[
https://issues.apache.org/jira/browse/STORM-1450?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15101195#comment-15101195
]
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_r49821028
--- Diff:
storm-core/src/jvm/org/apache/storm/scheduler/resource/RAS_Node.java ---
@@ -43,32 +43,84 @@
*/
public class RAS_Node {
private static final Logger LOG =
LoggerFactory.getLogger(RAS_Node.class);
- private Map<String, Set<WorkerSlot>> _topIdToUsedSlots = new
HashMap<String, Set<WorkerSlot>>();
- private Set<WorkerSlot> _freeSlots = new HashSet<WorkerSlot>();
+
+ //A map consisting of all workers on the node.
+ //The key of the map is the worker id and the value is the
corresponding workerslot object
+ Map<String, WorkerSlot> _slots = new HashMap<String, WorkerSlot> ();
+
+ // A map describing which topologies are using which slots on this
node. The format of the map is the following:
+ // {TopologyId -> {WorkerId -> {Executors}}}
+ private Map<String, Map<String, Collection<ExecutorDetails>>>
_topIdToUsedSlots = new HashMap<String, Map<String,
Collection<ExecutorDetails>>>();
+
private final String _nodeId;
private String _hostname;
private boolean _isAlive;
private SupervisorDetails _sup;
- private Double _availMemory;
- private Double _availCPU;
- private Cluster _cluster;
- private Topologies _topologies;
+ private Double _availMemory = 0.0;
+ private Double _availCPU = 0.0;
+ private final Cluster _cluster;
+ private final Topologies _topologies;
- public RAS_Node(String nodeId, Set<Integer> allPorts, boolean isAlive,
- SupervisorDetails sup, Cluster cluster, Topologies
topologies) {
+ public RAS_Node(String nodeId, SupervisorDetails sup, Cluster cluster,
Topologies topologies, Map<String, WorkerSlot> workerIdToWorker, Map<String,
Map<String, Collection<ExecutorDetails>>> assignmentMap) {
+ //Node ID and supervisor ID are the same.
_nodeId = nodeId;
- _isAlive = isAlive;
- if (_isAlive && allPorts != null) {
- for (int port : allPorts) {
- _freeSlots.add(new WorkerSlot(_nodeId, port));
- }
- _sup = sup;
+ if (sup == null) {
+ _isAlive = false;
+ } else {
+ _isAlive = !cluster.isBlackListed(_nodeId);
+ }
+
+ _cluster = cluster;
+ _topologies = topologies;
+
+ // initialize slots for this node
+ if (workerIdToWorker != null) {
+ _slots = workerIdToWorker;
+ }
+
+ //initialize assignment map
+ if (assignmentMap != null) {
+ _topIdToUsedSlots = assignmentMap;
+ }
+
+ //check if node is alive
+ if (_isAlive && sup != null) {
_hostname = sup.getHost();
+ _sup = sup;
_availMemory = getTotalMemoryResources();
_availCPU = getTotalCpuResources();
+
+ LOG.debug("Found a {} Node {} {}",
+ _isAlive ? "living" : "dead", _nodeId,
sup.getAllPorts());
+ LOG.debug("resources_mem: {}, resources_CPU: {}",
sup.getTotalMemory(), sup.getTotalCPU());
+ //intialize resource usages on node
+ intializeResources();
+ }
+ }
+
+ /**
+ * intializes resource usages on node
--- End diff --
will fix
> 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)
