Can't download files larger than /tmp mount size

[geerlingguy]

Describe the bug

I am trying to download some large files using files.download() to a Raspberry Pi. Due to the default filesystem layout, /tmp on the Pi is on tmpfs and on my 16 GB Pi CM5, I only get 8GB of data on that path:

$ df -h
Filesystem      Size  Used Avail Use% Mounted on
udev            7.8G     0  7.8G   0% /dev
tmpfs           3.2G  9.8M  3.2G   1% /run
/dev/sda2       906G   22G  848G   3% /
tmpfs           7.9G  468K  7.9G   1% /dev/shm
tmpfs           5.0M   16K  5.0M   1% /run/lock
tmpfs           1.0M     0  1.0M   0% /run/credentials/systemd-journald.service
tmpfs           7.9G  7.9G     0 100% /tmp

When downloading larger files (e.g. 10 GB LLM models), Pyinfra eventually fails.

To Reproduce

Using a Raspberry Pi (or some other system with a small-ish /tmp directory, use the following Pyinfra task to download a large file:

files.download(
    name="Downloading model: gpt-oss-20b-Q4_K_M.gguf",
    src="https://huggingface.co/unsloth/gpt-oss-20b-GGUF/resolve/main/gpt-oss-20b-Q4_K_M.gguf",
    dest="~/Downloads/llama.cpp/models/gpt-oss-20b-Q4_K_M.gguf".format(working_dir, filename),
)

Expected behavior

The file should download into the path ~/Downloads/llama.cpp/models/gpt-oss-20b-Q4_K_M.gguf

Instead, I get:

    Starting nested operation: Downloading model: gpt-oss-20b-Q4_K_M.gguf (file 1 of 1) 
    [cm5.local] nested curl: (23) Failure writing output to destination, passed 15495 returned 0
    [cm5.local] nested Error: executed 0 commands

Meta

$ pyinfra --support

    If you are having issues with pyinfra or wish to make feature requests, please
    check out the GitHub issues at https://github.com/Fizzadar/pyinfra/issues .
    When adding an issue, be sure to include the following:

    System: Darwin
      Platform: macOS-15.6.1-arm64-arm-64bit-Mach-O
      Release: 24.6.0
      Machine: arm64
    pyinfra: v3.2
      black: v25.1.0
      black: v25.1.0
      click: v8.1.8
      distro: v1.9.0
      gevent: v24.11.1
      importlib_metadata: v8.6.1
      jinja2: v3.1.6
      packaging: v24.2
      paramiko: v3.5.1
      python-dateutil: v2.9.0.post0
      pywinrm: v0.5.0
      pyyaml: v6.0.2
      pyyaml: v6.0.2
      setuptools: v80.9.0
      typeguard: v4.4.2
      typing-extensions: v4.13.2
      wheel: v0.45.1
    Executable: /opt/homebrew/bin/pyinfra
    Python: 3.13.7 (CPython, Clang 17.0.0 (clang-1700.0.13.3))

[geerlingguy]

One potential workaround would be to specify a TMPDIR as an environment variable for curl.

[Fizzadar]

Another option would be setting config.TEMP_DIR (https://github.com/pyinfra-dev/pyinfra/blob/a43146afc6a3c2b34078b74651a81f89e5c7c02b/src/pyinfra/api/config.py#L28).

But I think a nicer option would be an additional _temp_dir global argument (which would fallback to the config).

[wowi42]

@Fizzadar let me know

@geerlingguy you can backport locally if you want

[geerlingguy]

Thanks! This will help on my Pis, where my poor tmp dir is always 8GB and I'm often downloading rather large things for testing, like AI models that are 30+ GB :D