YARN-10343. Legacy RM UI should include labeled metrics for allocated, total, and reserved resources. Contributed by Eric Payne

This commit is contained in:
Jonathan Hung 2020-07-28 13:44:17 -07:00
parent 7350773b69
commit ffb920de2a
4 changed files with 72 additions and 9 deletions

View File

@ -200,6 +200,10 @@ public class ResourceUsage extends AbstractResourceUsage {
return _getAll(ResourceType.USED); return _getAll(ResourceType.USED);
} }
public Resource getAllReserved() {
return _getAll(ResourceType.RESERVED);
}
// Cache Used // Cache Used
public Resource getCachedUsed() { public Resource getCachedUsed() {
return _get(NL, ResourceType.CACHED_USED); return _get(NL, ResourceType.CACHED_USED);

View File

@ -22,6 +22,7 @@ import org.apache.hadoop.util.StringUtils;
import org.apache.hadoop.yarn.api.records.ResourceTypeInfo; import org.apache.hadoop.yarn.api.records.ResourceTypeInfo;
import org.apache.hadoop.yarn.server.resourcemanager.ResourceManager; import org.apache.hadoop.yarn.server.resourcemanager.ResourceManager;
import org.apache.hadoop.yarn.server.resourcemanager.webapp.dao.ClusterMetricsInfo; import org.apache.hadoop.yarn.server.resourcemanager.webapp.dao.ClusterMetricsInfo;
import org.apache.hadoop.yarn.server.resourcemanager.webapp.dao.ResourceInfo;
import org.apache.hadoop.yarn.server.resourcemanager.webapp.dao.SchedulerInfo; import org.apache.hadoop.yarn.server.resourcemanager.webapp.dao.SchedulerInfo;
import org.apache.hadoop.yarn.server.resourcemanager.webapp.dao.UserMetricsInfo; import org.apache.hadoop.yarn.server.resourcemanager.webapp.dao.UserMetricsInfo;
@ -60,7 +61,38 @@ public class MetricsOverviewTable extends HtmlBlock {
ClusterMetricsInfo clusterMetrics = new ClusterMetricsInfo(this.rm); ClusterMetricsInfo clusterMetrics = new ClusterMetricsInfo(this.rm);
DIV<Hamlet> div = html.div().$class("metrics"); DIV<Hamlet> div = html.div().$class("metrics");
long usedMemoryBytes = 0;
long totalMemoryBytes = 0;
long reservedMemoryBytes = 0;
long usedVCores = 0;
long totalVCores = 0;
long reservedVCores = 0;
if (clusterMetrics.getCrossPartitionMetricsAvailable()) {
ResourceInfo usedAllPartitions =
clusterMetrics.getTotalUsedResourcesAcrossPartition();
ResourceInfo totalAllPartitions =
clusterMetrics.getTotalClusterResourcesAcrossPartition();
ResourceInfo reservedAllPartitions =
clusterMetrics.getTotalReservedResourcesAcrossPartition();
usedMemoryBytes = usedAllPartitions.getMemorySize() * BYTES_IN_MB;
totalMemoryBytes = totalAllPartitions.getMemorySize() * BYTES_IN_MB;
reservedMemoryBytes = reservedAllPartitions.getMemorySize() * BYTES_IN_MB;
usedVCores = usedAllPartitions.getvCores();
totalVCores = totalAllPartitions.getvCores();
reservedVCores = reservedAllPartitions.getvCores();
// getTotalUsedResourcesAcrossPartition includes reserved resources.
usedMemoryBytes -= reservedMemoryBytes;
usedVCores -= reservedVCores;
} else {
usedMemoryBytes = clusterMetrics.getAllocatedMB() * BYTES_IN_MB;
totalMemoryBytes = clusterMetrics.getTotalMB() * BYTES_IN_MB;
reservedMemoryBytes = clusterMetrics.getReservedMB() * BYTES_IN_MB;
usedVCores = clusterMetrics.getAllocatedVirtualCores();
totalVCores = clusterMetrics.getTotalVirtualCores();
reservedVCores = clusterMetrics.getReservedVirtualCores();
}
div.h3("Cluster Metrics"). div.h3("Cluster Metrics").
table("#metricsoverview"). table("#metricsoverview").
thead().$class("ui-widget-header"). thead().$class("ui-widget-header").
@ -89,13 +121,14 @@ public class MetricsOverviewTable extends HtmlBlock {
clusterMetrics.getAppsFailed() + clusterMetrics.getAppsKilled() clusterMetrics.getAppsFailed() + clusterMetrics.getAppsKilled()
) )
). ).
td(String.valueOf(clusterMetrics.getContainersAllocated())). td(String.valueOf(
td(StringUtils.byteDesc(clusterMetrics.getAllocatedMB() * BYTES_IN_MB)). clusterMetrics.getTotalAllocatedContainersAcrossPartition())).
td(StringUtils.byteDesc(clusterMetrics.getTotalMB() * BYTES_IN_MB)). td(StringUtils.byteDesc(usedMemoryBytes)).
td(StringUtils.byteDesc(clusterMetrics.getReservedMB() * BYTES_IN_MB)). td(StringUtils.byteDesc(totalMemoryBytes)).
td(String.valueOf(clusterMetrics.getAllocatedVirtualCores())). td(StringUtils.byteDesc(reservedMemoryBytes)).
td(String.valueOf(clusterMetrics.getTotalVirtualCores())). td(String.valueOf(usedVCores)).
td(String.valueOf(clusterMetrics.getReservedVirtualCores())). td(String.valueOf(totalVCores)).
td(String.valueOf(reservedVCores)).
__(). __().
__().__(); __().__();

View File

@ -26,6 +26,7 @@ import org.apache.hadoop.yarn.server.resourcemanager.ResourceManager;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.QueueMetrics; import org.apache.hadoop.yarn.server.resourcemanager.scheduler.QueueMetrics;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.ResourceScheduler; import org.apache.hadoop.yarn.server.resourcemanager.scheduler.ResourceScheduler;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler; import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler;
import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue;
@XmlRootElement(name = "clusterMetrics") @XmlRootElement(name = "clusterMetrics")
@XmlAccessorType(XmlAccessType.FIELD) @XmlAccessorType(XmlAccessType.FIELD)
@ -67,6 +68,14 @@ public class ClusterMetricsInfo {
// Total registered resources of the cluster, including all partitions // Total registered resources of the cluster, including all partitions
private ResourceInfo totalClusterResourcesAcrossPartition; private ResourceInfo totalClusterResourcesAcrossPartition;
// Total reserved resources of the cluster, including all partitions.
private ResourceInfo totalReservedResourcesAcrossPartition;
// Total allocated containers across all partitions.
private int totalAllocatedContainersAcrossPartition;
private boolean crossPartitionMetricsAvailable = false;
public ClusterMetricsInfo() { public ClusterMetricsInfo() {
} // JAXB needs this } // JAXB needs this
@ -111,6 +120,11 @@ public class ClusterMetricsInfo {
cs.getRootQueue().getQueueResourceUsage().getAllUsed()); cs.getRootQueue().getQueueResourceUsage().getAllUsed());
totalClusterResourcesAcrossPartition = new ResourceInfo( totalClusterResourcesAcrossPartition = new ResourceInfo(
cs.getClusterResource()); cs.getClusterResource());
totalReservedResourcesAcrossPartition = new ResourceInfo(
cs.getRootQueue().getQueueResourceUsage().getAllReserved());
totalAllocatedContainersAcrossPartition =
((ParentQueue) cs.getRootQueue()).getNumContainers();
crossPartitionMetricsAvailable = true;
} }
} else { } else {
this.totalMB = availableMB + allocatedMB; this.totalMB = availableMB + allocatedMB;
@ -334,4 +348,16 @@ public class ClusterMetricsInfo {
public ResourceInfo getTotalClusterResourcesAcrossPartition() { public ResourceInfo getTotalClusterResourcesAcrossPartition() {
return totalClusterResourcesAcrossPartition; return totalClusterResourcesAcrossPartition;
} }
public ResourceInfo getTotalReservedResourcesAcrossPartition() {
return totalReservedResourcesAcrossPartition;
}
public int getTotalAllocatedContainersAcrossPartition() {
return totalAllocatedContainersAcrossPartition;
}
public boolean getCrossPartitionMetricsAvailable() {
return crossPartitionMetricsAvailable;
}
} }

View File

@ -459,7 +459,7 @@ public class TestRMWebServices extends JerseyTestBase {
Exception { Exception {
assertEquals("incorrect number of elements", 1, json.length()); assertEquals("incorrect number of elements", 1, json.length());
JSONObject clusterinfo = json.getJSONObject("clusterMetrics"); JSONObject clusterinfo = json.getJSONObject("clusterMetrics");
assertEquals("incorrect number of elements", 25, clusterinfo.length()); assertEquals("incorrect number of elements", 27, clusterinfo.length());
verifyClusterMetrics( verifyClusterMetrics(
clusterinfo.getInt("appsSubmitted"), clusterinfo.getInt("appsCompleted"), clusterinfo.getInt("appsSubmitted"), clusterinfo.getInt("appsCompleted"),
clusterinfo.getInt("reservedMB"), clusterinfo.getInt("availableMB"), clusterinfo.getInt("reservedMB"), clusterinfo.getInt("availableMB"),