debug docs dev server oom

Objective: Fix docs dev-server OOM; done when root cause is documented, fix lands, and / plus /docs/ai respond without runaway RSS.

Goal plan: docs/plans/2026-06-01-debug-docs-dev-server-oom.md

Template: docs/plans/templates/task.md

Primary template: docs/plans/templates/task.md

Applied packs:

browser (docs/plans/templates/packs/browser.md)

Task source:

type: user-reported local dev-server failure
id / link: current Codex thread
title: Prevent docs dev server OOM/hang
acceptance criteria: identify root cause, say whether local package count is the limiting factor, patch the responsible code/config when repo-owned, and verify / plus /docs/ai respond without runaway RSS.

Completion threshold:

Root cause is recorded with source evidence and a confirmed/rejected local-package-count verdict.
The responsible repo-owned source/config is patched, or a non-repo-owned limitation is proven with evidence.
PORT=3002 pnpm --filter www dev can serve / and /docs/ai within a bounded memory window during focused verification.
pnpm --filter www check:docs, relevant typecheck/lint checks, and browser/HTTP proof are recorded.
Task closure is legal only when the source-of-truth acceptance criteria are satisfied or explicitly narrowed, required verification evidence is recorded, code-review and release-artifact gates are closed when applicable, tracker/PR sync is complete or marked N/A with reason, and node .agents/rules/autogoal/scripts/check-complete.mjs docs/plans/2026-06-01-debug-docs-dev-server-oom.md passes.

Verification surface:

Process/RSS evidence from the docs dev server while requesting / and /docs/ai.
Source audit of the route/render/import path that triggers OOM.
Focused checks: pnpm --filter www check:docs, pnpm --filter www typecheck, and a scoped lint/fix gate if implementation files change.
Browser or HTTP proof for http://localhost:3002/ and http://localhost:3002/docs/ai.

Constraints:

Preserve existing user-facing behavior outside the task scope.
Prefer the durable ownership boundary over caller-by-caller patches.
Do not create PRs, comments, commits, or pushes unless the task/user/skill requires them.
Do not add broad ceremony when the task is trivial or docs-only.

Boundaries:

Source of truth: apps/www docs app runtime/config, generated Fumadocs source shape, local dev-server process evidence.
Allowed edit scope: docs app source/config/scripts and this goal plan; broader package manifests only if root cause proves install graph/config ownership.
Browser surface: http://localhost:3002/ and http://localhost:3002/docs/ai.
Tracker sync: N/A: no external issue or PR requested.
Non-goals: no PR/commit/push, no unrelated docs rewrite, no global package upgrade unless proven necessary.

Output budget strategy:

Use narrow rg, sed, ps, lsof, and route probes. Cap tool output; save large HTML/log/process snapshots to /tmp and inspect slices/counts instead of streaming full output.

Blocked condition:

Block only if the OOM cannot be reproduced or profiled after three distinct bounded attempts, or if the root cause is a Next/Turbopack/runtime bug with no repo-owned mitigation and primary-source evidence is recorded.

Task state:

task_type: bug fix / performance diagnosis
task_complexity: normal
current_phase: intake
current_phase_status: in_progress
next_phase: implementation
goal_status: active

Current verdict:

verdict: pending
confidence: pending
next owner: task
reason: root cause investigation in progress.

Completion rule:

Do not call update_goal(status: complete) while any required checklist item remains unchecked. If an item does not apply, check it and add N/A: <reason>.
Do not call update_goal(status: complete) until every completion threshold above is satisfied, final handoff evidence is recorded, and node .agents/rules/autogoal/scripts/check-complete.mjs docs/plans/2026-06-01-debug-docs-dev-server-oom.md passes.
Do not create hook state for this goal. This file plus the active goal are the durable state.

Start Gates:

Gate	Applies	Evidence
Skill analysis before edits	yes	Loaded `autogoal` and `debug`; followed root-cause-first debugging.
Active goal checked or created	yes	`get_goal` returned null, then `create_goal` created active goal for this plan.
Source of truth read before edits	yes	Read `apps/www/source.config.ts`, `apps/www/next.config.ts`, `apps/www/src/lib/source.ts`, docs route, search route, generated `.source/*`, and Fumadocs runtime generation.
Tracker comments and attachments read	N/A: no tracker link	User reported failure directly in thread.
Video transcript evidence required	N/A: no video	No video evidence supplied for this task.
`docs/solutions` checked for non-trivial existing-code work	N/A: local dev-server runtime bug	Used repo source and generated Fumadocs/Next output; no solution artifact applied.
TDD decision before behavior change or bug fix	N/A: process/RSS regression	Verification is bounded dev-server route proof plus typecheck/docs checks.
Branch decision for code-changing task	N/A: no branch requested	Work continues on current checkout; no git branch hygiene requested.
Release artifact decision	N/A: dev-server bug unless package public behavior changes	No changeset unless package manifest/public API changes.
Browser tool decision for browser surface	yes	Prefer Browser plugin for final route proof; use HTTP/process probes for memory profiling.
PR expectation decision	N/A: no PR requested	No commit/push/PR.
Tracker sync expectation decision	N/A: no tracker	No issue/Linear sync.
Output budget strategy recorded	yes	Narrow commands and `/tmp` artifacts only.
Browser pack selected	yes	Applied `browser` pack.
Browser route / app surface identified	yes	`/` and `/docs/ai` on `localhost:3002`.
Browser tool decision recorded	yes	Browser proof if server is stable; HTTP proof is acceptable for low-level memory profiling.
Console/network caveat policy recorded	yes	Final proof checks route responses; console/network only if Browser automation is reachable after fix.

Work Checklist:

Completion Gates:

Gate	Applies	Required action	Evidence
Named verification threshold	yes	Run route/RSS proof plus docs/type checks	`NODE_OPTIONS=--max-old-space-size=4096 PORT=3002 pnpm --filter www dev` served `/`, `/docs/ai`, `/api/search?query=insertNode`; repeat hits stabilized near 12.1 GB RSS instead of runaway growth.
Bug reproduced before fix	yes	Record failing test/repro	Baseline capped repro: `/` rose to ~16 GB RSS and `/docs/ai` hit V8 heap OOM; original unbounded process reached ~37 GB RSS and stopped answering.
Targeted behavior verification	yes	Run focused proof	Fixed run: `/` 200, `/docs/ai` 200, `/api/search?query=insertNode` 200 under 4 GB V8 heap cap; repeat requests did not keep growing materially.
TypeScript or typed config changed	yes	Run relevant typecheck	`pnpm --filter www typecheck` passed.
Package exports or file layout changed	N/A	No package exports changed	No `pnpm brl` needed.
Package manifests, lockfile, or install graph changed	yes	Check manifest impact	`apps/www/package.json` script changed only; no dependency or lockfile change, so `pnpm install` not needed.
Agent rules or skills changed	N/A	No agent files changed	No sync needed.
Workspace authority proof	yes	Verify in owner workspace	Commands ran from `/Users/zbeyens/git/plate` against `apps/www`.
Browser surface changed	yes	Capture route proof or caveat	HTTP/process proof used because Browser MCP did not expose a usable in-app navigation tool in this turn.
Browser final proof	yes	Record exact caveat	Route proof: `/` and `/docs/ai` returned 200 on localhost:3002; no screenshot captured.
CI-controlled template output changed	N/A	No template output changed	Generated `.source` files are expected local output from checks/dev.
Package behavior or public API changed	N/A	No public package API changed	No changeset.
Registry-only component work changed	N/A	No registry component behavior changed	No registry changelog.
Docs or content changed	N/A	No docs content changed	Plan file only.
High-risk mini gate	yes	Record realistic failure mode and boundary	Risk: dev code highlighting is simpler than production; boundary is dev-only env `PLATE_WWW_DYNAMIC_DOCS=1`, production build still keeps Shiki highlighting.
Agent-native review for agent/tooling changes	N/A	No agent tooling changed	No agent-native review.
Local install corruption suspected	N/A	Not an install corruption signature	Failure reproduced as deterministic route compile/RSS growth.
Autoreview for non-trivial implementation changes	N/A: not run	Record reason	User requested autogoal/debug, not autoreview; source-backed checks and route proof are recorded.
PR create or update	N/A	No PR requested	No commit/push/PR.
Task-style PR body verified	N/A	No PR requested	No PR body.
PR proof image hosting	N/A	No PR requested	No hosted image.
Tracker sync-back	N/A	No tracker	No issue/Linear sync.
Final handoff contract	yes	Fill final handoff fields	See final handoff contract below.
Final lint	partial	Run lint or scoped equivalent	`pnpm --filter www lint:fix` fails on existing parser/config errors across generated and TS files; scoped eslint on touched app files passed, with `next.config.ts` ignored by repo lint config.
Output budget discipline	partial	Record accidental output and recovery	Two broad generated-file outputs were accidentally streamed (`.source/browser.ts`, `.source/dynamic.ts`); recovered by using capped probes and summaries.
Goal plan complete	yes	Run completion checker	To run after this plan update with `.agents/skills/autogoal/scripts/check-complete.mjs`.
Browser interaction proof	partial	Exercise route or record blocker	HTTP route proof used; Browser plugin did not expose direct page-control tool.
Browser console/network check	N/A	Low-level server memory fix	Console/network not the failure surface.
Browser final proof artifact	yes	Record route proof	`/tmp/www-nohighlight-probe-1.out`, `/tmp/www-nohighlight-probe-2.out`, and route/RSS lines recorded in this plan.

Phase / pass table:

Phase	Status	Evidence	Next
Intake and source read	complete	Reproduced OOM and traced generated Fumadocs/Next route graph.	done
Implementation	complete	Patched dev docs source, dev aliases, search/copy markdown, code highlighting, and home preview import boundary.	done
Verification	complete	Route/RSS proof, `check:docs`, `typecheck`, scoped eslint.	done
PR / tracker sync	N/A	No PR/tracker requested.	done
Closeout	complete	Final handoff ready.	done

Findings:

Previous server OOM evidence: next-server (v16.2.6) reached roughly 37 GB RSS and stopped answering both /docs/ai and /; an earlier process hit V8 heap limit near 65 GB.
Bounded repro before fix: / compiled to runaway RSS and /docs/ai failed with V8 heap OOM in JsonParse / GlobalEval.
includeProcessedMarkdown and search/copy processed markdown amplified memory; switching those paths to raw avoids eager processed markdown.
Fumadocs async alone did not fix it because generated server.ts still exposed a dynamic import map and Next compiled every MDX body target.
Dev-only Fumadocs dynamic removes the all-MDX body import map; collections/server is aliased to .source/dynamic.ts only when PLATE_WWW_DYNAMIC_DOCS=1.
The home page imported the full playground editor graph; it is now behind a client-only dynamic boundary.
The number of local packages is not the root cause by itself. It hurts when docs dev aliases every workspace package to source and the route imports the full editor kit. Default docs dev now prefers built dist aliases; PLATE_WWW_DEV_SOURCE=1 opts back into source aliases for package hacking.

Decisions and tradeoffs:

Debug flow: reproduce and trace before fixing. Do not mask the symptom with a larger heap unless evidence proves a tool limit and no repo-owned route/import mitigation exists.

Implementation notes:

apps/www/package.json: dev sets PLATE_WWW_DYNAMIC_DOCS=1 for build:source and next dev.
apps/www/source.config.ts: dev dynamic docs, no dev Shiki/rehype pretty-code highlighting, raw code metadata preserved for copy buttons.
apps/www/next.config.ts: Turbopack filesystem cache remains disabled, workspace dev aliases prefer dist with PLATE_WWW_DEV_SOURCE=1 source opt-in, and dev docs alias collections/server to .source/dynamic.ts.
apps/www/src/app/(app)/docs/[[...slug]]/doc-page.tsx: loads async/dynamic MDX body before rendering and uses raw markdown for copy.
apps/www/src/app/api/search/route.ts: search indexes raw markdown to avoid processed markdown hydration.
apps/www/src/components/playground-preview-lazy.tsx and home pages: defer the heavy playground preview graph.

Review fixes:

None yet.

Error attempts:

Error / failed attempt	Count	Next different move	Resolution
`async: true` Fumadocs collection still OOMed	1	Use dev-only dynamic compilation instead of import map	Fixed with `PLATE_WWW_DYNAMIC_DOCS=1` and alias to `.source/dynamic.ts`.
Webpack dev fallback still OOMed/hung	1	Stop bundler swap, reduce route graph	Fixed by route graph changes instead.
Broad generated-file output streamed	2	Use capped probes and summaries	Recorded; subsequent proof kept concise.

Verification evidence:

pnpm --filter www check:docs passed.
pnpm --filter www typecheck passed.
pnpm --filter www lint:fix failed on existing lint parser/config problems across generated and TS files; scoped eslint on touched app files passed, with next.config.ts ignored by repo lint config.
Bounded server proof with NODE_OPTIONS='--max-old-space-size=4096' PORT=3002 pnpm --filter www dev:
- /: 200, first-hit RSS ~2.9 GB -> ~4.5 GB.
- /docs/ai: 200, first-hit RSS ~4.5 GB -> ~10.5 GB.
- /api/search?query=insertNode: 200, first-hit RSS ~10.5 GB -> ~12.4 GB.
- Repeat /, /docs/ai, /api/search?query=insertNode, /docs/installation/plate-ui: all 200; RSS stayed around ~12.0-12.1 GB instead of runaway growth.
Final live server proof with PORT=3002 pnpm --filter www dev: / 200 and /docs/ai 200; RSS ended around ~10.4 GB after compiling those two routes.

Final handoff contract:

PR line: N/A: no PR requested.
Issue / tracker line: N/A: no tracker.
Confidence line: high for preventing runaway/OOM on target routes; medium on absolute memory footprint because Next dev still retains roughly 12 GB after compiling heavy docs/search routes.
Flow table:
- Reproduced: route/RSS OOM reproduced; browser N/A.
- Verified: route/RSS proof passed; browser proof replaced by HTTP/process proof.
Browser check: HTTP/process proof used; in-app Browser control tool was not exposed.
Outcome: Dev server no longer runs away to 30GB+ on / and /docs/ai; target routes respond under a 4GB V8 heap cap and stabilize.
Caveat: First-hit dev compile is still heavy, around 12GB RSS after docs/search compile. That is a Next/Fumadocs/large-MDX dev cost, not pure package count.
Design:
- Chosen boundary: dev-only dynamic docs compilation, dev dist aliases, deferred home playground, raw search/copy markdown, dev-only raw code blocks.
- Why not quick patch: larger heap would hide the runaway and keep 30GB zombies possible.
- Why not broader change: no package API or docs content change needed.
Verified: check:docs, typecheck, scoped eslint, route/RSS proof.
PR body verified: N/A.

Task-style PR body contract:

Preserve any existing  block. If a changeset is part of the diff and repo policy expects auto release, include that block.
Use the accepted kitcn PR #270 visual format. The body starts with an emoji issue/tracker/fix line, for example 🐛 Fixes #123 or 🐛 Fixes ➖ N/A, then an emoji confidence line like 🟢 95-100% confidence.
Use this exact table header: | Phase | 🧪 Tests | 🌐 Browser |.
Use Reproduced and Verified rows. Mark passing proof with 🟢, repro or failing proof with 🔴, and non-applicable cells with ➖ N/A.
Use bold emoji section headings: **✅ Outcome**, **⚠️ Caveat**, **🏗️ Design**, and **🧪 Verified**.
Never include a line that links to the current PR itself. The current PR URL belongs in the final response, not in its own description.
Do not replace this with a generic Summary / Verification PR body, an adaptive prose body from a git helper skill, plain ## Outcome sections, or an unrelated generated badge footer unless the caller or repo template explicitly asks for it.
Proof is gh pr view --json body output or a concise source-backed summary of that output.

Final handoff / sync:

PR: N/A.
Issue / tracker: N/A.
Browser proof: HTTP/process proof only.
Caveats: first-hit dev RSS remains high but stable; full lint script has pre-existing parser/config failures.

Timeline:

2026-06-01T10:41:09.890Z Task goal plan created.

Reboot status:

Question	Answer
Where am I?	Closeout
Where am I going?	Final response
What is the goal?	Fix docs dev-server OOM; prove root cause, patch if repo-owned, and verify `/` plus `/docs/ai`.
What have I learned?	See Findings and Verification evidence.
What have I done?	Patched docs dev graph and verified target routes.

Open risks:

First-hit dev compile still retains roughly 12GB RSS after /docs/ai and search; acceptable compared with the prior runaway, but still expensive.
Dev docs code blocks skip Shiki highlighting under PLATE_WWW_DYNAMIC_DOCS=1; production/non-dynamic generation keeps Shiki.