Agreed, but maybe the servicemix-bpe component could be enhanced. If i recall, the main problem is that bpe does not have a way to provide an asynchronous response when a request/response pattern is used (need to check once again, though). However, for InOnly MEPs, we could easily use an asynchonous send. ... and you could use bpel correlations to use InOnly meps for request and response instead of an InOut mep. Would you please raise a JIRA ? imho, this problem is specific to servicemix-bpe and could be fixed asap.
On 8/9/06, Ramon Buckland <[EMAIL PROTECTED]> wrote:
It would defintiely be something that can cause problems. We are currently trying to resolve this. It seems to rear it's head (SM-512) on Windows, more than Linux, this has me baffled servicemix-bpe ... JbiInvokeAction only ever sends out a sendSync() when invoking a partner (JBI Component). The BPEL spec discussed correlation-sets as method of performing async calls. We can see that <flows> and <sequences> control the threading of servicemix-bpe but because there is a sendSync for every message going out, it means the threadpool can get eaten up pretty quickly. my 2 cents. Guillaume Nodet wrote: > SM-512 is a pretty big issue, but it does not have by itself any solution > (see my comment in JIRA). > However, SM-521 could be implemented to be able to easily tune thread > pools > and queue sizes to > avoid deadlocks. Though it may require quite an amount of > refactoring, i'm > wondering if we should > implement it for 3.0 ? >
-- Cheers, Guillaume Nodet
