Make chunk_size_mb control browser upload chunks

[usehoplite]

Problem

chunk_size_mb is documented as a setting but nothing reads it. The handler stored the value and never used it, and the browser always sent 10 MiB chunks. The default also disagreed with itself: 5 in the config loader, 10 in the code constant, and 5 in the bundled config.yaml. An operator behind a proxy with a small request-body limit had no way to shrink upload chunks.

Change

• Add GET /api/v1/system/upload, which returns { "chunkSizeBytes": N } from chunk_size_mb.
• The upload store reads it before the first upload and passes it to the existing chunkSize option. Workers that start together share one request, so they use the same size. If the request fails, they keep the 10 MiB fallback and the next upload asks again. An upload cancelled while the request is pending never starts.
• Use one default, DefaultChunkSizeMB (10). The bundled config.yaml now says 10, so default deployments keep the chunk size browsers sent before.
• Remove the unused chunk size from StreamHandler.
• Update docs/api.md, docs/configuration.md and the changelog.

An existing deployment that copied the old config.yaml with chunk_size_mb: 5 now uploads in 5 MiB chunks. That is the value they set, but it doubles the request count for large files.

Verification

• New handler test covers a configured value and the unset fallback.
• go vet ./..., handler, config and cmd tests pass. bun run check, lint and build pass.
• Ran the real server with chunk_size_mb: 3. GET /api/v1/system/upload returned {"chunkSizeBytes":3145728} with a token and 401 without one.
• In a real browser against the stack-tip build with chunk_size_mb: 3, a 8,000,000-byte upload called GET /api/v1/system/upload and then sent 3 chunk requests (200, 200, 201). The stored file's SHA-256 matched the source.
• Three files uploaded together made one GET /api/v1/system/upload, and all three stored files matched.
• Not tested: a deployment behind a proxy with a small body limit, and the fallback path when GET /system/upload fails.

Stack: 2 of 6. Targets the branch of the PR below it.

---

Written by anthropic/claude-sonnet-5-5 in the Hoplite agent harness.

[vercel]

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

Project	Deployment	Actions	Updated
box-box	Ready	Preview	Oct 1, 2026 8:02am UTC

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: ASSERTIVE

Plan: Advanced

Run ID: a3206309-c8a4-4569-b15c-f3803a24c7eb

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Autopilot is currently an internal CodeRabbit preview.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[greptile-apps]

Concurrent uploads use different chunk sizes

When uploads start together, each worker fetches the chunk setting separately. If one request fails while another succeeds, the first upload uses the 10 MiB fallback even though the server specifies a smaller size. If a proxy limits request bodies below 10 MiB, that upload fails while the other can succeed. Share the in-flight configuration request so the workers use the same result.

Knowledge Base Used: Frontend data and real-time state

[greptile-apps]

Cancellation waits for configuration

If a user cancels while the configuration request is pending, that request continues and keeps a worker slot occupied. When it finishes, processItem starts the upload flow without checking whether the signal was aborted. The subsequent progress and error updates can replace the upload’s cancelled state and report it as failed.

Knowledge Base Used: Frontend data and real-time state