A Phoenix.LiveView.UploadWriter that streams an upload straight into a Dropbox upload session — the bytes never touch your server's disk.

LiveView's default writer spools each entry into a temporary file, so a 500 MB upload is written to disk in full before consume_uploaded_entry/3 can hand it to Magpie.Files.upload_file/4. This writer skips that round trip: chunks are buffered in memory up to :chunk_size and appended to an upload session as they arrive.

Usage

socket =
  allow_upload(socket, :report,
    accept: ~w(.pdf),
    max_file_size: 500_000_000,
    writer: fn _name, entry, _socket ->
      {Magpie.LiveView.UploadWriter,
       client: MyApp.dropbox_client(), path: "/Reports/" <> entry.client_name}
    end
  )

consume_uploaded_entries/3 then receives whatever meta/1 returns, so the Dropbox metadata of the committed file is already there — nothing left to upload:

consume_uploaded_entries(socket, :report, fn %{metadata: metadata}, _entry ->
  {:ok, metadata}
end)

Options

• :client (required) — a Magpie.Client
• :path (required) — the destination path in Dropbox
• :chunk_size — bytes buffered before each append (default 8 MiB). Dropbox wants every append but the last to be a multiple of 4 MiB, so keep this a multiple of 4 MiB.
• :mode — "add" (default) or "overwrite"
• :autorename — default true
• :mute — default false

Notes

Magpie does not depend on :phoenix_live_view — the behaviour is a plain set of callbacks, and declaring it would drag Phoenix into every project that only wants a Dropbox client. The callbacks below match Phoenix.LiveView.UploadWriter as of LiveView 1.0.

They run in the per-entry Phoenix.LiveView.UploadChannel process, not in the LiveView itself: a slow append backpressures that one upload instead of blocking the rest of the page.

The session is opened in init/1. A cancelled or failed upload leaves it dangling, which is harmless — Dropbox drops incomplete sessions after 7 days and nothing is committed without a finish.