perf(share): cuts share card generation cost

[seferturan]

🎶 Notes 🎶

• Cold share card generation was dominated by two things, measured per stage
  • satori render was 60-72% of the request
  • the fonts were fetched from cdn-sveltekit-og.ethercorps.io on every cold generation, 6s to 19s in Sentry over the past week, which is long enough for a scraper to give up before the card renders
• Loads the fonts from R2 instead
  • reads assets/fonts/NotoSans-Regular.ttf and assets/fonts/NotoSans-Bold.ttf from the walter bucket the endpoint already binds, memoised per isolate
  • falls back to the sveltekit-og defaults when the bucket or an object is missing, so dev keeps working and a bad upload degrades instead of breaking
  • caches successes only, so a failed read retries instead of poisoning the isolate
  • fonts are already uploaded and verified reachable
• Collapses the poster shadow to a single layer
  • satori render, averaged over 3 samples per variant: open-graph 783ms to 538ms, feed 1413ms to 764ms. 31% and 46% off
  • cost tracks the area the blur covers rather than the number of layers, so alpha is free
  • ⚠️ keep the alpha low. A single layer at 0.5 concentrates the whole shadow into a visible slab with a hard edge along the poster radius, where five faint layers blended into a smooth falloff. 0.22 over a 32px blur reads like the original
• Resolves the poster and the fonts in parallel
  • neither depends on the other, and neither can reject, so Promise.all does not change the failure behaviour

🚀 Deploying 🚀

• No cache purge needed this time
  • cards already in R2 keep the old shadow, new ones get the new one
  • minor cosmetic drift only, sorts itself out once we prune automatically

🤔 Later 🤔

• Automatic pruning of cached cards, tackled separately
• In-flight dedupe. When several scrapers hit the same new url at once they each render it from scratch. Worth doing now that render is the bulk of the cost

[deepsource-io]

DeepSource Code Review

We reviewed changes in 4056525...68d9723 on this pull request. Below is the summary for the review, and you can see the individual issues we found as inline review comments.

See full review on DeepSource ↗

PR Report Card

Overall Grade

Security   Reliability   Complexity   Hygiene   Coverage

Code Review Summary

Analyzer	Status	Updated (UTC)	Details
JavaScript		Aug 18, 2026 11:16a.m.	Review ↗
Code coverage		Aug 18, 2026 11:16a.m.	Review ↗

Code Coverage Summary

Language	Line Coverage (Overall)
Aggregate	69% [▲ up 0.1% from main]
Javascript	69% [▲ up 0.1% from main]

➟ Additional coverage metrics may have been reported. See full coverage report ↗

---

Important

AI Review is run only on demand for your team. We're only showing results of static analysis review right now. To trigger AI Review, comment @deepsourcebot review on this thread.