Repository navigation
Description
2026-11-21 documentation preflight complete (not published): PR #93 on exact HEAD
1aa2b2151eb09f93bae8c7f7b79083e7bc201e64now has all workflows SUCCESS: CI 29/29, docs 15 pages / 588 links, CodeQL/changelog/title. Counteraudit package. Still DRAFT/unmerged pending review; do not silently imply documented 0.6.0 has been published. #85 was merged intostableas190d1059and #54 closed as implementation complete only. Remaining hard gates: #53 security disposition, #91 consumers 0/2, fresh 0.6 final wheel/sdist provenance/security/Dependabot, manual RC/tag/TestPyPI/PyPI, Pagesstableenvironment setup and actual URL verification. #50 events deferred by default to 0.7.0. Older notes below are historical.
2026-11-21 merge gate achieved, release still NO-GO: PR #85 MERGED at
190d105915972d436ddbf4ad152ad654a4db392aafter independent approval of exact HEAD098e37cb..., CI 29/29, migration 13/13, CodeQL/security and benchmarks. #54 source implementation is complete. #93 now updates staged-vs-published docs (draft awaiting CI). The 0.6.0 final artifact is still unbuilt/unapproved, #53 classification unresolved, #91 independent consumers at 0/2, #48 Pages unverified; #50 deferred by default to 0.7.0. No tag, RC, PyPI or promotion authorized. Older pre-merge notes below are historical.
2026-11-21 exact-head technical preflight: #85 now at
098e37cb4530a3d4fd282080e2d0c3251e089f4aafter read-onlymax_entriesfix. CI 29/29, migration Phase B 13/13, CodeQL, installed-security, Redis matrix, changelog all SUCCESS. Counteraudit evidence. Still unmerged: independent reviewer GO is required, so first checklist item remains unchecked. #91 0/2, #53 and final candidate release gates pending; 0.5.0 remains latest published.
Decision and current status (2026-11-21)
0.6.0 may enter release preparation soon, but final publication is NO-GO today. The latest published stable distribution is fastapi-cache-stable==0.5.0. This issue is the acceptance gate for the first intentionally promoted release (not the first PyPI release); it does not schedule or authorize a tag, a publication or external promotion. Track feature scope in #86 and promotion in #88.
Proposed limited scope: #54 / draft #85 only after independent approval; preserve fastapi_cache import and ordinary key goldens; privacy/tenant risk disposition #53; documentation and adoption evidence. Recommend deferring #50 to 0.7.0 if its opt-in privacy-safe instrumentation is not fully implemented/tested/reviewed before scope freeze. Do not add #49, stale serving, tags, single-flight, distributed locks, rate limiting or circuit breakers just to match feature-heavy competitors.
Phase A — code freeze and internal candidate
- feat(inmemory): optional max_entries LRU capacity without changing defaults (#54) #85 latest exact HEAD (
098e37cb4530a3d4fd282080e2d0c3251e089f4aas of issue creation) passes new, full CI/benchmark/Redis/security/migration after makingmax_entriesread-only. Obtain independent code-review GO on that exact SHA. Do not inherit the formerc33aa6f29/29 checks. Counteraudit request. - Merge feat(inmemory): optional max_entries LRU capacity without changing defaults (#54) #85 only when approved; record resulting
stableSHA, check required branch/status protections, re-run impacted checks. Documentmax_entriesas fixed at construction (no runtime resizing), per-instance entry count rather than payload bytes, bounded heap/LRU/TTL and explicit bulk-clear O(N + k log N). Keep published 0.5.0 docs/version separate. - Explicitly decide feat(observability): optional cache events after Redis modernization #50: if no audited PR before freeze, record deferred to 0.7.0, do not include unshipped events in 0.6 docs/marketing.
- security(cache): residual hidden tenant/identity context and header-cardinality safety after 0.5.0 #53: source-level / FastAPI test matrix of Authorization/Cookie bypass, explicit
Requestbypass,vary_headers, nonstandard tenant/role/locale dimensions, forwarded headers, header cardinality, untrusted middleware/contextvars and cache-data migration; classify package-owned defect vs application-owned risk. NO-GO for a reproduced unmitigated package-owned private disclosure. Document mitigations and residual limits regardless of whether security(cache): residual hidden tenant/identity context and header-cardinality safety after 0.5.0 #53 itself remains open. - Create a separate version/changelog/docs PR for
0.6.0candidate after feature and security scope freeze. Update API reference and feature matrix to state what's proposed vs published; preserve 0.5.0 historical docs.
Phase B — exact artifacts and real adoption
- Build wheels/sdists from the frozen source SHA; record SHA-256 for both. Validate fresh
python -Iinstalled-package smoke,pip check, metadata Requires-Python, supported Python 3.8–3.13 functionally (modern security 3.11–3.13), FastAPI/Pydantic and Redis client/version profiles. Do not conflate Python 3.8–3.10 functional legacy compatibility with a security guarantee. - Re-run real Redis 7.4 Phase B old 0.2.2 writer → 0.6 candidate reader, selective namespace clearing and safe package rollback (do not restore unsafe authenticated caching), clean venv and site-packages provenance. No Redis Cluster namespace-wide-clear claim.
- Fresh installed-security audit, CodeQL, authenticated Dependabot delta and package-owned reachability/mitigation record for the exact final artifacts. Reject blanket “0 vulnerabilities” claims and inherit no 0.5.0 release approval.
- test(0.6): two genuinely independent consumer exercises and evidence gate #91 2/2 genuine external consumers, one fresh application on InMemory+Redis and one separately maintained actual legacy migration/rollback. Currently 0/2; the docs(migration): runnable published before-and-after demo for adoption #90 in-repo before/after demo counts 0 and remains pinned to published 0.5.0 until publication.
- Release candidate (e.g.
0.6.0rc1) / TestPyPI only with an explicit maintainer go-ahead, accurate prerelease version and provenance. A successful candidate is not permission to publish final PyPI automatically.
Phase C — manual release and launch
- Obtain explicit final manual GO for exact SHA, tag, TestPyPI, PyPI, GitHub Release, wheel/sdist hashes and security/consumer findings. Publish only after final signed gates.
- After the 0.6.0 artifact is actually published, change the #90 tutorial from its accurate
0.5.0pin to exact 0.6.0 installed artifact, update all checks/expected versions, re-run and record hashes. Do not recast an earlier run as proof of a new wheel. - docs: build versioned documentation site after PR #52 quickstart #48 GitHub Pages:
github-pagesenvironment must allow stable only, run manual workflow, verify actual live URL, search/mobile/links, versioned 0.5.0 snapshot and 0.6.0 site. Can technically be configured earlier, but must be verifiably live before a website link enters any campaign. - Activate launch(0.6): first coordinated adoption campaign, public docs and verified migration demos #88 rollout only after successful publication and content verification. Focus on real
fastapi-cache2users and case-study evidence, not feature-count claims or synthetic popularity. See competitive audit/positioning.
Explicit exclusions
No rewrite of default cache keys, mandatory events/telemetry, distributed locks, tags, single-flight, rate limiting, circuit breaking, byte-level InMemory cap or automatic multi-tenant isolation. Experimental Memcached/DynamoDB and unsupported Redis Cluster clearing must be labeled accurately.
Release owner must record explicit date, tested SHA, artifact hashes, independent reviewer and final GO/NO-GO in this issue before any tag/publish action.
Activity