Initial take at adding JMH micro-benchmarks #12663
Conversation
|
@dweiss thank you for this! I tried it out and it works great. Glad to see the |
|
Let me review the minor remaining bits - I had to leave the computer before the tests completed. |
|
@dweiss this is awesome!!! Thank you so much! |
|
@dweiss one suggestion I have is adding a I tested this locally and was able to follow your directions and get the benchmarks to run without issue. |
|
I enabled forbiddenapis to the extent possible, to make Uwe happier. ;) I like those flame-graphs that Solr has, etc., but these can come later. I'll add a simple |
|
@benwtrent Can you take a look? If there's any wording you think would work better, please do change it. Explaining how JMH works is perhaps beyond the scope of the readme file... ;) |
| -h displays verbose help for all options | ||
| -l list available benchmarks | ||
| -lp list benchmarks that pass the filter and their parameters | ||
| regexp execute all benchmark containing regexp |
There was a problem hiding this comment.
| regexp execute all benchmark containing regexp | |
| regexp execute all benchmark containing regexp | |
| Here is an example running a single benchmark: | |
| java --module-path lucene\benchmark-jmh\build\benchmarks --module org.apache.lucene.benchmark.jmh org.apache.lucene.benchmark.jmh.BinaryCosineBenchmark.cosineDistanceNew | |
| Or running all of BinaryCosineBenchmark | |
| java --module-path lucene\benchmark-jmh\build\benchmarks --module org.apache.lucene.benchmark.jmh org.apache.lucene.benchmark.jmh.BinaryCosineBenchmark |
There was a problem hiding this comment.
Oh, apologies - I thought you added that change to the branch, not as github's suggested change. I added this manually.
|
Looks good @dweiss I added a line with an example for running a single benchmark or suite of benchmarks. |
3 participants
As per discussion in: #12641
This patch adds the ability to compile JMH microbenchmarks and run them from within Lucene codebase. I didn't use any plugins as they, in my opinion, obscure the view of what's happening and what's needed. I also wanted it to work with the module system and without uber-jars.
compiles microbenchmarks with JMH, then displays a short help on what to do next:
I didn't add any gradle-specific tasks to run the benchmarks directly - it's technically easy but it seems like a bad idea to fork a subprocess from (quite heavy) gradle when such delicate things are being measured? I'm not sure here.
Some future improvements/ remaining issues include: