Conversation
Zero-dependency bin (node builtins; Redis via the package's own adapters). init: Next-version detection (15 → ISR only, 16.1+ → both handlers), shim generation with the detected client type, next.config guidance (guidance only — the file is never auto-edited), .env.example append, idempotent marker-based rules injection into CLAUDE.md/AGENTS.md, optional local skill install. doctor: ping/latency, next-* key bucket counts, write/read round-trip; stable [ok]/[warn]/[action]/[error] prefixes for agent parsing. Phase 2 of the agent-optimization plan. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
seedBuildOutput() reads prerender-manifest.json and seeds App Router routes (html/rsc/meta plus complete PPR segment maps — partial segment sets are skipped as unsafe), Pages Router routes, and fetch-cache entries into Redis as the exact records the incremental handler serves (revalidateSec included so soft-SWR works on seeded entries; tags carried via meta headers so invalidation applies). Writes use SET NX so live instances' newer entries always win. Exposed as /seed and a CLI 'seed' command (redis→ioredis fallback). Validated end to end: demo app's real build seeded 15 routes + 1 fetch entry; a cold server's FIRST request returned x-nextjs-cache: HIT. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
updateTags publishes {tags, ts} on <keyPrefix>inval:<ns>; each instance
lazily subscribes on a dedicated duplicate connection (redis@5 duplicate +
subscribe, ioredis duplicate + message events) and merges timestamps into
its local mirror monotonically. Subscription failures retry with a 5s
cooldown and never surface to callers — the refreshTags scan remains the
consistency safety net, so pub/sub only ever buys latency. Cluster clients
(no duplicate-based subscribe wired) disable it with a single warning.
Measured cross-instance propagation over real Redis: 3ms on both client
types, vs ~2.1s end-to-end with scan-only propagation.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
scripts/cluster-test-env.sh spins up a local 3-master cluster (no docker); tests/cluster covers set-with-tags and revalidate-hard Lua under CROSSSLOT constraints with hashTag slotting, refreshTags via the per-master scanIterator, cross-handler soft propagation, tagPubSub's graceful cluster fallback, and the ISR handler round-trip. Wired as npm run test:cluster and a dedicated CI job (apt redis-server, no containers). README cluster claims upgraded from 'unit-tested only' to e2e-tested. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…hed) Standalone MCP server (stdio, runs locally against the user's Redis — no hosting) giving agents structured cache operations: cache_health, cache_search, cache_inspect (record decode incl. compression markers), tag_state (both layers), explain_key, simulate_swr, and invalidate_tag (dry-run unless confirm: true; publishes on the tagPubSub channel when executing). Verified via a real stdio JSON-RPC smoke (7 tools listed) and live tool calls against real Redis (health ping, ISR tag state read, dry-run plan). Separate package so the core keeps zero runtime deps. NOT published yet — first publish requires user action (see release checklist). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… document 0.4 features in skill/api docs Two overlapping gated ops could both pass the active check and open two duplicate subscriber connections, leaking one. An attempting flag makes the subscription attempt single-flight. SKILL.md and docs/api.md gain the 0.4 surface (seed command/entry point, tagPubSub, doctor/init, MCP pointer) so the 0.4.0 tarball ships accurate agent guidance. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
BLOCKER fixes: seeding now skips PPR routes carrying postponed resume state (a seeded copy would freeze the unresolved shell as final HTML); both adapters tear down the duplicate connection when SUBSCRIBE fails (an ACL without channel permissions previously leaked one connected duplicate per 5s retry — ~17k/day toward maxclients) and report connection death via a new onDown callback so pubsub.active un-latches and the retry path re-establishes. Also: seedBuildOutput closes redis@5 clients (dispose on adaptRedisV5 — programmatic callers no longer hang); getImpl folds memTagExp into tag freshness (publisher-clock skew could get a pushed timestamp dropped by the scan rebuild's carry-over); pub/sub message handler prunes the mirrors; CLI seed gains --namespace/--key-prefix/ --hash-tag/--compression plus key-shape output (config mismatches were silently seeding unreadable keys) and error-parity on the ioredis path; ISR readTagStates uses per-key GETs instead of a cross-slot MGET that CROSSSLOT- failed multi-tag entries on Cluster; route-handler outputs get their own seed counter; init writes .mcp.json (registers the MCP server, never clobbers an existing file). release.yml gains a publish_mcp dispatch input using the same trusted-publisher workflow identity. 4 blocker regression tests added (158 unit total). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…tion
HASH_TAG env mirrors the handler option: marker reads/writes, namespace
discovery (braces preserved), and the pubsub channel now match {ns}-braced
key shapes — previously every 'use cache' operation silently missed for
Cluster deployments while reporting success. cache_inspect guards with
STRLEN (>2MB → sizes/TTL only) and bounds zlib output (16MB) against
decompression bombs. Agent-supplied tags are glob-escaped in SCAN patterns;
explain_key parses un-namespaced ISR tags containing colons; invalidate_tag
reports warnings when no use-cache namespace was found and documents its
non-atomic hard mode. Verified with a live braced-marker probe.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The 0.4.0 core roadmap plus agent-optimization Phases 2–3, held as a draft for final approval per the release plan (docs/release-checklist-0.4.0.md).
/seed+ `nextjs-cache-handler seed`): NX-protected inserts of prerendered routes (incl. PPR segment maps), Pages routes, and fetch-cache. Verified on the real demo app: cold server's first request = `x-nextjs-cache: HIT`.Test evidence
Unit 153 · integration 25 (real Redis) · cluster 5 (real 3-master) · real-app seeding e2e · MCP stdio/live smokes. Core stays zero-dependency.
Release plan on approval
See docs/release-checklist-0.4.0.md — merge → two-phase release workflow → npm 0.4.0 via trusted publishing; MCP 0.1.0 first publish needs one manual `npm publish` (new package can't pre-configure trusted publishing).
🤖 Generated with Claude Code