Yes, I started looking into this because of a memory regression. While upgrading our project from Java 17 to Java 21, I noticed an increase in memory consumption. After dumping the heap and analyzing the differences, I found that there are a few thousand instances of the class "java.util.jar.Attributes$Name" which all contain the same string which is "SHA-384-Digest". In our case, it was a 2MB increase in a process that uses 35MB normally so that's almost a 5% increase. The issue here is that we are loading multiple jars that each contain a MANIFEST file which contains hundreds of classes and so many times the string "SHA-384-Digest".
It would be nice if the fix could be backported to java 21 since it's the current LTS that's impacted. On Fri, Apr 4, 2025 at 7:07 PM Jaikiran Pai <jai.forums2...@gmail.com> wrote: > The caching of some attribute names in the java.util.jar.Attributes > class appears to have been introduced in > https://bugs.openjdk.org/browse/JDK-6805750 (RFR > https://mail.openjdk.org/pipermail/core-libs-dev/2018-April/052697.html). > > Given the precedence, it may be OK to add "SHA-384-Digest" to that set, > but I don't know how much (if any) would it help with the performance. > Did you happen to notice any performance issues which prompted you to > look at this cache? > > -Jaikiran > > On 04/04/25 10:25 pm, Jaikiran Pai wrote: > > Since this is about caching values in the java.util.jar.Attributes > > class, the discussion is more appropriate in the core-libs-dev mailing > > list. I've added it to the "To" now and "Bcc"ed the jdk-dev mailing > > list. Please subscribe to core-libs-dev > > https://mail.openjdk.org/mailman/listinfo/core-libs-dev if you haven't > > already and we can continue the discussion there. > > > > -Jaikiran > > > > On 04/04/25 8:54 pm, Ayman wrote: > >> Hello, > >> > >> Starting from JDK 19, SHA-384 is replacing SHA-256 as the default > >> digest algorithm in the jarsigner: > >> https://bugs.openjdk.org/browse/JDK-8283475 > >> > >> This means that the string "SHA-384" is now written multiples times > >> is the MANIFEST files, this has been tackled in the past by adding > >> the string "SHA-256" as a KNOWN_NAME that shouldn't be duplicated in > >> memory: > >> > https://github.com/openjdk/jdk/blob/master/src/java.base/share/classes/java/util/jar/Attributes.java#L729 > >> > >> > >> Is it possible to do the same with "SHA-384"? > >> > >> Thanks, >
