This makes a lot of sense. The purpose of the RollSink is simply to break the data stream into smaller chunks so exact timing isn't necessary. Thread.sleep() probably makes more sense in this case. If you have a patch prepared I'd love to see it!
On Sun, Dec 4, 2011 at 11:19 AM, 音田司 <[email protected]> wrote: > Hello, My name is Otoda. > > I found that RollSink kill performance. > I tried checking TextFileSink 's performance when RollSink uses or not. As > a result, performance using RollSink was haler than not using RollSink. > Maybe, this is caused by RollSink#synchronousAppend() that checks a timing > of rotate using TimeTrigger#isTriggered(). > > And, I think RollSink is too strict to only use flashing data and closing > files like CollectorSink. > In this use case , I don't mind the worst case in which RollSink's rotate > is late for several second. > > So, do you mind my idea using Thread#sleep() in RollSink's sub thread or > other new RollSink instead of TimeTrigger#isTriggered() ? > I tried and that didn't kill performance, but I mind bat effects caused by > this.
