All our calls to retrieve email/establish a connection to the Mail Server do not specify a timeout for the connection. The default behavior is to block indefinitely; if this happens, all resources associated with that thread are never released. This is responsible for https://support.atlassian.com/browse/CSP-14098 as we obtain a cluster wide lock and do not release until the mail is retrieved. No doubt other bugs exist.

User Symptoms: Confluence performance degrades over time, particularly during load. A restart resolves the issue temporarily. A thread dump while Confluence is slow shows many threads waiting on mail connections