Bug #16382
closed
Improve performance of policy generation writer
Added by François ARMAND almost 5 years ago.
Updated over 4 years ago.
Description
We lose quite some performance in policy generation in 6.0.
The ticket (and companion pr) are here to try to see if we can get back some of it.
- Status changed from New to In progress
- Assignee set to François ARMAND
- Status changed from In progress to Pending technical review
- Assignee changed from François ARMAND to Nicolas CHARLES
- Pull Request set to https://github.com/Normation/rudder/pull/2666
- Status changed from Pending technical review to Pending release
- Status changed from Pending release to In progress
I uncorrectly push the branch on a branch in "origin", and redmine thought it was corrected. I'm not sure how to remove all the links.
It should be corrected now.
- Target version changed from 6.0.1 to 6.0.2
- Target version changed from 6.0.2 to 6.0.3
Two things:
- movePromisesToFinalPosition(folders: Seq[NodePoliciesPaths]) is not parallel. It seems to be because the order seems to be important, certainly because of relay. It feels like it can be parallelised (at worse, we can parallelize each policy server direct children, recursively).
- the whole fillTemplate is monothreaded while we use a paralleleSequence. Now, the parallelism is made on fibers, not thread, but still it's strange that they don't use more thread. John advice to use MUCH bigger parallelism factor than core/2.
- Related to Bug #16557: CachedFindRuleNodeStatusReports is a huge source of contention added
- Status changed from In progress to Pending release
- Fix check changed from To do to Checked
- Name check changed from To do to Reviewed
This bug has been fixed in Rudder 6.0.3 which was released today.
- Related to Bug #16629: Use only one ZIO threadpool to enhance performances added
- Status changed from Pending release to Released
Also available in: Atom
PDF
Updated by 
Updated by 
Updated by