fix: cap cache durations at one year

[lan17]

Summary

• cap cache TTLs at 31,536,000 seconds (a fixed 365-day duration)
• cap invalidation future buffers at 31,536,000,000 milliseconds
• enforce the shared range in static defaults, dynamic policy resolution, cache conversion, high-level invalidation, and the exported Lua protocol
• document the supported range and add exact boundary coverage

Why

DialCache previously accepted any positive safe-integer TTL even though local, Redis, tracked-watermark, and invalidation paths later multiplied or added those values independently. That made rejection timing inconsistent and allowed fallback serialization to happen before an oversized Redis TTL was rejected.

This change gives every duration path one simple supported maximum. Static/default TTLs reject before fallback work, malformed dynamic TTLs preserve the existing invalid_ttl fail-open behavior, and invalidations reject before Redis mutation. Existing inputs within 365 days are unchanged; larger inputs are now explicitly unsupported.

Validation

• Node.js 22.22.0
• corepack pnpm install --frozen-lockfile
• corepack pnpm check (289 unit tests, typecheck, coverage, build, packed consumers)
• corepack pnpm test:integration (77 Redis 6.2, Valkey 8, and Redis Cluster tests)
• git diff --check

Closes#33

[lan17]

Summary

• cap cache TTLs at 31,536,000 seconds (a fixed 365-day duration)
• cap invalidation future buffers at 31,536,000,000 milliseconds
• enforce the shared range in static defaults, dynamic policy resolution, cache conversion, high-level invalidation, and the exported Lua protocol
• document the supported range and add exact boundary coverage

Why

DialCache previously accepted any positive safe-integer TTL even though local, Redis, tracked-watermark, and invalidation paths later multiplied or added those values independently. That made rejection timing inconsistent and allowed fallback serialization to happen before an oversized Redis TTL was rejected.

This change gives every duration path one simple supported maximum. Static/default TTLs reject before fallback work, malformed dynamic TTLs preserve the existing invalid_ttl fail-open behavior, and invalidations reject before Redis mutation. Existing inputs within 365 days are unchanged; larger inputs are now explicitly unsupported.

Validation

• Node.js 22.22.0
• corepack pnpm install --frozen-lockfile
• corepack pnpm check (289 unit tests, typecheck, coverage, build, packed consumers)
• corepack pnpm test:integration (77 Redis 6.2, Valkey 8, and Redis Cluster tests)
• git diff --check

Closes#33

[lan17]

Summary

• cap cache TTLs at 31,536,000 seconds (a fixed 365-day duration)
• cap invalidation future buffers at 31,536,000,000 milliseconds
• enforce the shared range in static defaults, dynamic policy resolution, cache conversion, high-level invalidation, and the exported Lua protocol
• document the supported range and add exact boundary coverage

Why

DialCache previously accepted any positive safe-integer TTL even though local, Redis, tracked-watermark, and invalidation paths later multiplied or added those values independently. That made rejection timing inconsistent and allowed fallback serialization to happen before an oversized Redis TTL was rejected.

This change gives every duration path one simple supported maximum. Static/default TTLs reject before fallback work, malformed dynamic TTLs preserve the existing invalid_ttl fail-open behavior, and invalidations reject before Redis mutation. Existing inputs within 365 days are unchanged; larger inputs are now explicitly unsupported.

Validation

• Node.js 22.22.0
• corepack pnpm install --frozen-lockfile
• corepack pnpm check (289 unit tests, typecheck, coverage, build, packed consumers)
• corepack pnpm test:integration (77 Redis 6.2, Valkey 8, and Redis Cluster tests)
• git diff --check

Closes#33

[lan17]

Summary

• cap cache TTLs at 31,536,000 seconds (a fixed 365-day duration)
• cap invalidation future buffers at 31,536,000,000 milliseconds
• enforce the shared range in static defaults, dynamic policy resolution, cache conversion, high-level invalidation, and the exported Lua protocol
• document the supported range and add exact boundary coverage

Why

DialCache previously accepted any positive safe-integer TTL even though local, Redis, tracked-watermark, and invalidation paths later multiplied or added those values independently. That made rejection timing inconsistent and allowed fallback serialization to happen before an oversized Redis TTL was rejected.

This change gives every duration path one simple supported maximum. Static/default TTLs reject before fallback work, malformed dynamic TTLs preserve the existing invalid_ttl fail-open behavior, and invalidations reject before Redis mutation. Existing inputs within 365 days are unchanged; larger inputs are now explicitly unsupported.

Validation

• Node.js 22.22.0
• corepack pnpm install --frozen-lockfile
• corepack pnpm check (289 unit tests, typecheck, coverage, build, packed consumers)
• corepack pnpm test:integration (77 Redis 6.2, Valkey 8, and Redis Cluster tests)
• git diff --check

Closes#33

[lan17]

Summary

• cap cache TTLs at 31,536,000 seconds (a fixed 365-day duration)
• cap invalidation future buffers at 31,536,000,000 milliseconds
• enforce the shared range in static defaults, dynamic policy resolution, cache conversion, high-level invalidation, and the exported Lua protocol
• document the supported range and add exact boundary coverage

Why

DialCache previously accepted any positive safe-integer TTL even though local, Redis, tracked-watermark, and invalidation paths later multiplied or added those values independently. That made rejection timing inconsistent and allowed fallback serialization to happen before an oversized Redis TTL was rejected.

This change gives every duration path one simple supported maximum. Static/default TTLs reject before fallback work, malformed dynamic TTLs preserve the existing invalid_ttl fail-open behavior, and invalidations reject before Redis mutation. Existing inputs within 365 days are unchanged; larger inputs are now explicitly unsupported.

Validation

• Node.js 22.22.0
• corepack pnpm install --frozen-lockfile
• corepack pnpm check (289 unit tests, typecheck, coverage, build, packed consumers)
• corepack pnpm test:integration (77 Redis 6.2, Valkey 8, and Redis Cluster tests)
• git diff --check

Closes#33

[lan17]

Summary

• cap cache TTLs at 31,536,000 seconds (a fixed 365-day duration)
• cap invalidation future buffers at 31,536,000,000 milliseconds
• enforce the shared range in static defaults, dynamic policy resolution, cache conversion, high-level invalidation, and the exported Lua protocol
• document the supported range and add exact boundary coverage

Why

DialCache previously accepted any positive safe-integer TTL even though local, Redis, tracked-watermark, and invalidation paths later multiplied or added those values independently. That made rejection timing inconsistent and allowed fallback serialization to happen before an oversized Redis TTL was rejected.

This change gives every duration path one simple supported maximum. Static/default TTLs reject before fallback work, malformed dynamic TTLs preserve the existing invalid_ttl fail-open behavior, and invalidations reject before Redis mutation. Existing inputs within 365 days are unchanged; larger inputs are now explicitly unsupported.

Validation

• Node.js 22.22.0
• corepack pnpm install --frozen-lockfile
• corepack pnpm check (289 unit tests, typecheck, coverage, build, packed consumers)
• corepack pnpm test:integration (77 Redis 6.2, Valkey 8, and Redis Cluster tests)
• git diff --check

Closes#33

[lan17]

Summary

• cap cache TTLs at 31,536,000 seconds (a fixed 365-day duration)
• cap invalidation future buffers at 31,536,000,000 milliseconds
• enforce the shared range in static defaults, dynamic policy resolution, cache conversion, high-level invalidation, and the exported Lua protocol
• document the supported range and add exact boundary coverage

Why

DialCache previously accepted any positive safe-integer TTL even though local, Redis, tracked-watermark, and invalidation paths later multiplied or added those values independently. That made rejection timing inconsistent and allowed fallback serialization to happen before an oversized Redis TTL was rejected.

This change gives every duration path one simple supported maximum. Static/default TTLs reject before fallback work, malformed dynamic TTLs preserve the existing invalid_ttl fail-open behavior, and invalidations reject before Redis mutation. Existing inputs within 365 days are unchanged; larger inputs are now explicitly unsupported.

Validation

• Node.js 22.22.0
• corepack pnpm install --frozen-lockfile
• corepack pnpm check (289 unit tests, typecheck, coverage, build, packed consumers)
• corepack pnpm test:integration (77 Redis 6.2, Valkey 8, and Redis Cluster tests)
• git diff --check

Closes#33

[lan17]

Summary

• cap cache TTLs at 31,536,000 seconds (a fixed 365-day duration)
• cap invalidation future buffers at 31,536,000,000 milliseconds
• enforce the shared range in static defaults, dynamic policy resolution, cache conversion, high-level invalidation, and the exported Lua protocol
• document the supported range and add exact boundary coverage

Why

DialCache previously accepted any positive safe-integer TTL even though local, Redis, tracked-watermark, and invalidation paths later multiplied or added those values independently. That made rejection timing inconsistent and allowed fallback serialization to happen before an oversized Redis TTL was rejected.

This change gives every duration path one simple supported maximum. Static/default TTLs reject before fallback work, malformed dynamic TTLs preserve the existing invalid_ttl fail-open behavior, and invalidations reject before Redis mutation. Existing inputs within 365 days are unchanged; larger inputs are now explicitly unsupported.

Validation

• Node.js 22.22.0
• corepack pnpm install --frozen-lockfile
• corepack pnpm check (289 unit tests, typecheck, coverage, build, packed consumers)
• corepack pnpm test:integration (77 Redis 6.2, Valkey 8, and Redis Cluster tests)
• git diff --check

Closes#33