[diegosouzapw]

Problem

The chatCore memory-pressure guard returned 503 Service temporarily unavailable due to resource pressure once heapUsed passed a hardcoded 200MB. The app's own working set is ~260MB, so on the default 512MB heap — and on the no-cap PM2 deploy (V8 ceiling ~2240MB) — the guard rejected every request after warmup. Observed live on the Local VPS: all connections failing, surfaced to clients as [anthropic/claude-haiku-4.5] Service temporarily unavailable....

Fix

Derive the threshold from the live V8 heap ceiling (v8.getHeapStatistics().heap_size_limit) at 85%, floored at 400MB, so it tracks --max-old-space-size across 1GB / 2GB / large VPS and always clears the runtime baseline. Explicit HEAP_PRESSURE_THRESHOLD_MB still overrides.

VPS	OMNIROUTE_MEMORY_MB	heap ceiling	auto threshold
1GB/1core	512 (default)	~704	598
2GB	1024	~1216	1034
4GB	2048	~2240	1904
no-cap (PM2)	—	2240	1904

Changes

• open-sse/utils/heapPressure.ts — pure computeHeapPressureThresholdMb + resolved constant
• chatCore.ts — import the constant instead of the inline parse-or-200
• .env.example — document auto-calibration + per-tier OMNIROUTE_MEMORY_MB guidance (incl. the container/cgroup caveat)
• tests — heap-pressure.test.ts (full matrix) + replace the obsolete "defaults to 200" assertion with one tied to the live production constant

13/13 heap tests pass; typecheck / lint / cycles clean.

[gemini-code-assist]

Code Review

This pull request replaces the fixed 200MB heap-pressure threshold with an auto-calibrating mechanism that dynamically sets the threshold to 85% of the V8 heap ceiling (with a 400MB floor), preventing premature 503 rejections. The changes include a new utility 'heapPressure.ts', updates to 'chatCore.ts', and corresponding unit tests. The review feedback correctly points out that using 'Number(override)' to parse the environment variable is less robust than 'parseInt' because it fails on values with unit suffixes like '512mb', and suggests a backward-compatible fix.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Using Number(override) to parse the environment variable is less robust than parseInt and changes the previous behavior in chatCore.ts (which used parseInt). Specifically, Number will fail (returning NaN and falling back to auto-calibration) on values with common unit suffixes like "400MB" or "512mb". Using parseInt with a fallback to typeof override === 'number' maintains backward compatibility and gracefully handles trailing non-numeric characters.

Suggested change

	const explicit = Number(override);
	if (Number.isFinite(explicit) && explicit > 0) return Math.floor(explicit);
	const explicit = typeof override === "number" ? override : parseInt(String(override || ""), 10);
	if (Number.isFinite(explicit) && explicit > 0) return Math.floor(explicit);

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a514420ce9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Keep the auto threshold below the heap ceiling

For supported low-memory settings, this floor puts the shed threshold above the process's actual V8 ceiling: the launchers accept OMNIROUTE_MEMORY_MB down to 64 MB (bin/cli/commands/serve.mjs and scripts/build/runtime-env.mjs), and e.g. --max-old-space-size=128 reports a heap limit of about 320 MB. In that case computeHeapPressureThresholdMb() returns 400, so chatCore's heapUsed > HEAP_PRESSURE_THRESHOLD_MB check cannot trip before V8 reaches its own limit and hard-OOMs, regressing the guard specifically for low-memory deployments.

Useful? React with 👍 / 👎.