Fix corrupted or incomplete files after interrupted download

[BigRoy]

Stacked on #358 (same download/upload loop). Review only the last commit; the base changes to develop once that PR is merged.

Bug

When a download is interrupted and retried, the resulting file can be corrupted or incomplete without any error:

1. Duplicated content — the retry sends Range: bytes=<offset>-. If the server (or a proxy) ignores Range and answers 200 with the whole file, the whole file is appended after the already written part.
2. Incomplete file accepted — if the connection is closed before all content arrives without raising (or raises ChunkedEncodingError, which wasn't caught), the download is reported as finished.
3. Content-Length was stored in progress as a string, and a response without it (chunked transfer) raised KeyError.

Cause

• Response status was never checked for 206 Partial Content before appending.
• Received bytes were never compared with the expected size.
• response.headers["Content-length"] was passed directly to set_content_size.

Fix

• If continuing (offset > 0) and the response is not 206, truncate the output stream and reset progress, then write the full content.
• After the response is consumed, compare transferred bytes with the content size. A mismatch raises ConnectionError, which is retried with Range like other connection errors. ChunkedEncodingError is retried as well.
• Content-Length is converted to int (for 206 the already downloaded offset is added) and a missing header is allowed.

Reproduce

Hard to trigger on demand against a real server, tests/test_download_resume.py simulates:

• a response closed after 4 of 10 bytes, continued with a 206,
• a server ignoring Range (develop result: b"01230123456789"),
• a response without Content-Length (develop: KeyError).

Testing notes

• Normal downloads: con.download_file(...) of a thumbnail/file gives identical bytes and progress.content_size is now an int.
• Manual interruption: download a large file through a proxy that drops the connection (e.g. toxiproxy limit_data toxic), compare the checksum with the source file. Test once with a proxy that strips Range to cover the 200 case.

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

Clear the stale Range header after resetting progress and add the requested chunked-error regression test.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Fixes resumable downloads by detecting incomplete responses, handling ignored ranges, and supporting missing Content-Length headers.

Changes:

• Resets partial output when resumed responses are not 206.
• Validates transfer sizes and retries incomplete downloads.
• Adds regression tests for interrupted and range-ignored downloads.

File summaries

File	Summary
tests/test_download_resume.py	Adds interrupted-download regression coverage.
ayon_api/server_api.py	Implements resume integrity checks and retry handling; a critical stale Range header issue remains, and ChunkedEncodingError coverage is missing.

Review details

Suppressed comments (1)

ayon_api/server_api.py:1723

• The new ChunkedEncodingError branch is not exercised by these regression tests: all FakeResponse iterators complete normally, so a future removal or misordering of this exception handler could go unnoticed. Add a response/iterator that raises ChunkedEncodingError after yielding a prefix and assert the retry uses the Range offset and produces the complete content.

                requests.exceptions.ChunkedEncodingError,

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟡 Changes recommended

Correct Content-Length parsing and missing-header handling before approval.

Get a fresh assessment by requesting another Copilot review.

Review details

• Files reviewed: 3/3 changed files
• Comments generated: 1
• Review effort level: Lite