This page from rolling-metrics summarizes the issues of the EDS-based histogram from Coda Hale/Yammer/Dropwizard metrics and how to fix them pretty well. https://github.com/vladimir-bukhtoyarov/rolling-metrics/blob/master/histograms.md …
-
Show this thread
-
Also, this post nicely illustrates the issues with experiments and charts.https://medium.com/hotels-com-technology/your-latency-metrics-could-be-misleading-you-how-hdrhistogram-can-help-9d545b598374 …
2 replies 0 retweets 1 likeShow this thread -
Replying to @shuheikagawa
I wish heatmaps would be more mainstream; they solve some of the problems outlined in this post.
1 reply 0 retweets 1 like -
Replying to @yoshuawuyts
Do you mean showing a histogram as a heatmap, like x-axis: time and y-axis: logarithmic buckets?
1 reply 0 retweets 1 like -
Replying to @shuheikagawa
Yes, for example! Scales should be tweaked for the domain, but overall sounds about right!
1 reply 0 retweets 1 like -
Replying to @yoshuawuyts
Makes sense, especially for monitoring! On top of it, I think percentiles are nice because we can easily track improvements/regressions. For example ”the p99 response time dropped by 200ms with this change”
1 reply 0 retweets 1 like
Yes, totally. Id like response times to be measured in proportion to a budget tho. It's important to know how much changes are. Saying: response times are now 4x as slow is more significant than saying they're now 20ms longer (for example).
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.