benchmark argument into the default path in and out. The follow-up post shifts the argument to the read side, which is the half QWP actually changes: folding Arrow egress into the same protocol points the database at the Python analytics stack directly rather than through a SQL driver, something the Polars and ConnectorX content has been rehearsing for months. Live views and agent notebooks are earlier-stage, and the Enterprise line continues on its own track.
◆Prediction
Expect QuestDB Enterprise 4.0 next, built on 10.0's storage work and leading with cold storage, plus client libraries catching up to QWP one language at a time. Live views and the agent notebooks look likeliest to leave beta on a later minor rather than in the next release.

◆Recent moves

  1. 6d ago

    Streaming 500 million rows into Apache Arrow in 2.3 seconds

    A follow-up benchmark on the just-shipped QWP read path: a naive parallel reader streaming 500M rows into Arrow in 2.3 seconds, measured against ClickHouse and TimescaleDB with three measurement artifacts stripped out. Methodology content backing the 10.0 claims rather than a release of its own.

    View source ↗
  2. 7d ago

    QuestDB 10.0: QWP, one binary streaming protocol for writes and Arrow reads

    ⚡ SPARK

    The major-version release that turns QuestDB's protocol argument into shipped default behaviour: QWP carries writes in and Arrow out through one client, alongside live views in beta and agent-driven notebooks. It also lays the storage foundation Enterprise 4.0 is said to build cold storage on, so the Enterprise track now depends on this release.

    View source ↗
  3. 8d ago

    Introducing QuestDB's new binary ingestion protocol: QWP

    The technical introduction of QWP a day ahead of the 10.0 release, with the TSBS numbers behind it: 33M rows/s on a single machine at a million series, about 3.6x InfluxDB Line Protocol over a network. The protocol itself lands as part of 10.0 — this is the benchmark case for it, not a separate shipment.

    View source ↗
  4. 24d ago

    Transaction Cost Analysis with QuestDB and Polars: VWAP, Slippage and Markout

    A worked transaction-cost-analysis pipeline — VWAP, arrival slippage, markout curves — in QuestDB SQL with Polars over ConnectorX. Vertical content for the capital-markets audience, and a preview of the dataframe-side integration QWP's Arrow egress is aimed at.

    View source ↗
  5. 27d ago

    HDFC Bank uses QuestDB for mule account detection across all major 25+ banking channels

    A customer case study: HDFC Bank runs QuestDB inside its streaming platform for mule-account detection at 5,000-7,000 transactions per second on a single instance. Reference material rather than a product change, though it marks fraud detection as a second anchor vertical alongside trading.

    View source ↗
  6. 29d ago

    The Most Expensive Instruction Might Be… cmov

    An engineering deep-dive on HotSpot's C2 branch-to-cmov heuristic and why measuring branch bias rather than predictability costs up to 2.9x in a tight loop. JVM internals content, no product change.

    View source ↗