apupier opened a new pull request, #30341:
URL: https://github.com/apache/camel/pull/30341

   CAMEL-27163: Only instrument `org.apache.camel` classes with the JaCoCo 
agent in the coverage profile
   
   ## Problem
   
   On Jenkins, the coverage profile (`-Pcoverage`) moves the build and test 
time of the ubuntu-avx jdk 21 build from about 34 min to about 55 min. The test 
JVMs also use noticeably more memory.
   
   The `prepare-agent` execution of the `coverage` profile in `parent/pom.xml` 
had no `includes`, so the JaCoCo agent instrumented every class loaded by the 
forked test JVMs: Camel classes, but also all the third party libraries and the 
test dependencies. This costs CPU and memory in every forked test JVM and 
brings nothing: the coverage reported to Sonar 
(`coverage/target/site/jacoco-aggregate/jacoco.xml`) is only about the Camel 
main sources.
   
   ## Change
   
   Add `<includes><include>org/apache/camel/**</include></includes>` to the 
agent configuration of the `coverage` profile in `parent/pom.xml`. The existing 
`CamelThreadFactory*` exclusion (workaround of jacoco/jacoco#407) is kept.
   
   All the main sources of the repository are in the `org.apache.camel` 
package. The only other package is `com.acme`, a file in the test resources of 
`camel-jbang-core` used as an example, which does not need coverage.
   
   The generated agent argument is now:
   
   ```
   
-javaagent:...org.jacoco.agent-0.8.15-runtime.jar=destfile=.../target/jacoco.exec,includes=org/apache/camel/**,excludes=**/CamelThreadFactory*
   ```
   
   ## Measurements
   
   Both runs were done on the same machine (22 cores, 30 GB), with the exact 
command of the `Jenkinsfile` for the ubuntu-avx jdk 21 build (`./mvnw -T1C -B 
-e -fae -V -Dnoassembly -Dsurefire.rerunFailingTestsCount=2 
-Dfailsafe.rerunFailingTestsCount=1 -Dci.env.name=apache.org 
-Darchetype.test.skip -Dmaven.test.failure.ignore=true -Dcheckstyle.skip=true 
install -Dcamel.threads.virtual.enabled=false -Pcoverage`), once with the 
change and once without.
   
   The build without the change was stopped by an out of memory at module 646 
of 729, so the comparison is on the part of the build that completed in both 
runs. The build with the change completed (`BUILD SUCCESS`, 37 min).
   
   | Measure | Without the change | With the change | Ratio |
   |---|---|---|---|
   | Sum of the test time of the 11,866 test classes run in both builds | 
49,685 s | 41,942 s | 0.84 |
   | Sum for the 237 modules with more than 30 s of test time | 46,661 s | 
39,562 s | 0.85 |
   | Median per module ratio | | | 0.85 |
   | Size of `jacoco.exec` for `camel-netty` | 241 KB | 94 KB | 0.39 |
   
   Individual modules are noisy (the machine was loaded and swapping at times, 
some modules are slower, others faster), the totals are what is meaningful. The 
real effect on the Jenkins node will be visible on the next build of `main`.
   
   ## Notes
   
   - The coverage report generation (`report-aggregate`) is unchanged.
   - No test change: this only changes the configuration of the `coverage` 
profile, which is not active by default.
   
   ---
   _This PR description is AI-generated by IBM Bob 2.2.1 on behalf of @apupier._
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to