Gil Tene

Note that the Hytrix issue referred is not about this. It was a misuse of non-thread safe (and non-cocurrent) histograms. The issue here (I suspect) is localized to ConcurrentDoubleHistogram. It...

Ok. I think we now have the ConcurrentDoubleHistogram race fixed, and only auto-range-triggering and scaling recordings will be blocked (recordings that do not stretch the edges will continue to be...

Rarely reproducing races are actually worse than more frequently reproducing ones ;-( There was probably more than one race, and with the more common one (found by inspection) plugged up...

@chapj2001: your observations case (AIOOB exceptions during getMean() calls on ConcurrentHistogram) are expected behavior. The ConcurrentHistogram states "... It is important to note that concurrent recording, auto-sizing, and value shifting...

Edited: I take it (the below) back. @stepnav your reproducer test's use of a ReaderWriterLock to prevent getMean being called while modifications are in flight makes the use case supported,...

Any news on whether this still reproduces? I had recently fixed another concurrent race in ConcurrentDoubleHistogram as part of addressing https://github.com/HdrHistogram/HdrHistogram/issues/156 . not yet pushed to maven central, but buolding...

I think that this is a case of misunderstanding the meaning of the lowest discernible value, with the misunderstanding made much more likely due to an unfortunate naming of the...

Good point. Suggestions for specific way to indicate that?

The issue with a methodology that runs at e.g. 1/2 of the max is that bad latency behavior is not caused (primarily) by qps. While there is definitely a qps...

TLDR: Zulu binaries are real, 100% Open Source. It is diligently packaged as such. Unlike many other binary packages that are not quite that. (copied from prior clear response to...