[ https://issues.apache.org/jira/browse/ARTEMIS-2604?focusedWorklogId=375448&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-375448 ]
ASF GitHub Bot logged work on ARTEMIS-2604: ------------------------------------------- Author: ASF GitHub Bot Created on: 22/Jan/20 07:37 Start Date: 22/Jan/20 07:37 Worklog Time Spent: 10m Work Description: franz1981 commented on issue #2950: ARTEMIS-2604 Save Message decoding when unnecessary URL: https://github.com/apache/activemq-artemis/pull/2950#issuecomment-577046495 @clebertsuconic Sadly seems that while exchanging messages, we trigger the full decode anyway, due to other properties accessed eg `CoreMessage:.getRoutingType` ![image](https://user-images.githubusercontent.com/13125299/72872704-48d4a580-3cee-11ea-8a32-8c13934fd45d.png) But, good news about the change already done: that's from *ARTEMIS 2.11* ![image](https://user-images.githubusercontent.com/13125299/72873309-c6e57c00-3cef-11ea-988c-b0b72dbc0ef4.png) And the same heap configuration ie `-XX:+UseG1GC -Xms4G -Xmx4G -XX:+AlwaysPreTouch` for *this PR*: ![image](https://user-images.githubusercontent.com/13125299/72873378-f98f7480-3cef-11ea-893a-9e13b9b604e2.png) The difference in max pause time is just too much (*7 sec vs 150 ms* to compare them :) There are other metrics, but the point is that we won't produce tons of garbage to decode messages if the `ScheduledDeliveryTime` property isn't set and should cover the most of the cases. The previous results are obtained by loading a journal created by running 30 pairs of Core JMS producers/consumers sending persistent byte messages of 10 bytes size until the broker start to page, with slow consumers (ie performing `Thread::sleep(200)` while consuming messages): this has filled the message journal with a mix of unacked/acked small messages that would stress the heap/GC part of the loading activity instead of the disk. Indeed the whole `data` folder is just `2.5 GB`. I would try with different sizes too, to put under stress the new search property method, that is being benchmarked separately to make it as fast as possible. I strongly encourage any user that make heavy use of the journal and has stress tests while loading journal to try this (@michaelandrepearce @wy96f ?) being aware that ATM it just improve this for Core protocol, but with some help, will get the same for AMQP as well. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org Issue Time Tracking ------------------- Worklog Id: (was: 375448) Time Spent: 50m (was: 40m) > Save Message decoding when unnecessary > -------------------------------------- > > Key: ARTEMIS-2604 > URL: https://issues.apache.org/jira/browse/ARTEMIS-2604 > Project: ActiveMQ Artemis > Issue Type: Improvement > Affects Versions: 2.12.0 > Reporter: Francesco Nigro > Priority: Major > Time Spent: 50m > Remaining Estimate: 0h > > There are few cases eg while loading journal or on common message routing > where decoding a full message isn't necessary if certain properties aren't > present. > Performing a "brute-force" scan into the message not-yet-decoded raw contant > would be welcome if compared to a full decoding that could be saved for later > (or never). -- This message was sent by Atlassian Jira (v8.3.4#803005)