[
https://issues.apache.org/jira/browse/STORM-898?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15062741#comment-15062741
]
ASF GitHub Bot commented on STORM-898:
--------------------------------------
Github user d2r commented on a diff in the pull request:
https://github.com/apache/storm/pull/921#discussion_r47956948
--- Diff:
storm-core/src/jvm/backtype/storm/scheduler/resource/ResourceAwareScheduler.java
---
@@ -128,56 +308,121 @@ private void updateSupervisorsResources(Cluster
cluster, Topologies topologies)
cluster.setSupervisorsResourcesMap(supervisors_resources);
}
- private Map<String, Double> getUserConf() {
- Map<String, Double> ret = new HashMap<String, Double>();
- ret.put(Config.TOPOLOGY_COMPONENT_RESOURCES_ONHEAP_MEMORY_MB,
- (Double)
_conf.get(Config.TOPOLOGY_COMPONENT_RESOURCES_ONHEAP_MEMORY_MB));
- ret.put(Config.TOPOLOGY_COMPONENT_RESOURCES_OFFHEAP_MEMORY_MB,
- (Double)
_conf.get(Config.TOPOLOGY_COMPONENT_RESOURCES_OFFHEAP_MEMORY_MB));
- return ret;
+ public User getUser(String user) {
+ return this.userMap.get(user);
+ }
+
+ public Map<String, User> getUserMap() {
+ return this.userMap;
}
/**
- * print scheduling for debug purposes
- * @param cluster
+ * Intialize scheduling and running queues
+ *
* @param topologies
+ * @param cluster
*/
- public String printScheduling(Cluster cluster, Topologies topologies) {
- StringBuilder str = new StringBuilder();
- Map<String, Map<String, Map<WorkerSlot,
Collection<ExecutorDetails>>>> schedulingMap = new HashMap<String, Map<String,
Map<WorkerSlot, Collection<ExecutorDetails>>>>();
- for (TopologyDetails topo : topologies.getTopologies()) {
- if (cluster.getAssignmentById(topo.getId()) != null) {
- for (Map.Entry<ExecutorDetails, WorkerSlot> entry :
cluster.getAssignmentById(topo.getId()).getExecutorToSlot().entrySet()) {
- WorkerSlot slot = entry.getValue();
- String nodeId = slot.getNodeId();
- ExecutorDetails exec = entry.getKey();
- if (schedulingMap.containsKey(nodeId) == false) {
- schedulingMap.put(nodeId, new HashMap<String,
Map<WorkerSlot, Collection<ExecutorDetails>>>());
- }
- if
(schedulingMap.get(nodeId).containsKey(topo.getId()) == false) {
- schedulingMap.get(nodeId).put(topo.getId(), new
HashMap<WorkerSlot, Collection<ExecutorDetails>>());
- }
- if
(schedulingMap.get(nodeId).get(topo.getId()).containsKey(slot) == false) {
-
schedulingMap.get(nodeId).get(topo.getId()).put(slot, new
LinkedList<ExecutorDetails>());
- }
-
schedulingMap.get(nodeId).get(topo.getId()).get(slot).add(exec);
+ private void initUsers(Topologies topologies, Cluster cluster) {
+ this.userMap = new HashMap<String, User>();
+ Map<String, Map<String, Double>> userResourcePools =
this.getUserResourcePools();
+ LOG.debug("userResourcePools: {}", userResourcePools);
+
+ for (TopologyDetails td : topologies.getTopologies()) {
+ //Get user that submitted topology. If topology submitter is
null or empty string, the topologySubmitter
+ //will be set to anonymous
+ String topologySubmitter = td.getTopologySubmitter();
+ //additional safety check to make sure that topologySubmitter
is going to be a valid value
+ if (topologySubmitter == null || topologySubmitter.equals(""))
{
+ LOG.error("Cannot determine user for topology {}. Will
skip scheduling this topology", td.getName());
+ continue;
+ }
+ if (!this.userMap.containsKey(topologySubmitter)) {
+ this.userMap.put(topologySubmitter, new
User(topologySubmitter, userResourcePools.get(topologySubmitter)));
+ }
+ if (cluster.getUnassignedExecutors(td).size() > 0) {
+ LOG.debug("adding td: {} to pending queue", td.getName());
+
this.userMap.get(topologySubmitter).addTopologyToPendingQueue(td);
+ } else {
+ LOG.debug("adding td: {} to running queue with existing
status: {}", td.getName(), cluster.getStatusMap().get(td.getId()));
+
this.userMap.get(topologySubmitter).addTopologyToRunningQueue(td);
+ if (cluster.getStatusMap().get(td.getId()) == null ||
cluster.getStatusMap().get(td.getId()).equals("")) {
+ cluster.setStatus(td.getId(), "Fully Scheduled");
}
}
}
+ }
- for (Map.Entry<String, Map<String, Map<WorkerSlot,
Collection<ExecutorDetails>>>> entry : schedulingMap.entrySet()) {
- if (cluster.getSupervisorById(entry.getKey()) != null) {
- str.append("/** Node: " +
cluster.getSupervisorById(entry.getKey()).getHost() + "-" + entry.getKey() + "
**/\n");
- } else {
- str.append("/** Node: Unknown may be dead -" +
entry.getKey() + " **/\n");
+ private void initialize(Topologies topologies, Cluster cluster) {
+ this.cluster = cluster;
+ this.topologies = topologies;
+ this.nodes = new RAS_Nodes(this.cluster, this.topologies);
+ initUsers(topologies, cluster);
+ }
+
+ /**
+ * Get resource guarantee configs
+ *
+ * @return
+ */
+ private Map<String, Map<String, Double>> getUserResourcePools() {
+ Object raw =
this.conf.get(Config.RESOURCE_AWARE_SCHEDULER_USER_POOLS);
+ Map<String, Map<String, Double>> ret = new HashMap<String,
Map<String, Double>>();
+
+ if (raw != null) {
+ for (Map.Entry<String, Map<String, Number>> UserPoolEntry :
((Map<String, Map<String, Number>>) raw).entrySet()) {
+ String user = UserPoolEntry.getKey();
+ ret.put(user, new HashMap<String, Double>());
+ for (Map.Entry<String, Number> resourceEntry :
UserPoolEntry.getValue().entrySet()) {
+ ret.get(user).put(resourceEntry.getKey(),
resourceEntry.getValue().doubleValue());
+ }
}
- for (Map.Entry<String, Map<WorkerSlot,
Collection<ExecutorDetails>>> topo_sched :
schedulingMap.get(entry.getKey()).entrySet()) {
- str.append("\t-->Topology: " + topo_sched.getKey() + "\n");
- for (Map.Entry<WorkerSlot, Collection<ExecutorDetails>> ws
: topo_sched.getValue().entrySet()) {
- str.append("\t\t->Slot [" + ws.getKey().getPort() + "]
-> " + ws.getValue() + "\n");
+ }
+
+ Map fromFile =
Utils.findAndReadConfigFile("user-resource-pools.yaml", false);
+ Map<String, Map<String, Number>> tmp = (Map<String, Map<String,
Number>>) fromFile.get(Config.RESOURCE_AWARE_SCHEDULER_USER_POOLS);
+ if (tmp != null) {
+ for (Map.Entry<String, Map<String, Number>> UserPoolEntry :
tmp.entrySet()) {
+ String user = UserPoolEntry.getKey();
+ ret.put(user, new HashMap<String, Double>());
+ for (Map.Entry<String, Number> resourceEntry :
UserPoolEntry.getValue().entrySet()) {
+ ret.get(user).put(resourceEntry.getKey(),
resourceEntry.getValue().doubleValue());
}
}
}
- return str.toString();
+ return ret;
+ }
+
+ private SchedulingState checkpointSchedulingState() {
+ LOG.debug("/*********Checkpoint scheduling state************/");
+ for (User user : this.getUserMap().values()) {
+ LOG.debug(user.getDetailedInfo());
+ }
+ LOG.debug(ResourceUtils.printScheduling(this.cluster,
this.topologies));
+ LOG.debug("nodes:\n{}", this.nodes);
+ LOG.debug("/*********End************/");
+ return new SchedulingState(this.userMap, this.cluster,
this.topologies, this.nodes, this.conf);
+ }
+
+ private void restoreCheckpointSchedulingState(SchedulingState
schedulingState) {
+ LOG.debug("/*********restoring scheduling state************/");
+ //reseting cluster
+ //Cannot simply set this.cluster=schedulingState.cluster since
clojure is immutable
+
this.cluster.setAssignments(schedulingState.cluster.getAssignments());
+
this.cluster.setSupervisorsResourcesMap(schedulingState.cluster.getSupervisorsResourcesMap());
+ this.cluster.setStatusMap(schedulingState.cluster.getStatusMap());
+
this.cluster.setResourcesMap(schedulingState.cluster.getResourcesMap());
+ //don't need to explicitly set data structues like Cluster since
nothing can really be changed
+ //unless this.topologies is set to another object
+ this.topologies = schedulingState.topologies;
+ this.conf = schedulingState.conf;
+ this.userMap = schedulingState.userMap;
+ this.nodes = schedulingState.nodes;
+
+ for (User user : this.getUserMap().values()) {
+ LOG.debug(user.getDetailedInfo());
+ }
+ LOG.debug(ResourceUtils.printScheduling(cluster, topologies));
+ LOG.debug("nodes:\n{}", this.nodes);
+ LOG.debug("/*********End************/");
--- End diff --
Could we just make a SchedulingState a member of the
ResourceAwareScheduler? Then restoring would be simply pointing to the desired
SchedulingState instance.
> Add priorities and per user resource guarantees to Resource Aware Scheduler
> ---------------------------------------------------------------------------
>
> Key: STORM-898
> URL: https://issues.apache.org/jira/browse/STORM-898
> Project: Apache Storm
> Issue Type: New Feature
> Components: storm-core
> Reporter: Robert Joseph Evans
> Assignee: Boyang Jerry Peng
> Attachments: Resource Aware Scheduler for Storm.pdf
>
>
> In a multi-tenant environment we would like to be able to give individual
> users a guarantee of how much CPU/Memory/Network they will be able to use in
> a cluster. We would also like to know which topologies a user feels are the
> most important to keep running if there are not enough resources to run all
> of their topologies.
> Each user should be able to specify if their topology is production, staging,
> or development. Within each of those categories a user should be able to give
> a topology a priority, 0 to 10 with 10 being the highest priority (or
> something like this).
> If there are not enough resources on a cluster to run a topology assume this
> topology is running using resources and find the user that is most over their
> guaranteed resources. Shoot the lowest priority topology for that user, and
> repeat until, this topology is able to run, or this topology would be the one
> shot. Ideally we don't actually shoot anything until we know that we would
> have made enough room.
> If the cluster is over-subscribed and everyone is under their guarantee, and
> this topology would not put the user over their guarantee. Shoot the lowest
> priority topology in this workers resource pool until there is enough room to
> run the topology or this topology is the one that would be shot. We might
> also want to think about what to do if we are going to shoot a production
> topology in an oversubscribed case, and perhaps we can shoot a non-production
> topology instead even if the other user is not over their guarantee.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
