fix(traces): support larger execution trace archives

[TheodoreSpeaks]

Summary

• Support execution trace archives up to 512 MiB while preserving the 64 MiB cap on individual workflow values. Completion writes and backfills use a dedicated archive writer with the existing storage reference format, ownership metadata, and cleanup behavior.
• Enforce the archive cap on reads and before backfill payload loading. Preserve the scheduler's serialized-byte budget and checkpoint behavior; size or storage failures stop the run without advancing past an incomplete page.
• Keep the default in-flight payload budget at 512 MiB. Parsed objects, serialization buffers, and the bounded shared cache use additional memory, which the existing RSS telemetry reports.

Type of Change

• Bug fix

Testing

• 821 tests passed across 32 related suites. Coverage includes archive admission above the ordinary value limit, the archive cap, byte-budget rejection before payload reads, checkpoint preservation, durable ownership, and storage failures.
• Round-trip tests exercise real serialization and cache-miss reads with mocked object storage and reduced test caps, including multibyte data and oversized reference rejection.
• bun run lint, all 47 check:audits checks, block registry validation, and docs manifest validation passed.
• bun run type-check passed in apps/sim.

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
docs	Skipped		Sep 23, 2026 1:29am UTC

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge; no actionable correctness, security, or repository-rule violations were identified.

Summary

This PR adds an explicit --skip-oversized mode to the trace-span backfill while retaining fail-fast behavior by default.

• Detects payloads above the durable-storage limit before reading or uploading them.
• Logs each oversized record and reports an aggregate oversized count.
• Advances a page checkpoint only after all eligible records in that page finish successfully.
• Adds coverage for mixed and fully oversized pages, both scan orders, boundary sizes, interrupted pages, and storage failures.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Scan metadata page] --> B[Query payload byte sizes]
  B --> C{Payload oversized?}
  C -->|No| D[Process eligible record]
  C -->|Yes, flag disabled| E[Fail the run]
  C -->|Yes, flag enabled| F[Warn and leave payload inline]
  D --> G{All eligible records completed?}
  F --> G
  G -->|No: failure or shutdown| H[Retain previous checkpoint]
  G -->|Yes| I[Checkpoint after full page]
  I --> J[Continue to next page]

Loading

_{Reviews (1) · Last reviewed commit: "fix(traces): allow explicitly skipping o..."}

[cubic-dev-ai]

No issues found across 2 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}