Details
-
Improvement
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
When a JD returns no work, each worker JP thread sleeps for a bit and than asks JD again. At large scaleout many threads repeat this cycle causing the JD to spend a lot of time handling WI requests.
Modify the JP to use a single thread in each process to retry request for work. While a single thread repeatedly retries other threads block until:
1) Retry thread receives work item
2) JP terminates