At fisrt load into browser response is very slow. Set ExtractingRoot as context ressources fix this issue by pre-extracting the resources. Maybe this commit can be linked this issue #6177
Comment From: pivotal-cla
@dlemaignent Please sign the Contributor License Agreement!
Click here to manually synchronize the status of this Pull Request.
See the FAQ for frequently asked questions.
Comment From: pivotal-cla
@dlemaignent Thank you for signing the Contributor License Agreement!
Comment From: dlemaignent
Note : to reproduce issue you need to launch spring boot web application with a war package set as executable (./my_app.war)
Comment From: wilkinsona
Thanks for the suggestion.
At fisrt load into browser response is very slow. Set ExtractingRoot as context ressources fix this issue
Can you please provide a minimal sample application that illustrates the difference?
Maybe this commit can be linked this issue https://github.com/spring-projects/spring-boot/issues/6177
Our belief is that https://github.com/spring-projects/spring-boot/issues/6216 accounts for much of the difference when running a war. A sample application that illustrates the problem you're facing will help us to see if that's not the case.
Note : to reproduce issue you need to launch spring boot web application with a war package set as executable (./my_app.war)
Looking at Tomcat's code, I cannot see why it would not also reproduce when running with java -jar my_app.war. Again, a sample application will help us to understand exactly what's happening.
Comment From: dlemaignent
Thanks for your quick response.
I've created this repo https://github.com/dlemaignent/first-war-start-slow to reproduce the issue.
I confirm that the application can be launched with java -jar my_app.war , the issue is the same.
Comment From: wilkinsona
Thanks, @dlemaignent. I'm seeing a significant difference in startup time with ./mvnw spring-boot:run (3.078s) vs java -jar (8.561s):
2022-03-10 15:14:02.082 INFO 64384 --- [ restartedMain] c.e.f.FirstWarStartSlowApplication : Started FirstWarStartSlowApplication in 3.078 seconds (JVM running for 3.544)
2022-03-10 15:15:24.511 INFO 64501 --- [ main] c.e.f.FirstWarStartSlowApplication : Started FirstWarStartSlowApplication in 8.561 seconds (JVM running for 9.487)
Switching to Undertow, the difference is much less pronounced:
2022-03-10 15:25:40.502 INFO 68556 --- [ restartedMain] c.e.f.FirstWarStartSlowApplication : Started FirstWarStartSlowApplication in 2.893 seconds (JVM running for 3.343)
2022-03-10 15:19:44.830 INFO 68316 --- [ main] c.e.f.FirstWarStartSlowApplication : Started FirstWarStartSlowApplication in 4.61 seconds (JVM running for 5.209)
Switching back to Tomcat but with jar packaging, it is significantly quicker with performance that is similar to Undertow's:
2022-03-10 15:27:43.556 INFO 69563 --- [ main] c.e.f.FirstWarStartSlowApplication : Started FirstWarStartSlowApplication in 4.739 seconds (JVM running for 5.265)
There's obviously work to do here so that Tomcat starts quickly when packaged as a war. However, I don't think that extracting by default should be the answer. It would mean that changing packaging alters the runtime behaviour in terms of how the application writes to the filesystem and I think that might catch a lot of people out. I'd prefer that we find a way to close the performance gap without making a functional change. In the meantime, you can opt in as you have done using a customizer. Thanks anyway for the suggestion.