feat(platform-node): a second host over better-sqlite3 + run the conformance TCK against it (spike) - #277
Merged
Merged
Conversation
Stand up @lunora/platform-node — a Node host over better-sqlite3 plus an in-process socket/directory/scheduler registry — and run the existing @lunora/platform and @lunora/shard-engine conformance TCKs against it (38/38 pass). Register NODE_CAPABILITIES honestly rated per feature, wire a `node` target into codegen's PLATFORM_MATRICES, and exercise the fail-closed capability gate against a matrix that is mostly unsupported/emulated. Registering `node` exposed that @lunora/config's driver-registry invariant conflated "codegen can gate this target" with "the CLI can deploy to this target" — narrow that test to the direction that actually matters (every deploy driver needs a matrix; the reverse no longer holds now that a dev/test-only host exists). Every place construction needed something the contracts didn't promise is classified in plans/234-node-host-findings.md, including two latent bugs found by contrast in the existing node:sqlite reference host and an architecture gap where ShardDirectory's jurisdiction contract member has no real caller in the codebase.
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. 🗂️ Base branches to auto review (2)
Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
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. Comment |
Contributor
|
Thank you for following the naming conventions! 🙏 |
Contributor
|
Thank you for confirming the Contributor License Agreement! 🙏 |
prisis
changed the title
…an't fire on a closed DB and timers/handles don't leak Nothing previously closed the better-sqlite3 handle or cleared pending timers: a caller that set a future alarm and then closed the database hit an uncaught TypeError from inside a setTimeout callback (a crash that never touches the caller's own try/catch), and the composition root plus the conformance host's cleanup both leaked the alarm and scheduler timers, keeping the process alive well past teardown. - Guard the alarm callback on `database.open` so a closed connection can never take a `.run()` again. - Give createNodeShardHost/createNodeSchedulerHost a `dispose()` that clears their own timers (and, for the shard host, closes the DB). - Add NodePlatform.close()/[Symbol.dispose] composing both disposers, and route the conformance host's cleanup() through the same path instead of a bare `database.close()`. Co-Authored-By: Claude Opus 4.8 <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.
What
Turns portability from a claim into a tested property:
PLATFORM_MATRICESheld exactly one entry (platformMatrixIds()'s own comment: "both hold exactlycloudflare, which is why nothing has noticed"). This stands up a real second host —@lunora/platform-node(5 contract adapters + composition root overbetter-sqlite3and an in-process socket registry) — and runs the existing conformance TCK against it.Headline result: the contract generalizes
@lunora/platform/conformance: 28/28.@lunora/shard-engine/conformance(OCC/RLS/reactive fan-out): 10/10. Plus 2 composition-root smokes → 40/40.idForstability, the fail-closed capability gate, the tagged fan-out) hold on a structurally different host, not just Cloudflare. There was nothing to "resist patching."Honest
nodecapability matrixNODE_CAPABILITIESadded to@lunora/platform(rates every feature Cloudflare rates — enforced by a same-key-set test), registered asnodein codegen'sPLATFORM_MATRICES. This is the first time the fail-closed gate has run against a mostly-unsupported/emulatedmatrix: verified end-to-end thatgatePlatformFeatures(usage, "node")flipsbrowser/containeroff withplatform_unsupported_featurediagnostics while leavingkv/scheduleron, and that{ "target": "node" }resolves through the real registry (noplatform_unknown_target).Findings (the deliverable) —
plans/234-node-host-findings.md, 7 classifiednode:sqlitereference host, found by contrast: a write-onlydurableAttachmentsmap (dead code hidden by a blanket lint exemption) and a text-sniffing ("select") read/write heuristic weaker thanbetter-sqlite3'sStatement.reader.ShardDirectory.jurisdictionhas zero real callers (five packages reimplement the check against the raw CF type instead); "durable survives host recycling" doesn't distinguish in-process recycle from process restart;ShardKvStore.put's "structured-clonable" pins no wire format (three hosts, three serializers).databaseSizehas no conformance coverage on any host.Two edits flagged for your judgment (both documented)
scripts/api-snapshot.js:platform-nodeexcluded from the API-surface guard until it graduates past spike stage (nolunora dev/deploy wiring yet, surface still moving) — alongside the other experimental packages. api:check is 43/43 with this.packages/configinvariant test: registeringnodebroke a strict-equality invariant (deployTargetIds() === platformMatrixIds()) that only ever held because both were[cloudflare]. Relaxed to the one dangerous direction — every deploy driver must have a matrix (else codegen emits an un-gated surface) — while allowing the benign codegen-gateable-but-not-yet-deployable case thatnodelegitimately is. Reasoning is inline + in the findings doc.Verification
platform-node 40, platform 11, codegen 11, config 16; full regression across the 6 affected packages +
api:check(43/43),dist:check,lint:package-json, prettier, eslint all green. Out of scope (follow-ups):lunora devwiring, a real deploy driver.Plan:
plans/234-*.md.🤖 Generated with Claude Code