Closed MainRo closed 8 years ago
Thank you, @MainRo! :)
@timofey-barmin: shouldn't MZBench do this by itself (ie. ensuring that mz_histogram:notify values are positive). Or rather: is there a specific reason for not doing it?
Hey @ioolkos,
As you can see mzbench throws {value_out_of_range,-63} in that case. I don't think we should always silently write 0 in histogram. mz_histogram is based on hdr_histogram hence only not negative values make sense (they are supposed to be latencies). If we get negative value we don't know what to do with it. In your case maybe it's ok to store 0 but in another case maybe you'll want to crash or increment some kind of "number of anomalies" counter for example. Only mzbench user knows how to handle it. Does it make sense?
@timofey-barmin @MainRo Yes, this makes perfect sense. Use cases are different & some "anomalies counter" might be useful for some. I'll merge the Set to 0 solution for now, so that we can collect experience on this.
Thanks a lot, Timofey and Romain, for your time & effort :+1:
Add some checks to ensure that computing time differences is always positive. This fixes the following worker errors:
08:20:37.094 [error] <0.227.0> gen_fsm <0.227.0> in state connected terminated with reason: {value_out_of_range,-63} in mz_histogram:notify/2 line 73