[SandipBajracharya]

Summary

Files larger than 150 MiB failed to upload to Dropbox with payload_too_large because uploadFileInDropbox used the single-shot /files/upload endpoint (150 MiB cap). Large files now stream through a Dropbox upload session (OUT-4176).

What changed

• chunkStream helper — reads a byte stream into fixed-size buffers (last one smaller); peak memory ≈ one chunk.
• DropboxClient upload session — _uploadFileSession drives start → append_v2 (per chunk) → finish, exposed as retry-wrapped uploadFileSession. getBody yields a fresh stream so a retry restarts the whole session; the streamed byte count is checked against the expected size before finish, so a short/expired download can never commit a partial file.
• Routing — uploadFileInDropbox branches on file.size: over 150 MiB (or unknown size) → session, else the unchanged single-shot path. A shared download step fails on a non-ok response so an error page is never uploaded as content. A log line records which path each file took.
• Constants — DROPBOX_SINGLE_UPLOAD_MAX_BYTES (150 MiB), DROPBOX_UPLOAD_CHUNK_BYTES (16 MiB, a multiple of 4 MiB per Dropbox guidance).
• Logging — raised the logger inspect depth 2 → 4 (with a LOG_DEPTH override) so nested payloads stop printing as [Object]. Cross-cutting fix, bundled here.

Design notes

• 16 MiB is the chunk size, not per file: peak memory ≈ one chunk per in-flight upload (so chunk size × queue concurrency overall).
• On any failure the whole session restarts with a fresh session_id; the orphaned partial session expires on Dropbox's side (no duplicate, mode: 'add').
• Strict > boundary: a file of exactly 150 MiB stays on the single-shot path.

Testing

• pnpm typecheck, pnpm lint, pnpm test — 311 unit tests pass.
• pnpm test:integration — 189 integration tests pass (22 files), including a new large-file flow (assembly-to-dropbox.large-file.integration.test.ts) that drives the session path end-to-end and asserts start/finish ran and the mapping row was created. The single-shot upload is left unmocked, so a routing regression trips msw.
• Unit coverage: chunkStream edges, session sequence/offsets/error, empty-body, retry re-fetch, routing (single/session/unknown), non-ok download, byte-count guard, logger depth.

Before merge

• Confirm file.size is actually populated from Assembly on real files (log it in staging). If it's ever missing, files still upload correctly but always take the slower session path and skip the byte-count guard.

🤖 Generated with Claude Code

[vercel]

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

Project	Deployment	Actions	Updated
dropbox-integration	Ready	Preview	Sep 11, 2026 2:46pm UTC

[linear-code]

OUT-4176

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge with no outstanding correctness, security, or repository-rule violations.

Summary

• Streams large or unknown-size files through Dropbox session start, append, and finish requests.
• Restarts the complete session with a fresh Assembly download when retrying.
• Verifies known file sizes before committing and rejects unsuccessful Assembly downloads.
• Adds unit and integration coverage for chunking, upload-session sequencing, routing, retries, and durable file mappings.
• Makes logger inspection depth configurable while retaining bounded output.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Assembly file] --> B{Download URL present?}
    B -- No --> X[Fail upload]
    B -- Yes --> C{Size known and ≤ 150 MiB?}
    C -- Yes --> D[Fetch Assembly body]
    D --> E[Dropbox files/upload]
    C -- No --> F[Fetch fresh Assembly body]
    F --> G[Upload-session start with first chunk]
    G --> H[Append remaining chunks with offsets]
    H --> I{Known size matches streamed bytes?}
    I -- No --> X
    I -- Yes --> J[Finish and commit Dropbox file]
    E --> K[Complete durable file mapping]
    J --> K

Loading

_{Reviews (2) · Last reviewed commit: "test(OUT-4176): add large-file upload-se..."}

[SandipBajracharya]

@greptileai rereview

[priosshrsth]

lgtm