fix(kernels): stream kernels output files through download_file

[nefayran]

Summary

kaggle kernels output now saves each file through download_file, the path dataset and competition downloads already take. Files are written in chunks instead of held in memory, checked against Content-Length, and resumed after a dropped connection.

Problem

kernels_output fetches each file with requests.get(url, stream=True) and writes response.content, which reads the whole body into memory. The request has no read timeout, nothing retries it, and nothing compares the written size with Content-Length. The target file is opened before the body is read, so when the connection drops, the command fails and leaves an empty file under the final name.

Measured with an in-process HTTP server (macOS arm64, Python 3.14, requests 2.34.2):

case	main	this branch
512 MiB output: growth of the client's peak RSS	1,175 MB	7 MB
3 MiB output, connection cut after 1.5 MiB	ChunkedEncodingError, 0-byte file left on disk	resumed from the first 1 MiB chunk, all 3,145,728 bytes written

Solution

Call download_file for kernel output files and pass timeout=300 to the first request, the same read timeout download_file uses for its retries. Resume is on unless --force is given, as in datasets download.

download_file does not use its http_client argument, and the client opened for the output listing is closed by the time the files are fetched, so the call passes None.

Tests

• New tests/unit/test_kernels_output_download.py runs kernels_output against a local HTTP server that supports Range requests. One test checks that the file goes through download_file and arrives intact. The other cuts the first connection halfway through the second 1 MiB chunk and checks that the retry asks for bytes=1048576- and that no .kaggle-partial marker is left. Both fail on main.
• test_kernels_output_paths.py and two pagination tests in test_kernels_logs.py faked the old .content read. They now stub download_file, and the requests.get assertion includes the timeout.

pytest tests/unit: 1397 passed. black --check and mypy report no issues.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[stevemessick]

/gcbrun

[stevemessick]

Nice!. Thanks.