A long-lived connection to /jobs/take, delivering jobs to an owner process as they arrive.

This is the low-level consumer. Most applications want Zizq.Worker, which runs one of these and handles concurrency, acknowledgement and shutdown. Use this directly only if you need to drive the loop yourself.

{:ok, _pid} = Zizq.Stream.Take.start_link(client: MyApp.Zizq, prefetch: 10)

receive do
  {:zizq_stream, _pid, {:job, job}} -> handle(job)
end

Messages

The owner (by default whoever called start_link/1) receives:

• {:zizq_stream, pid, {:connected, url}}
• {:zizq_stream, pid, {:job, %Zizq.Job{}}}
• {:zizq_stream, pid, {:disconnected, reason}} — reason is a Zizq.Error or :closed for a clean end of stream

Flow control

The server sends at most :prefetch unacknowledged jobs before it pauses, so acknowledging is what asks for more. That bounds this process's mailbox by construction: it cannot be flooded faster than jobs are being completed, however far behind the consumer falls.

Reconnection

Disconnects are expected on a connection meant to live forever, so they are retried with exponential backoff rather than crashing the process. A response the server will answer identically next time — a rejected queue name, say — is not retried; the process stops instead, because retrying could only loop.