elastic / hey-apm

Basic load generation for apm-server built on hey
Apache License 2.0
16 stars 16 forks source link

Ensure chained/tree exceptions are performance measured. #120

Open simitt opened 5 years ago

simitt commented 5 years ago

Extend current implementation on creating exceptions to reflect https://github.com/elastic/apm/issues/40.

jalvz commented 5 years ago

We can easily proxy this, eg. instead of testing X exceptions with 10 causes per second, test 10X exceptions / second. Scanning a tree with a few levels of depth is really not a problem.

simitt commented 5 years ago

I am not sure I follow your comment @jalvz . Afaik there are no micro-benchmarks on chained exceptions in the server repo, so what I suggested here was to add some chained exceptions to the benchmark test cases that we have defined. Previous observations suggested that a larger amount of stack traces can highly influence CPU usage, so having a test case for multiple chained exceptions seems reasonable to me.

jalvz commented 5 years ago

i meant that there is no much difference between testing performance of 10 exceptions, than 1 exception with 10 causes; so we rather test the performance of 10 exceptions to get an idea.

stacktraces is another story because it is a different data type, and with much larger variability (5-100 frames/error?)... 10's of levels of cause exceptions per error I find it really bizarre.

i'm fine reopening if you are concerned about it