[lcian]

Changelog Entry

• Session::put now uses resumable uploads automatically for buffers and paths to regular files of 256 MiB or more, and resends only what the server hasn't stored after a failed write
  • Change the threshold with Usecase::with_resumable_threshold_bytes or per request with .resumable_threshold_bytes(...); None disables automatic resumable uploads.
  • Set ClientBuilder::spool_dir to choose where payloads compressed on the fly are spooled (defaults to the system temp dir).
  • The resumable-upload-api feature only gates the manual resumable upload API. This new behavior is enabled regardless of that feature flag.

Details

Session::put now uses resumable uploads automatically for eligible uploads, matching the Python client (#667). An upload is eligible if:

• It's a buffer or a path to a regular file (put, put_path)
• Its input size, before client compression, is at least the resumable threshold (default 256 MiB)
• It's sent standalone. Uploads queued through many() keep their existing batch or direct routing and ignore the threshold.

Open file handles (put_file), streams, generic readers, non-regular paths and inputs of unknown size use a direct upload.
The API and behavior mirrors the Python client:

let usecase = Usecase::from_static("debug_files")
    .with_resumable_threshold_bytes(Some(256 * 1024 * 1024)); // None disables; Some(0) includes every supported payload

session.put(payload).resumable_threshold_bytes(None).send().await?; // per-request override

ClientBuilder::new(url).spool_dir("/var/tmp/objectstore"); // optional

Path and reader lifetime. Each attempt opens an independent reader. Source paths must remain readable, unchanged, and must not be renamed, removed, or replaced until the upload completes. Compressed payloads spool to named temporary files and retries reopen them. The HTTP layer may retain old readers after the upload returns or is dropped, delaying handle closure and temporary-storage reclamation.

For reviewers

• The Error variants ResumableUploadUnavailable, UploadOffsetMismatch and the chunk-granularity variant are no longer feature-gated. Since Error is #[non_exhaustive] (feat(rs-client)!: Mark Error as non-exhaustive #706), this is additive.
• Most of the logic is in put.rs (routing), put/payload.rs (preparation and readers) and put/automatic.rs (recovery loop).

Closes FS-561

[github-actions]

Semver Impact of This PR

This PR lands in these workspaces:

• clients/rust: 🟡 Minor (new features)

📋 Changelog Preview

This is how your changes will appear in the changelog.
Entries from this PR are highlighted with a left border (blockquote style).

---

New Features

• Session::put now uses resumable uploads automatically for buffers and paths to regular files of 256 MiB or more, and resends only what the server hasn't stored after a failed write by lcian with sentry-junior in #700
  • Change the threshold with Usecase::with_resumable_threshold_bytes or per request with .resumable_threshold_bytes(...); None disables automatic resumable uploads.
  • Set ClientBuilder::spool_dir to choose where payloads compressed on the fly are spooled (defaults to the system temp dir).
  • The resumable-upload-api feature only gates the manual resumable upload API. This new behavior is enabled regardless of that feature flag.

---

_{🤖 This preview updates automatically when you update the PR.}

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 3f64519. Configure here.}

[jan-auer]

This means automatic resumable uploads are enabled by default. This is something we should call out explicitly in the changelog.

Let's check with Ingest team what their thoughts on the behavior are.