send websocket messages without flushing, with periodic flush

[ornicar]

not used yet, just implementing the behaviour

[lenguyenthanh]

should it be counter = 0 🤔

[ornicar]

Dammit discarded values!!

[schlawg]

i don't think it really matters, the clientEmit closure seems to recreate itself with some regularity.

[schlawg]

probably because i'm reconnecting..

[schlawg]

i was thinking something like:

final class FlushPeriodically()(using scheduler: Scheduler, ec: Executor):
  val flushQ = scala.collection.mutable.LinkedHashSet.empty[Channel]

  scheduler.scheduleAtFixedRate(0 seconds, 1 millisecond)(() => flushEm())

  def apply(channel: Channel) =
    synchronized: // maybe ConcurrentHashMap could avoid synchronized
      flushQ += channel

  private def flushEm(): Unit =
    val toFlush = synchronized:
      val f = flushQ.take(max(100, flushQ.size / 500) // try to stick to 100/1ms
      flushQ --= f
      f
    toFlush.foreach(_.flush())

mine doesn't do any batching but it does smooth traffic, so maybe we'll need a combo of the two approaches

[schlawg]

i cannot test the code in this PR because clients just keep reconnecting with the emitToChannelWithPeriodicFlush lambda path. could scala hate me more?

  private def emitToChannel(channel: Channel): ClientEmit =
    // partial function with direct pattern matching
  private def emitToChannelWithPeriodicFlush(channel: Channel): ClientEmit =
    // closure & lambda wrapper
    msg =>
      msg match
        // pattern matching

[ornicar]

the client reconnects because lila-ws doesn't respond to its null ping with the expected 0 pong.

[schlawg]

except for adding:

        case ipc.ClientIn.Pong =>
          channel.writeAndFlush("0")

doesn't help.

the client reconnects because lila-ws doesn't respond to its null ping with the expected 0 pong.

yeah, i had commented out all of the periodically stuff, changed to use writeAndFlush in both, and was left with the only difference between the two methods being one was wrapped in another layer of lambda. and swapping between them the wrapped one was still reconnecting every few seconds. but this morning things are behaving, so maybe i only thought i was rebuilding? at least it's me that's crazy and not scalac or netty.