With DSpace 6.0 as released you can intentionally create idle in transaction database connections simply by calling the start page. The connections does not recover and flood the connection pool. Calling the start page e.g. 100 times in 5 seconds fills the DB pool and DSpace answers only with a 500 internal server error until the next tomcat restart. We tested this in different environments (Tomcat 7 and 8, Java 7 and 8, Windows and Linux, Apache as (reverse) Proxy and not etc.) and with different configurations. I even tested it with a fresh installed vanilla DSpace. Allowing DSpace/PostgreSQL a larger DB pool or longer DB timeouts does not change the behaviour. I also tried all possible configurations from the last DCAT meeting: https://wiki.duraspace.org/display/cmtygp/DCAT+Meeting+April+2017 @christian.scheible kindly reproduced the bug at his institution and it seems that only JSPUI is affected. I verified that with JSPUI version 5.x is not affected. With the help of @pbecker I was able to test it against demo.dspace.org/jspui I assume that not only the home page is affected. Currently the consequence of this bug is, that we can not roll out the 6.0 release to production. Especially local crawlers like the Google Search Appliance triggers to much of the idle in transaction connections by crawling patterns like this:
|