LoginSign Up
← Back to all sparks
D

Dragonfly

INFRA · APIS
Velocity6.3

In-memory datastore compatible with Redis and Memcached

www.dragonflydb.io ↗

Dragonfly tags v2.0.0 while hardening memory accounting and patching a HyperLogLog CVE across the 1.x line

redis-compatiblein-memory-databasememory-managementperformancecaching
◆Current state
Dragonfly is shipping across two concurrent tracks: the 1.40.x stable line is receiving steady improvements to connection memory accounting correctness, tiering metrics, and Redis ecosystem compatibility (RedisShake RDB format), while v2.0.0 has been tagged. The 2.0.0 entry's visible content is minimal — scope-based memory tracking is added but disabled — suggesting 2.0.0 is an architectural milestone marker for accumulated work rather than a single user-visible feature introduction.
◆Where it's heading
The project's operational focus is sharpening around production reliability and Redis compatibility: O(1) connection memory tracking, per-shard tiering metrics, CVE-2025-32023 remediation in HyperLogLog, and AVX2/NEON SIMD acceleration for dense HLL operations (8-29x measured gains). The staged 2.0.0 release and the disabled scope-based memory tracking point to Dragonfly building toward granular per-connection memory visibility as a differentiating feature for operators running large clusters.
◆Prediction
Scope-based memory tracking will be re-enabled in a near-term 2.x patch — the code is already shipped in 2.0.0, just gated off. That re-enable will be the actual headline for what 2.0.0 unlocks operationally.

◆Recent moves

  1. 7d ago

    Dragonfly v2.0.0: major version milestone with scope-based memory tracking staged

    ⚡ SPARK

    Dragonfly v2.0.0 is tagged, with scope-based memory tracking included in the codebase but disabled by default pending a follow-on PR. The major version bump is itself the signal — v2.0.0 marks the architectural threshold beyond which breaking changes become acceptable and the 2.x differentiation story begins.

    View source ↗
  2. 9d ago

    i1.40.8: fix(server): make connection memory accounting constant time (#8245)

    Connection memory accounting is restructured to O(1) by tracking subscription, WATCH, and MULTI allocations at mutation sites and caching ACL glob memory at install time — eliminates per-command traversal overhead that grew with connection state.

    View source ↗
  3. 12d ago

    Fix RDB listpack compatibility with RedisShake migrations

    Patches RDB_TYPE_SET_LISTPACK compatibility with RedisShake — a commonly used Redis data migration tool. Teams doing live migrations from Redis to Dragonfly via RedisShake were hitting deserialization failures on set data encoded as listpack.

    View source ↗
  4. 1mo ago

    Tiering: unified pending-bytes limit and metrics export

    Tiering consolidates to a single tiered_max_pending_bytes limit and exports it to metrics — aligns with the trajectory of making Dragonfly's disk-tiering behavior more observable and tunable for operators running hot/cold workloads.

    View source ↗
  5. 1mo ago

    i1.40.2: backport: mem accounting fix, journal flushslots in tx (#8061)

    Backports memory accounting correctness fix to the 1.40.x line and fixes a cluster bug where FLUSHSLOTS wasn't being journaled from inside the flush transaction, risking replication inconsistencies on cluster resharding.

    View source ↗
  6. 1mo ago

    HyperLogLog: CVE-2025-32023 patch, SIMD acceleration (8-29x), and three correctness fixes

    ⚡ SPARK

    Dragonfly syncs its HyperLogLog implementation with Valkey upstream, fixing CVE-2025-32023 (wild-pointer write in sparse HLL decoding) and three local bugs including a dropped-registers defect in multi-key PFCOUNT. AVX2 and ARM NEON SIMD kernels arrive with 8-29x measured gains over scalar for dense HLL merge and compress operations — a meaningful capability upgrade for cardinality-heavy workloads.

    View source ↗