[infinispan-issues] [JBoss JIRA] (ISPN-4390) CHMv8 leaks ThreadLocal

[ https://issues.jboss.org/browse/ISPN-4390?page=com.atlassian.jira.plugin.... ] Dan Berindei edited comment on ISPN-4390 at 4/7/16 11:42 AM: ------------------------------------------------------------- I should add that our CHMv8 code comes from Doug Lea's repository, not from OpenJDK, so the original code would also also have a problem with "leaked" {{ThreadLocal}}s. Still, I have a hunch that the non-deterministic cleanup of stale {{ThreadLocalMap}} entries isn't such a big problem when you have a limited number of {{ThreadLocals}}. It's only when the {{ThreadLocalMap}} grows very large, e.g. because we create one per cache in {{InvocationContextContainerImpl}}, that {{cleanSomeSlots()}} becomes ineffective and it takes multiple rounds to clean up the stale entries created by the previous deployment. That and maybe worker threads that don't process requests often enough :) was (Author: dan.berindei): I should add that our CHMv8 code comes from Doug Lea's repository, not from OpenJDK, so the original code would also also have a problem with "leaked" the {{ThreadLocal}}. Still, I have a hunch that the non-deterministic cleanup of stale {{ThreadLocalMap}} entries isn't such a big problem when you have a limited number of {{ThreadLocals}}. It's only when the {{ThreadLocalMap}} grows very large, e.g. because we create one per cache in {{InvocationContextContainerImpl}}, that {{cleanSomeSlots()}} becomes ineffective and it takes multiple rounds to clean up the stale entries created by the previous deployment. That and maybe worker threads that don't process requests often enough :) -- This message was sent by Atlassian JIRA (v6.4.11#64026)