Bug 1088600

Summary: [GSS] (6.2.x) Full scheduledReferences traversal in every call to ScheduledDeliveryHandlerImpl$ScheduledDeliveryRunnable.run()
Product: [JBoss] JBoss Enterprise Application Platform 6 Reporter: Jimmy Wilson <jawilson>
Component: HornetQAssignee: Clebert Suconic <csuconic>
Status: CLOSED CURRENTRELEASE QA Contact: Martin Svehla <msvehla>
Severity: unspecified Docs Contact: Russell Dickenson <rdickens>
Priority: unspecified    
Version: 6.2.1CC: msvehla
Target Milestone: CR2   
Target Release: EAP 6.2.3   
Hardware: Unspecified   
OS: Unspecified   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
HornetQ would consume a lot of CPU if lots of messages were scheduled with a small interval between each message because it would transverse the whole list of messages at every scheduled time.
 Story Points: ---
Clone Of:
: 1089838 (view as bug list) Environment:
Last Closed: 2014-06-09 12:47:59 UTC Type: Bug
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Bug Depends On: 1089838    
Bug Blocks: 1067532, 1084639    

Description Jimmy Wilson 2014-04-16 20:57:49 UTC 
The method ScheduledDeliveryHandlerImpl$ScheduledDeliveryRunnable.run() unnecessarily does a full traversal of the scheduledReferences linked list on each and every call.

When the number of scheduled messages is very large, this traversal severely impacts HornetQ's performance when consuming scheduled messages.

We are using HornetQ 2.2.24 in an application where we schedule ~400,000 messages in a single queue to be consumed several hours into the future, with large bursts of messages to be consumed as quickly as possible. In this application, we have found that HornetQ's rate of consuming scheduled messages is limited by CPU. From performance profiling, we found that the method

org.hornetq.core.server.impl.ScheduledDeliveryHandlerImpl$ScheduledDeliveryRunnable.run()

is consuming approximately 70-80% of the CPU cycles during bursts. Looking at the code, this is due to the full traversal of the scheduledReferences linked list on every call to run():
https://github.com/hornetq/hornetq/blob/HornetQ_2_2_24_EAP_GA/src/main/org/hornetq/core/server/impl/ScheduledDeliveryHandlerImpl.java#L181
Comment 1 Martin Svehla 2014-05-06 14:08:18 UTC 
Verified with EAP 6.2.3.CP.CR2 / HornetQ 2.3.14.1.Final