Long-running streams UX

[kylecarbs]

When creating a long-running agent, especially one that runs indefinitely, the writer can grow quite large.

I’m persisting all messages to the database, so ideally the reader would resume after the last message stored there. Right now the only obvious way to do this is by counting chunks and tracking that offset in the database.

It would be great if I could reset the writer state so that future connections only read from the new starting point. I can use startIndex, but if a client has never connected before, this can still return megabytes (or more) of data that the client already has.

For now I can read the sum of parts and index based off that, it's just kinda janky.

[TooTallNate]

Not quite sure I understand, if the client has never connected then how would it already have the data?

In any case, just an idea: would it be helpful if startIndex could be a negative value to start reading from the tail end of the stream? So you'd do startIndex: -1 and it would only return new data starting from the current end of the stream.

[kylecarbs]

So I have a loop:

function startAgent() {
  "use workflow";

  const writer = getWriter();
  while (true) {
    // do stream step with writer
    // store message in the database
  
    await sleep("10s");
  }
}

So I don't need the data from the writer anymore, because I'll be fetching messages on page load.

I was thinking a negative index would be great too.