[cachekit-release-bot]

🤖 I have created a release beep boop

0.25.0 (2026-10-27)

⚠ BREAKING CHANGES

• config: cachekit.config.nested.CircuitBreakerConfig.validate(), which DecoratorConfig runs, now raises ConfigurationError, and cachekit.reliability.CircuitBreakerConfig (exported as the top-level cachekit.CircuitBreakerConfig) now raises ValueError, unless each of failure_threshold, success_threshold and half_open_requests is an int or a NumPy integer scalar, not a bool, and >= 1. Before, the decorator config refused a count only when value < 1 and the top-level class only when value <= 0, so these used to pass and are now refused: a float such as 5.0 (in the top-level class also 0.5), a Decimal such as Decimal(5), a Fraction such as Fraction(5) (in the top-level class also Fraction(1, 2)), a bool, a NumPy bool scalar, NaN, inf, a np.timedelta64, and a 0-d or one-element NumPy array; pass an int. recovery_timeout / timeout_seconds must now be an int, a float, a Fraction or a NumPy integer or float scalar, not a bool, finite and > 0; a bool, a NumPy bool scalar, a Decimal, a 0-d or one-element NumPy array, and a np.timedelta64 that is unitless or in ns, M or Y units, which passed before when positive and finite, are now refused (a timedelta in other units, or NaT, used to raise TypeError and now raises the class's error). BackpressureConfig.validate() now refuses a 0-d integer NumPy array, which passed before, for max_concurrent_requests, queue_size and timeout, and an object with __index__ that is not a numbers.Integral for all three.
• cachekit.monitoring.CircuitBreakerProtocol now requires a get_stats() -> dict method instead of the state, failure_count, success_count and last_failure_time properties. The dict holds state (the upper-case state name: "CLOSED", "OPEN" or "HALF_OPEN"), failure_count, success_count and last_failure_time (epoch seconds of the last counted failure, 0.0 before the first). HealthChecker reads a circuit breaker only through get_stats(), and a state it does not recognise now reports UNHEALTHY instead of falling through to HEALTHY. cachekit.reliability.CircuitBreaker already implements the method. BackpressureConfig.queue_size and timeout now take effect. BackpressureConfig.validate(), which DecoratorConfig runs, raises ConfigurationError when max_concurrent_requests or queue_size is not a whole number >= 1 (a bool, a float or NaN included), or when timeout is not an int or float from 0 to threading.TIMEOUT_MAX (None, a str, a bool, a Decimal, a Fraction or NaN included). Those queue_size and timeout values were previously accepted and ignored, and such a max_concurrent_requests was accepted and disabled the limit.

Features

• build: publish a free-threaded CPython 3.15 wheel (LAB-10969) (#638) (c9794b4)
• build: publish CPython 3.15 wheels (LAB-10967) (#633) (2f9a661)
• decorators: opt-in write_behind returns a miss before its L2 write lands (LAB-7161) (#623) (cbf4e85)
• drop Python 3.10 support, require >=3.11 (LAB-10968) (#637) (2911337)

Bug Fixes

• config: name the accepted types in the timeout messages (LAB-10984) (#634) (4188c1c)
• config: refuse non-whole circuit-breaker thresholds and non-numeric timeouts (LAB-10531) (#631) (c87f83b)
• deps: floor filelock at 4.2.1 on Python 3.11+ for pip-audit (LAB-10807) (#629) (234e3e2)
• make: remove the pytest base temp after a passing test run (LAB-11059) (#642) (2e7443e)
• perf: wake the instruction-budget gate's wait so a signal is never slept through (LAB-10970) (#641) (f930482)
• readme: replace the unmeasured performance figures with measured ones (LAB-10638) (#622) (9d00c5e)
• redis: replace deprecated setex with set(ex=) (LAB-11029) (#651) (6b18341)
• redis: run on redis-py 8: invalidation listener and RESP3 error-frame tests (LAB-10979) (fcd7f27)
• security: make security-medium mirror the Security Medium workflow (LAB-10606) (#621) (60c5182)
• serializers: count empty collections toward the nesting bound; the guard's ValueError replaces msgpack's StackError (LAB-6485) (#618) (39320e5)
• six defects in health checks, backpressure wiring and serializer and breaker docs (LAB-10106) (#615) (a46c9c1)

Performance Improvements

• decorators: memoise generated keys and serve L1-only hits on a fast path (LAB-10663) (#625) (75440ff)
• deps: bump cachekit-core to 0.7.2 (LAB-11012) (#654) (0b8aa83)
• deps: update cachekit-core to 0.7.1 (LAB-10985) (#636) (c8ddc7d)
• encryption: precompute the AAD prefix and bind the keyring to the tenant (LAB-7126) (#644) (986ed04)
• encryption: read decrypt ciphertext from any buffer without a bytes() copy (LAB-7127) (#647) (ffa8cc0)

---

This PR was generated with Release Please. See documentation.

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

Run configuration

• Source mode: Auto
• Repository: cachekit-io/cachekit-py
• Branch: main
• File: .coderabbit.yaml
• Commit: 7321f51
• Layers: Repository YAML
• Review profile: ASSERTIVE
• Plan: Advanced
• Run ID: badf4c7a-8352-4822-8c99-72a11cbaf2a2

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!

[github-actions]

Instruction budget: PASS

Head f87e7df2afcf6b3b69c65150febd153b0e3098ac measured against base 7321f51868e8cc5ad8160eccfabdfa926a65e023 on the same runner, on cpython-3.12-x86_64 (3.12.3). Workflow run

Verdict	Path	Head Ir/op	Base Ir/op	Delta
ok	byte_storage_store_40b	2,756	2,756	+0.00%
ok	byte_storage_store_64k	724,751	724,749	+0.00%
ok	file_set	86,986	86,979	+0.01%
ok	l1_hit	13,123	13,124	-0.01%
ok	l2_hit	195,328	196,053	-0.37%
ok	l2_hit_async_metrics	183,020	182,975	+0.02%
ok	minimal_l1_hit	11,983	11,982	+0.01%
ok	miss	232,127	232,192	-0.03%
ok	secure_l1_hit	177,044	177,580	-0.30%
ok	serializer_arrow	2,214,479	2,214,416	+0.00%
ok	serializer_auto	72,528	72,558	-0.04%
ok	serializer_default	64,421	64,304	+0.18%
ok	serializer_default_records	1,482,214	1,482,617	-0.03%
ok	serializer_encrypted	109,159	109,279	-0.11%
ok	serializer_orjson	24,894	24,888	+0.02%

A path fails at +1% or more against the base (+2% for serializer_orjson). This check is not required: a deliberate cost increase is accepted at merge. Method: docs/performance.md, Instruction budgets.