Skip to content

HydraCache 0.67.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 27 Jul 07:01
· 579 commits to main since this release
v0.67.0
e0477ec
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

HydraCache 0.67.0 - Performance measurement infrastructure

Status: shipped 2026-07-27 as a performance measurement infrastructure release without numerical performance claims.

Release scope

Release 0.67 delivers the performance-testing infrastructure and methodology implemented by W0-W10:

  • a fixed-rate open-loop scheduler with scheduled-send latency, repeat/reset discipline, bounded histograms, and a complete sustainable-throughput-at-SLO predicate;
  • typed, digest-bound reports for the real embedded cache, the in-process Axum client surface, one selected node-local RESP endpoint, real daemon control-plane operations, and explicitly labeled library/model primitives;
  • surface-specific brownout and overload scenarios, pinned same-box Redis comparison tooling, and exported-only metrics cross-checks;
  • immutable-anchor and rolling-main budget validation that rejects silent rebaseline, unstable hosts, mixed fingerprints, candidate self-baselining, and incomplete evidence;
  • release-scoped canaries, registered exact commands, prebuilt-binary identity binding, receipt aggregation, hosted tripwires, and a protected manual self-hosted reference workflow.

The load generator remains publish = false. The release adds no product protocol, distributed RESP backend, consistency level, or performance optimization.

Claim boundary and deferred evidence

The code and methodology above are the 0.67 ship scope. Official numerical evidence is not.

The reference-v1 anchor, rolling baseline, and budgets remain unbootstrapped because no protected non-shared bare-metal runner is connected. Bootstrap and the final core, RESP/Redis, control-plane, budget, and aggregate reference run are tracked by TD-0013.

Until TD-0013 is closed, HydraCache publishes no official:

  • capacity floor or sizing guidance;
  • node-local RESP or distributed cluster capacity result;
  • Redis comparative or superiority claim;
  • metrics-agreement performance result; or
  • portable numerical baseline.

GitHub-hosted ci-shared measurements are broad regression tripwires only. The five manual reference-v1 execution gates remain registered and fail-closed, but are deferred evidence gates and do not block shipping the 0.67 tooling release.

Surface boundaries

The tooling must never relabel:

  • the in-process /client/v1/* Axum router as a daemon wire;
  • independent node-local RESP stores as a distributed value plane;
  • summed per-node throughput as cluster capacity;
  • library/model primitive costs as end-to-end daemon-grid performance; or
  • one same-box Redis comparison as universal superiority.

See ../PERFORMANCE.md for the methodology and quotation rules. Tag v0.67.0 publishes this claim-free tooling scope; it does not bootstrap or claim the deferred reference evidence.

Assets 2
Loading