[YUNIKORN-1746] Improve the performance of nodeInfoListerImpl.List()

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.3.0
• Labels: pull-request-available
• Target Version: 1.3.0
• Assignee: Peter Bacsko
• Reporter: Peter Bacsko
• Created: 17/May/23 17:18
• Updated: 20/Jun/23 14:41
• Resolved: 25/May/23 15:40

Description

Performance data (cpu and heap profiles) indicate that we can improve the performance of nodeInfoListerImpl.List().

This is the code which is located in the shim:

func (n nodeInfoListerImpl) List() ([]*framework.NodeInfo, error) {
	nodes := n.cache.GetNodesInfoMap()
	nodeList := make([]*framework.NodeInfo, 0, len(nodes))
	for _, node := range nodes {
		nodeList = append(nodeList, node)
	}

	return nodeList, nil
}

We're supposed to be having a read lock on the cache, which is done outside this method.

The problem is, the nodes info map only changes if a node joins or leaves the cluster. A lot of copying is done here, we generate a lot of waste. This method is called very frequently, because predicates need this list and predicates always run for each pod by default.

Issue Links

• links to: GitHub Pull Request #595