[
https://issues.apache.org/jira/browse/STORM-1450?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15101200#comment-15101200
]
ASF GitHub Bot commented on STORM-1450:
---------------------------------------
Github user jerrypeng commented on a diff in the pull request:
https://github.com/apache/storm/pull/1016#discussion_r49821113
--- Diff:
storm-core/src/jvm/org/apache/storm/scheduler/resource/RAS_Node.java ---
@@ -247,19 +253,23 @@ private void freeMemory(double amount) {
private void freeCPU(double amount) {
LOG.debug("freeing {} CPU on node...avail CPU: {}", amount,
getHostname(), _availCPU);
- if ((_availCPU + amount) > getAvailableCpuResources()) {
- LOG.warn("Freeing more CPU than there exists!");
+ if ((_availCPU + amount) > getTotalCpuResources()) {
+ LOG.warn("Freeing more CPU than there exists! CPU trying to
free: {} Total CPU on Node: {}", (_availMemory + amount),
getTotalCpuResources());
return;
}
_availCPU += amount;
}
+ /**
+ * get the amount of memory used by a worker
+ */
public double getMemoryUsedByWorker(WorkerSlot ws) {
TopologyDetails topo = findTopologyUsingWorker(ws);
if (topo == null) {
return 0.0;
}
Collection<ExecutorDetails> execs = getExecutors(ws, _cluster);
+ LOG.info("getMemoryUsedByWorker executors: {}", execs);
--- End diff --
will get rid of
> Fix bugs and refactor code in ResourceAwareScheduler
> ----------------------------------------------------
>
> Key: STORM-1450
> URL: https://issues.apache.org/jira/browse/STORM-1450
> Project: Apache Storm
> Issue Type: Improvement
> Reporter: Boyang Jerry Peng
> Assignee: Boyang Jerry Peng
> Priority: Minor
>
> Code refactored:
> 1. Refactor RAS_Nodes. Pushed some of the functionality in to RAS_Nodes.
> Each RAS_Node will now be initialized with a map of all its assignments.
> Each RAS_Node will also figure out resources used and available. Removed
> unnecessary functions.
> 2. Made WorkerSlot immutable so that a scheduling strategy won't mistakenly
> modify it
> 3. Added a wrapping layer for RAS_Node to feed into scheduling strategies so
> that the semantics of what a scheduling strategy should do will be more
> clear. Each scheduling strategy shouldn't be actually assigning anything.
> The strategy should only calculate a scheduling.
> Bug fixes:
> 1. Minor bug in displaying the assigned resources for a supervisor on the UI.
> The function updateSupervisorResources was placed in the wrong place
> 2. Minor bug fix in freeing memory in RAS_Node there was some wrong math that
> was done.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
