Felix Geisendörfer
Felix Geisendörfer
I'm no longer involved with this project, but decided to take a quick look since it seems interesting. As far as I can tell, my theory from 2012 is probably...
@logidelic yeah, they should help. I.e you should be able to see better timer latencies in exchange of slightly lower overall throughput.
@dougwilson are you sure? I just modified my example to use `socket.pause()` and `setImmediate` and was able to reduce the timer starvation from ~1800ms at a time to ~80ms: https://gist.github.com/felixge/d16ee6b128af7256862bf83fe2f34d8d#file-blocked-by-net-js...
> Nice! It would be very interesting to know if this impacts the wall clock of large query results, but fixing scheduling lock ups is more important than raw throughput...
@logidelic I replied to the node thread. I think node has a scheduling problem and the suggested workaround seems unreasonable for many scenarios, including node-mysql. Let’s wait for the discussion...
@logidelic any sort of attempt to implement user-land scheduling will be a tradeoff between latency and throughput. That being said, I suspect your proposal will have a higher overhead than...
@Hueristic please explain.
@Hueristic this links to the first page of a 192 page thread. I don't see the connection to this issue right away. Care to explain?
@dougwilson thx. I suppose that is what happens if you write financial software in a dynamically typed language ...
https://github.com/felixge/node-formidable/issues/412