Skip to content

HydraCache 0.70.0

Choose a tag to compare

@github-actions github-actions released this 19 Aug 14:08
· 336 commits to main since this release
75719b0

HydraCache 0.70.0

0.70 is an allocation-path and retained-owner correctness release. It adds exact subsystem owner
snapshots and deterministic cleanup contracts before making any host-level RSS claim.

Implemented evidence includes bounded tag-generation and load-breaker ownership, bounded client
idempotency/audit/event state, quiescent cache flush and TTL cleanup, and HC/2
invocation/subscription/session release. The release lane now includes behavioral defect canaries,
a 100-cycle core retention soak, a million-mutation fixed-keyspace plateau proof, and exact-SHA
admission receipts.

GitHub-hosted HydraCache/Redis measurements remain explicitly non-promotable. A scheduled
diagnostic stores raw evidence for 30 days and a compact fingerprinted summary/trend for 90 days;
numeric RSS/PSS changes are reported but never used as an absolute PR threshold.

Release workflows no longer rely on deprecated Node.js 20 action runtimes: artifact downloads use
actions/download-artifact@v8, including fail-closed digest verification, and the operator Helm
proof uses azure/setup-helm@v5.0.1; both execute on Node.js 24.
The operator proof also gives its background controller an explicit 120-minute timeout, preventing
the GitHub runner from failing an otherwise successful proof during the final cancel step.

Conditional delete tombstones are reclaimed only through a bounded per-partition watermark built
from the minimum ordered applied prefix acknowledged by every effective replica in the current
authority epoch. Invalid or incomplete progress fails closed, and stale replicated records below a
reclaimed prefix cannot resurrect deleted keys.

The exact candidate passed release-evidence --release 0.70 --require-ship, including the
candidate-bound fast, canary and retention-soak receipts.

The corrected TTL row and local WSL2/Docker measurement campaign are transferred to the 0.71 W0
causal baseline. They are not 0.70 ship gates and cannot be used as 0.70 sizing or product-ranking
evidence.