Bump Lambda to 4096 MB and cap Node.js heap to prevent OOM

CLAUDE.md

@@ -123,27 +123,26 @@ assert find_test_files("foo.ts", all_files, None) == ["foo.test.ts"]
 2. `scripts/git/list_changed_files.sh` — store for staging reference
 3. Verify not on main: `git branch --show-current`
 4. `git fetch origin main && git merge origin/main`
-5. Stage only YOUR files (**NEVER `git add .`**). Other sessions may have changes. If unsure, check content.
-6. `git commit -m "descriptive message"`
+5. `git commit -m "descriptive message"` — user has already run `git add` before saying "lgtm"
    - Pre-commit hook runs automatically (see `scripts/git/pre_commit_hook.sh`): pip-freeze, generate-types, black, ruff, print/logging checks, then pylint + pyright + pytest concurrently
    - Install: `ln -sf ../../scripts/git/pre_commit_hook.sh .git/hooks/pre-commit`
    - **If hooks fail**: fix, re-stage, commit again. Don't stage other sessions' files.
    - **`--no-verify`** only for trivial non-code changes
    - Unused mock params: `# pyright: reportUnusedVariable=false` at top
    - NO co-author lines or `[skip ci]`
-7. Check for existing PR: `gh pr list --head $(git branch --show-current) --state open` — if exists, **STOP and ask**
-8. `git push`
-9. `gh pr create --title "PR title" --assignee @me` — create PR immediately, no body
-10. Check recent posts: `scripts/git/recent_social_posts.sh gitauto` and `scripts/git/recent_social_posts.sh wes`
-11. `gh pr edit <number> --body "..."` — add summary and social posts after checking recent posts
+6. Check for existing PR: `gh pr list --head $(git branch --show-current) --state open` — if exists, **STOP and ask**
+7. `git push`
+8. `gh pr create --title "PR title" --body "" --assignee @me` — create PR immediately, no body
+9. Check recent posts: `scripts/git/recent_social_posts.sh gitauto` and `scripts/git/recent_social_posts.sh wes`
+10. `gh pr edit <number> --body "..."` — add summary and social posts after checking recent posts
     - Technical, descriptive title. **No `## Test plan`**.
     - **Two posts** (last section, customer-facing only): GitAuto (changelog) + Wes (personal voice, don't emphasize "GitAuto")
     - Format: `## Social Media Post (GitAuto)` and `## Social Media Post (Wes)` headers (parsed by `extract-social-posts.js`)
     - **GitAuto post**: Changelog format — one-liner headline + change bullets. No storytelling.
     - **Wes post**: Honest stories. Vary openers — check recent posts first.
     - Guidelines: No em dashes (—). Under 280 chars. No marketing keywords. No negative framing. No internal names. No small numbers — use relative language.
-12. If Sentry issue: `python3 scripts/sentry/get_issue.py AGENT-XXX` then `python3 scripts/sentry/resolve_issue.py AGENT-XXX ...`
-13. **Blog post** in `../website/app/blog/posts/`:
+11. If Sentry issue: `python3 scripts/sentry/get_issue.py AGENT-XXX` then `python3 scripts/sentry/resolve_issue.py AGENT-XXX ...`
+12. **Blog post** in `../website/app/blog/posts/`:
     - `YYYY-MM-DD-kebab-case-title.mdx`. Universal dev lesson, not GitAuto internals (exception: deep technical content).
     - **Skip if lesson is thin** — argue back if no real insight.
     - `metadata.title`: **34-44 chars** (layout appends `- GitAuto Blog` for 50-60 total). Verify no duplicate slug.
@@ -173,7 +172,7 @@ assert find_test_files("foo.ts", all_files, None) == ["foo.test.ts"]
       - Unsplash API: `source .env && curl "https://api.unsplash.com/search/photos?query=QUERY&orientation=landscape&client_id=$UNSPLASH_ACCESS_KEY"`, download with `?w=1200&h=630&fit=crop&crop=entropy`
       - Convert to PNG: `sips -s format png downloaded.jpg --out ../website/public/og/blog/{slug}.png`
       - Dev.to crops to 1000x420 — keep important content centered.
-14. **Docs page** in `../website/app/docs/`: Create new or update existing. Browse for best-fit category. New pages: 3 files (`page.tsx`, `layout.tsx`, `jsonld.ts`).
+13. **Docs page** in `../website/app/docs/`: Create new or update existing. Browse for best-fit category. New pages: 3 files (`page.tsx`, `layout.tsx`, `jsonld.ts`).
 
 ## CRITICAL: Fixing Foxquilt PRs
 

infrastructure/deploy-lambda.yml

@@ -129,8 +129,8 @@ Resources:
       PackageType: Image
       Code:
         ImageUri: !Ref ECRImageUri
-      # 3072 MB to prevent OOM (2048 MB hit limit during agent fix loops)
-      MemorySize: 3072
+      # Python (~800 MB) + Node.js/Jest (~2 GB) + mongod (~500 MB) + OS (~100 MB)
+      MemorySize: 4096
       EphemeralStorage:
         Size: 2048
       Timeout: 900

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "GitAuto"
-version = "1.6.1"
+version = "1.6.3"
 requires-python = ">=3.14"
 dependencies = [
     "annotated-doc==0.0.4",

services/jest/run_js_ts_test.py

@@ -18,6 +18,7 @@
 from utils.error.handle_exceptions import handle_exceptions
 from utils.logging.logging_config import logger
 from utils.logs.minimize_jest_test_logs import minimize_jest_test_logs
+from utils.memory.is_lambda_oom_approaching import NODE_MAX_OLD_SPACE_SIZE_MB
 from utils.process.kill_processes_by_name import kill_processes_by_name
 
 
@@ -95,6 +96,9 @@ async def run_js_ts_test(
                 mongodb_server_version
             )
 
+    # Cap Node.js heap so it OOMs gracefully (catchable JS error) instead of killing the Lambda
+    env["NODE_OPTIONS"] = f"--max-old-space-size={NODE_MAX_OLD_SPACE_SIZE_MB}"
+
     # Kill any lingering mongod processes from previous verify_task_is_complete calls.
     # MongoMemoryServer uses a fixed port (e.g. 34213) hardcoded in customer tests.
     # If a previous jest run's globalTeardown didn't fully clean up, the stale mongod causes "namespace already exists, but with different options" errors.

services/jest/test_run_js_ts_test.py

@@ -5,6 +5,7 @@
 import pytest
 
 from services.jest.run_js_ts_test import run_js_ts_test
+from utils.memory.is_lambda_oom_approaching import NODE_MAX_OLD_SPACE_SIZE_MB
 
 
 @pytest.mark.asyncio
@@ -403,6 +404,28 @@ async def test_run_js_ts_test_sets_mongoms_md5_check_false(
     assert env["MONGOMS_MD5_CHECK"] == "false"
 
 
+@pytest.mark.asyncio
+@patch("services.jest.run_js_ts_test.subprocess.run")
+@patch("services.jest.run_js_ts_test.os.path.exists")
+async def test_run_js_ts_test_sets_node_max_old_space_size(
+    mock_exists, mock_subprocess, create_test_base_args
+):
+    mock_exists.return_value = True
+    mock_subprocess.return_value = MagicMock(returncode=0, stdout="", stderr="")
+
+    base_args = create_test_base_args(clone_dir="/tmp/clone")
+    await run_js_ts_test(
+        base_args=base_args,
+        test_file_paths=["src/index.test.ts"],
+        source_file_paths=[],
+        impl_file_to_collect_coverage_from="",
+    )
+
+    call_kwargs = mock_subprocess.call_args_list[0].kwargs
+    env = call_kwargs["env"]
+    assert env["NODE_OPTIONS"] == f"--max-old-space-size={NODE_MAX_OLD_SPACE_SIZE_MB}"
+
+
 # Real Jest output captured from foxden-rating-quoting-backend on 2026-03-23.
 # Jest writes PASS/FAIL to stderr, coverage tables to stdout.
 # This was the root cause of a bug where run_js_ts_test checked only result.stdout.

utils/memory/is_lambda_oom_approaching.py

@@ -1,9 +1,16 @@
+# Standard imports
+import os
+
 # Local imports
 from utils.error.handle_exceptions import handle_exceptions
 from utils.memory.get_rss_mb import get_rss_mb
 
-# Must match MemorySize in infrastructure/deploy-lambda.yml
-LAMBDA_MEMORY_MB = 3072
+# On Lambda, AWS sets this env var automatically from MemorySize in infrastructure/deploy-lambda.yml
+# https://docs.aws.amazon.com/lambda/latest/dg/configuration-envvars.html
+LAMBDA_MEMORY_MB = int(os.environ.get("AWS_LAMBDA_FUNCTION_MEMORY_SIZE", "4096"))
+
+# Reserve 1.5 GB for Python (~800 MB) + mongod (~500 MB) + OS (~100 MB) so Node.js OOMs gracefully (catchable) before Lambda is killed
+NODE_MAX_OLD_SPACE_SIZE_MB = LAMBDA_MEMORY_MB - 1536
 
 
 @handle_exceptions(default_return_value=(False, 0), raise_on_error=False)

utils/memory/test_is_lambda_oom_approaching.py

@@ -1,16 +1,19 @@
 # Standard imports
+import importlib
 from unittest.mock import MagicMock, patch
 
 # Third-party imports
 import pytest
 
 # Local imports
+import utils.memory.is_lambda_oom_approaching as oom_module
 from utils.memory.is_lambda_oom_approaching import (
     LAMBDA_MEMORY_MB,
+    NODE_MAX_OLD_SPACE_SIZE_MB,
     is_lambda_oom_approaching,
 )
 
-# 90% of 3072 = 2764.8 MB
+# 90% of 4096 = 3686.4 MB
 THRESHOLD_MB = LAMBDA_MEMORY_MB * 90 / 100
 
 
@@ -35,12 +38,12 @@ def test_below_threshold_linux(self, mock_resource):
     @patch("utils.memory.get_rss_mb._IS_MACOS", False)
     @patch("utils.memory.get_rss_mb.resource")
     def test_above_threshold_linux(self, mock_resource):
-        # 2800 MB in KB (above 2764.8 MB threshold)
-        mock_resource.getrusage.return_value = _mock_rusage(2800 * 1024)
+        # 3700 MB in KB (above 3686.4 MB threshold)
+        mock_resource.getrusage.return_value = _mock_rusage(3700 * 1024)
         mock_resource.RUSAGE_SELF = 0
         is_approaching, used_mb = is_lambda_oom_approaching()
         assert is_approaching is True
-        assert used_mb == 2800.0
+        assert used_mb == 3700.0
 
     @patch("utils.memory.get_rss_mb._IS_MACOS", True)
     @patch("utils.memory.get_rss_mb.resource")
@@ -55,42 +58,42 @@ def test_below_threshold_macos(self, mock_resource):
     @patch("utils.memory.get_rss_mb._IS_MACOS", True)
     @patch("utils.memory.get_rss_mb.resource")
     def test_above_threshold_macos(self, mock_resource):
-        # 2800 MB in bytes (macOS units)
-        mock_resource.getrusage.return_value = _mock_rusage(2800 * 1024 * 1024)
+        # 3700 MB in bytes (macOS units)
+        mock_resource.getrusage.return_value = _mock_rusage(3700 * 1024 * 1024)
         mock_resource.RUSAGE_SELF = 0
         is_approaching, used_mb = is_lambda_oom_approaching()
         assert is_approaching is True
-        assert used_mb == 2800.0
+        assert used_mb == 3700.0
 
     @patch("utils.memory.get_rss_mb._IS_MACOS", False)
     @patch("utils.memory.get_rss_mb.resource")
     def test_exact_threshold_not_approaching(self, mock_resource):
-        # Exactly at threshold (2764.8 MB) - use 2764 MB, not greater, so False
-        mock_resource.getrusage.return_value = _mock_rusage(2764 * 1024)
+        # Exactly at threshold (3686.4 MB) - use 3686 MB, not greater, so False
+        mock_resource.getrusage.return_value = _mock_rusage(3686 * 1024)
         mock_resource.RUSAGE_SELF = 0
         is_approaching, used_mb = is_lambda_oom_approaching()
         assert is_approaching is False
-        assert used_mb == 2764.0
+        assert used_mb == 3686.0
 
     @patch("utils.memory.get_rss_mb._IS_MACOS", False)
     @patch("utils.memory.get_rss_mb.resource")
     def test_just_above_threshold(self, mock_resource):
-        # 2765 MB - just above 2764.8 threshold
-        mock_resource.getrusage.return_value = _mock_rusage(2765 * 1024)
+        # 3687 MB - just above 3686.4 threshold
+        mock_resource.getrusage.return_value = _mock_rusage(3687 * 1024)
         mock_resource.RUSAGE_SELF = 0
         is_approaching, used_mb = is_lambda_oom_approaching()
         assert is_approaching is True
-        assert used_mb == 2765.0
+        assert used_mb == 3687.0
 
     @pytest.mark.parametrize(
         "used_kb, expected_approaching",
         [
             (0, False),
             (512 * 1024, False),
             (1024 * 1024, False),
-            (2764 * 1024, False),
-            (2765 * 1024, True),
-            (3072 * 1024, True),
+            (3686 * 1024, False),
+            (3687 * 1024, True),
+            (4096 * 1024, True),
         ],
         ids=["zero", "512mb", "1024mb", "at_threshold", "above_threshold", "at_limit"],
     )
@@ -105,5 +108,16 @@ def test_parametrized_linux(self, mock_resource, used_kb, expected_approaching):
     def test_has_handle_exceptions_decorator(self):
         assert hasattr(is_lambda_oom_approaching, "__wrapped__")
 
-    def test_constant_matches_infrastructure(self):
-        assert LAMBDA_MEMORY_MB == 3072
+    def test_default_matches_infrastructure(self):
+        assert LAMBDA_MEMORY_MB == 4096
+
+    @patch.dict("os.environ", {"AWS_LAMBDA_FUNCTION_MEMORY_SIZE": "4096"})
+    def test_reads_memory_from_env_var(self):
+        importlib.reload(oom_module)
+        assert oom_module.LAMBDA_MEMORY_MB == 4096
+        assert oom_module.NODE_MAX_OLD_SPACE_SIZE_MB == 4096 - 1536
+        # Restore default for other tests
+        importlib.reload(oom_module)
+
+    def test_node_max_old_space_size_derivation(self):
+        assert NODE_MAX_OLD_SPACE_SIZE_MB == LAMBDA_MEMORY_MB - 1536