Streaming

• In HTTP, when you make a request, the server can send the response all at once or in a streamed/chunked way
• This concept is similar to what you have with database cursors
• While streaming, your HTTP connection stays open
• The server keeps the connection open while sending data in chunks.
• When the server finishes sending all data, it closes the connection (or signals the end of the stream, e.g., with a zero-length chunk in HTTP/1.1).

> GET /streaming/mystream HTTP/1.1
> Host: localhost:8000
> User-Agent: curl/8.7.1
> accept: application/json

< HTTP/1.1 200 OK
< date: Sun, 17 Aug 2025 02:56:00 GMT
< server: uvicorn
< content-type: application/x-ndjson
< x-process-time: 0.000737959002435673
< transfer-encoding: chunked

• HTTP streaming responses typically use the header Transfer-Encoding: chunked
• Content-Type
• application/x-ndjson
• text/event-stream: server-side events

Streaming formats

• The key is that the format must allow partial consumption, you shouldn't need the whole file or array in memory to process the first piece.

JSON Lines (NDJSON)

• Each record is a separate line.
• Easy to parse as it arrives.
• Example: logs, telemetry, large datasets.

{"a":1}
{"b":2}
{"c":3}

CSV / TSV

• Each row can be processed immediately.
• Often used for bulk exports or analytics data.

id,name\n1,Alice\n2,Bob\n

Binary formats with framing

• Examples: Protocol Buffers (with length-prefix), Avro, MessagePack.
• Efficient for large datasets; you can deserialize records as they stream in.

Media / large files

• Audio (.mp3, .wav), video (.mp4, .mkv), images.
• Typically streamed as raw bytes or in chunks over HTTP.

Server-Sent Events (SSE) or WebSockets messages

• For continuous updates or real-time feeds.
• Each message is discrete and can be handled as it arrives.