If you do nothing on dispatch policies in OSB proxy or business services, all is done in the Default Workmanager. But since some constructions, not only service call-outs, need other threads to finish the job, you can get stuck threads because the workmanager's threadpool gets empty having all or near to all threads waiting, leaving no threads to pickup work to free the others.
More on this in the terrific blog of Anthony Reynolds on the subject: Following the thread.
By the way, I've seen that some people by default use service call-outs for almost everything. But the default use should be the routing node with a route activity. Even in some services you need to gather information from several sources, while you can only have one route node, pick or choose a 'driving-service' to use in the Routing node. Just like creating a query on several tables where you have to choose a 'driving-table'. Then use service call outs only to do the extra enrichment.
From that earlier project I got the following recommendations, based on the blog of Anthony Reynolds. Since I refer back to it regularly, I think it would be good to share it.
For OSB to work optimally and prevent floading WebLogic’s threadpool with hogged/stuck threads you should create 3 FairShareRequest classes in ratio of 33/33/33, to distinguish different “kinds of threadpools”.
Then create 4 workmanagers:
- FTPPollingWorkManager: file based inbound OSB proxy services. Polling a filesystem (or FTP). Uses FairShareReqClass-1, and ignores stuck threads.
- InboudWorkManager: inbound OSB proxy services, not polling file based. Also uses FairShareReqClass-1, not ignoring stuck threads.
- CallOutWorkManager: Service Call Out operations in a OSB proxy. Uses FairShareReqClass-2.
- DeliveryWorkManager: outbound business services in OSB. Uses FairShareReqClass-3.
I don't have screendumps of 12c at hand. But the idea would be the same there. I haven't learned that the thread model in 12c is architecturally different.
No comments :
Post a Comment