[
https://issues.apache.org/jira/browse/CAMEL-4009?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Arkadi Shishlov updated CAMEL-4009:
-----------------------------------
Attachment:
apache-camel-r1128875-fast-TypeConverter-loading-in-uber-jar-try2.diff
After thinking a little bit I implemented in-memory cache of jar entries names
via SoftRefence in loadImplementationsInJar(). Please see attached path \-try2.
The memory will be release by GC eventually. I'm not sure the
PackageScanClassResolver meant to be thread-safe or not - current usage is
single threaded, but anyway, memory ordering is protected by volatile and hash
is synchronized to be on the safe side.
The peformance is good - same 2 seconds as with previous patch vs 15 sec of
original approach.
> Fast @Converter loading in uber-jar
> -----------------------------------
>
> Key: CAMEL-4009
> URL: https://issues.apache.org/jira/browse/CAMEL-4009
> Project: Camel
> Issue Type: Improvement
> Components: camel-core
> Affects Versions: 2.7.0
> Reporter: Arkadi Shishlov
> Priority: Minor
> Labels: uber-jar
> Fix For: Future
>
> Attachments:
> apache-camel-r1127037-fast-TypeConverter-loading-in-uber-jar.diff,
> apache-camel-r1128875-fast-TypeConverter-loading-in-uber-jar-try2.diff
>
>
> Currently, the @Converter converters are loaded as follows
> {noformat}
> for package in TypeConverter meta-inf resource
> get package url from classloader
> if url is jar
> for entries in jar
> if entry match package
> check annotation
> {noformat}
> This is very inefficient approach when application is packaged into uber-jar
> as expensive iteration is performed multiple times over same jar.
> The attached patch changes loop into:
> {noformat}
> for package in TypeConverter meta-inf resource
> get package url from classloader
> create map url -> package names
> for url in distinct url
> if jar
> for entries in jar
> for package in packages
> if entry match package
> check annotation
> {noformat}
> With this patch I was able to cut startup time significantly from 15sec spent
> on @Converter-s load to 2sec on 45MB 29k entries uber-jar.
> I understand that the solution is not pretty but resolves a very practical
> problem. Let me know if you want me to polish it and update the javadoc.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
