Improve PlexusIoZipFileResourceCollection performance by using JarFile #106
Conversation
|
Hi, this is my attempt to fix codehaus-plexus/plexus-io#12 without steping onto codehaus-plexus/plexus-io#2. What do you think? Of course for me the long term solution would be to deprecate @rbjorklin @vojtechhabarta would be great if you can confirm that this PR solves your performance issues. |
|
Yes, I can confirm that this change solves the issue with performance (6 minutes shortened to 30 seconds in our case). BTW in addition to building this PR it was also needed to change and build |
|
If there are no objections I plan to merge this PR. @vojtechhabarta thank you for testing the patch. And for pointing that plexus-archiver 3.7.0 is not exactly backward compatible as it introduces new method in the interface - I've overlooked that. Do you think that we need to bump the major version? /cc @michael-o @krosenvold @khmarbaise |
|
I'd bump it. |
Currently `PlexusIoZipFileResourceCollection` uses `PlexusIoURLResource` to get the `InputStream` of the JAR entries. `PlexusIoURLResource` uses `URL` and `URLConnection` to get the input stream. The problem is that they create a new `JarFile` for every entry and in some cases the `JarFile` initialization could be expensive (for example when he JAR is signed). Using the `URLConnection` cache would solve the performance issues but opens new one. The cache is global for the build so if the JAR file have changed during the build you may get the cached instance (see plexus-io#2). Modify `PlexusIoZipFileResourceCollection` to use `JarFile` directly instead of using `PlexusIoURLResource`. That would solve solves the two issues - `JarFile` is initialized once so there is no performance penalty and it is local so if the file changed during the build the latest version will be picked up. Under the hood `PlexusIoURLResource` uses `JarFile` as well and keeping `ZipFileResource` extending `PlexusIoURLResource` would both keep the behaviour and interface backward compatible. Closes #106
Currently
PlexusIoZipFileResourceCollectionusesPlexusIoURLResourceto get theInputStreamof the JAR entries.PlexusIoURLResourceusesURLandURLConnectionto get the input stream. The problem is that they create a newJarFilefor every entry and in some cases theJarFileinitialization could be expensive (for example when he JAR is signed). Using theURLConnectioncache would solve the performance issues but opens new one. The cache is global for the build so if the JAR file have changed during the build you may get the cached instance (see plexus-io#2).Modify
PlexusIoZipFileResourceCollectionto useJarFiledirectly instead of usingPlexusIoURLResource. That would solve solves the two issues -JarFileis initialized once so there is no performance penalty and it is local so if the file changed during the build the latest version will be picked up.Under the hood
PlexusIoURLResourceusesJarFileas well and keepingZipFileResourceextendingPlexusIoURLResourcewould both keep the behaviour and interface backward compatible.