perf: Cloud Run CPU optimization, server-side pagination, GCS cache by MarkusNeusinger · Pull Request #4710 · MarkusNeusinger/anyplot

MarkusNeusinger · 2026-03-08T20:52:34Z

Summary

Cloud Run CPU 2→1: Both frontend (nginx) and backend (uvicorn async) don't need 2 cores. Backend upgraded to gen2 execution environment for better CPU/network performance.
Server-side pagination: /plots/filter now accepts limit and offset query params. Counts and totals are still computed from all filtered images. Fully backward-compatible — no params = all images.
GCS Cache-Control 1h→1d: All gsutil cp commands in impl-generate, impl-merge, and impl-repair workflows now set Cache-Control: public, max-age=86400.

Changed files

File	Change
`app/cloudbuild.yaml`	CPU 2→1
`api/cloudbuild.yaml`	CPU 2→1, add `--execution-environment=gen2`
`api/schemas.py`	Add `offset`/`limit` fields to `FilteredPlotsResponse`
`api/routers/plots.py`	Add pagination query params, cache key update, slice logic
`tests/unit/api/test_routers.py`	4 new pagination tests + cached mock update
`.github/workflows/impl-generate.yml`	Cache-Control header on gsutil cp
`.github/workflows/impl-merge.yml`	Cache-Control header on gsutil cp
`.github/workflows/impl-repair.yml`	Cache-Control header on gsutil cp

Test plan

uv run ruff check passes on all changed Python files
uv run pytest tests/unit/api/test_routers.py — 101 tests pass
Verify /plots/filter?limit=10&offset=0 returns 10 images with correct total
Verify /plots/filter without params returns all images (backward compat)
Verify Cloud Run deploys successfully with new CPU/gen2 settings

🤖 Generated with Claude Code

… API preconnect - Lazy-load CatalogPage, LegalPage, McpPage, InteractivePage, DebugPage - Switch react-syntax-highlighter to PrismLight with only Python grammar - Add manualChunks to split vendor (React) and MUI into separate cached chunks - Add preconnect hint for api.pyplots.ai Closes #4704 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Added LazyFallback component for better user experience during lazy loading - Updated router configuration to include HydrateFallback for lazy-loaded pages - Improved manual chunking strategy in Vite config for better performance

- Reduce frontend Cloud Run CPU from 2 to 1 (nginx doesn't need 2 cores) - Reduce backend Cloud Run CPU from 2 to 1, enable gen2 execution environment - Add server-side pagination (limit/offset) to /plots/filter endpoint - Increase GCS image Cache-Control from 1h to 1 day (86400s) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

codecov · 2026-03-08T20:54:53Z

Codecov Report

❌ Patch coverage is 80.64516% with 6 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
api/routers/plots.py	82.14%	5 Missing ⚠️
app/src/components/SpecTabs.tsx	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

- Replace .fn attribute access on decorated functions (no longer wrapped) - Use list_tools() instead of removed get_tools() method Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Copilot

Pull request overview

This PR focuses on performance optimizations across the deployment pipeline and runtime: reducing Cloud Run CPU allocation, adding pagination to the plots filtering endpoint, and improving caching behavior for generated assets uploaded to GCS.

Changes:

Reduced Cloud Run CPU allocation for frontend and backend; backend deployment now targets Gen2 execution environment.
Added limit/offset pagination support to /plots/filter (with response schema updates and new unit tests).
Increased GCS object cacheability by setting Cache-Control: public, max-age=86400 during workflow uploads/copies, and introduced frontend bundle-splitting / lazy-loading adjustments.

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
`app/cloudbuild.yaml`	Reduce frontend Cloud Run CPU from 2 → 1.
`api/cloudbuild.yaml`	Reduce backend Cloud Run CPU from 2 → 1 and add `--execution-environment=gen2`.
`api/schemas.py`	Extend `FilteredPlotsResponse` with `offset` and `limit`.
`api/routers/plots.py`	Add `limit`/`offset` query params, pagination slicing, and cache key changes.
`tests/unit/api/test_routers.py`	Add unit tests covering pagination behavior and update cached response mock shape.
`app/vite.config.ts`	Add Rollup `manualChunks` configuration for bundle splitting.
`app/src/router.tsx`	Convert several routes to lazy-loaded route modules with a shared fallback.
`app/src/components/SpecTabs.tsx`	Switch syntax highlighter import to `prism-light` and register Python language explicitly.
`app/index.html`	Add `<link rel="preconnect">` for the API domain.
`.github/workflows/impl-generate.yml`	Upload artifacts with `Cache-Control: public, max-age=86400`.
`.github/workflows/impl-merge.yml`	Promote staging → production with `Cache-Control: public, max-age=86400`.
`.github/workflows/impl-repair.yml`	Upload repair artifacts with `Cache-Control: public, max-age=86400`.

You can also share your feedback on Copilot code review. Take the survey.

Copilot · 2026-03-08T20:56:45Z

+    cache_key = _build_cache_key(filter_groups, offset=offset, limit=limit)
    try:
        cached = get_cache(cache_key)
        if cached:


Including offset/limit in the cache key means each paginated request creates a distinct entry in the in-process TTLCache (maxsize=1000). With real pagination (offset stepping through pages), this can quickly thrash/evict other cache entries and reduce overall cache effectiveness. Consider caching the unpaginated filtered result per filter_groups (or caching just the computed lookups/counts), then apply slicing after a cache hit so pages reuse the same cached payload.

Copilot · 2026-03-08T20:56:45Z

-        return "filter:all"
+        base = "filter:all"
+    else:
+        cache_parts = [f"{g['category']}={','.join(sorted(g['values']))}" for g in filter_groups]


Cache key construction currently depends on the incoming query parameter order because filter_groups is not normalized (groups/categories aren’t sorted before joining). That makes lib=matplotlib&plot=scatter and plot=scatter&lib=matplotlib produce different cache keys for the same logical filter, reducing cache hit rate. Consider sorting a copy of filter_groups by category (and optionally de-duping identical groups) when building the cache key so cache behavior is stable without changing response semantics.

Suggested change

cache_parts = [f"{g['category']}={','.join(sorted(g['values']))}" for g in filter_groups]

# Normalize group ordering so cache keys do not depend on query param order.

normalized_groups = sorted(filter_groups, key=lambda g: g.get("category", ""))

cache_parts = [f"{g['category']}={','.join(sorted(g['values']))}" for g in normalized_groups]

Copilot · 2026-03-08T20:56:46Z

+  build: {
+    rollupOptions: {
+      output: {
+        manualChunks(id) {
+          if (id.includes('node_modules/@mui/')) return 'mui';
+          if (id.includes('node_modules/react/') || id.includes('node_modules/react-dom/') || id.includes('node_modules/react-router-dom/')) return 'vendor';
+        },
+      },


The PR description’s “Changed files” list and summary don’t mention these frontend bundle-splitting changes (manual Rollup chunking) even though they can affect load behavior and caching in production. Please update the PR description (and ideally the test plan) to include these frontend perf changes so reviewers know they’re intentional and can validate them during deploy smoke tests.

Address Copilot review feedback: - Cache the full filtered result, apply pagination after cache hit (avoids cache thrashing with per-page entries) - Sort filter groups by category in cache key so param order doesn't affect cache hits Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

MarkusNeusinger and others added 3 commits March 8, 2026 21:11

Copilot AI review requested due to automatic review settings March 8, 2026 20:52

Copilot started reviewing on behalf of MarkusNeusinger March 8, 2026 20:53 View session

fix(tests): update MCP test_tools for current FastMCP API

989b952

- Replace .fn attribute access on decorated functions (no longer wrapped) - Use list_tools() instead of removed get_tools() method Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Copilot AI reviewed Mar 8, 2026

View reviewed changes

MarkusNeusinger merged commit b242d4f into main Mar 8, 2026
7 of 8 checks passed

MarkusNeusinger deleted the perf/cloud-run-optimization branch March 8, 2026 21:00

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

perf: Cloud Run CPU optimization, server-side pagination, GCS cache#4710

perf: Cloud Run CPU optimization, server-side pagination, GCS cache#4710
MarkusNeusinger merged 5 commits intomainfrom
perf/cloud-run-optimization

MarkusNeusinger commented Mar 8, 2026

Uh oh!

codecov bot commented Mar 8, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Mar 8, 2026

Uh oh!

Copilot AI Mar 8, 2026

Uh oh!

Copilot AI Mar 8, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

-        cache_parts = [f"{g['category']}={','.join(sorted(g['values']))}" for g in filter_groups]
+        # Normalize group ordering so cache keys do not depend on query param order.
+        normalized_groups = sorted(filter_groups, key=lambda g: g.get("category", ""))
+        cache_parts = [f"{g['category']}={','.join(sorted(g['values']))}" for g in normalized_groups]

Conversation

MarkusNeusinger commented Mar 8, 2026

Summary

Changed files

Test plan

Uh oh!

codecov bot commented Mar 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 8, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

codecov bot commented Mar 8, 2026 •

edited

Loading