fix: add upload retry

[not-matthias]

No description provided.

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 7 untouched benchmarks

---

_{Comparing cod-2839-retry-upload-on-connection-errorstimeouts (2298ed2) with main (41f4db9)}

[greptile-apps]

Greptile Summary

Adds upload retry for on-disk (streaming) profile archives. Previously STREAMING_CLIENT had no retry middleware because the stream body can't be replayed; this PR introduces send_streamed_with_retry, which reopens the file from disk on each attempt and drives the shared ExponentialBackoff policy manually.

• upload_backoff() is extracted to request_client.rs so both the middleware on REQUEST_CLIENT and the new manual retry loop share the same policy; under #[cfg(test)] the intervals are shrunk to milliseconds.
• send_streamed_with_retry uses DefaultRetryableStrategy to classify responses and ExponentialBackoff::should_retry to gate each attempt, rebuilding the tokio file stream on every retry.
• Two new unit tests (streamed_upload_is_retried, in_memory_upload_is_retried) use a raw TCP mock server to assert that both code paths make exactly 1 + UPLOAD_RETRY_COUNT attempts on persistent 503 responses.

Confidence Score: 5/5

Safe to merge. The production retry logic is correct, response-level errors are handled by the existing caller check, and both new tests pass through realistic code paths.

The production retry logic is correct — send_streamed_with_retry correctly rebuilds the stream body, classifies transient errors via DefaultRetryableStrategy, and gates retries through ExponentialBackoff. The only finding is a test-infrastructure concern that does not affect production behaviour.

No files require special attention.

Important Files Changed

Filename	Overview
src/request_client.rs	Extracts upload_backoff() as a shared ExponentialBackoff factory; uses #[cfg(test)] to shrink intervals to milliseconds so tests don't sleep through real backoff. UPLOAD_RETRY_COUNT is made pub so the uploader can reference it.
src/upload/uploader.rs	Adds send_streamed_with_retry which manually rebuilds the stream body from disk on each attempt and drives ExponentialBackoff + DefaultRetryableStrategy for transient errors; replaces the previous single-shot streaming upload. Two new unit tests verify the retry counts for both the streaming and in-memory code paths using a raw TCP mock server.

Sequence Diagram

sequenceDiagram
    participant U as upload_profile_archive
    participant S as send_streamed_with_retry
    participant F as File (disk)
    participant SC as STREAMING_CLIENT
    participant SV as Upload Server

    U->>S: call (path, upload_data, …)
    loop up to 1 + UPLOAD_RETRY_COUNT attempts
        S->>F: File::open(path)
        F-->>S: file handle
        S->>SC: PUT upload_url (streaming body)
        SC->>SV: HTTP PUT
        SV-->>SC: 503 (transient)
        SC-->>S: "Ok(Response{503})"
        S->>S: DefaultRetryableStrategy → Transient
        S->>S: policy.should_retry(start, n_past_retries)
        alt Retry allowed
            S->>S: "sleep(backoff), n_past_retries += 1"
        else Retries exhausted
            S-->>U: "Ok(Response{503})"
        end
    end
    U->>U: !response.status().is_success() → bail!

Loading

_{Reviews (2): Last reviewed commit: "fix(upload): retry streamed uploads on t..." | Re-trigger Greptile}

[GuillaumeLagrange]

lgtm