You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@ambari.apache.org by ja...@apache.org on 2015/05/02 03:49:50 UTC
ambari git commit: AMBARI-10893. Update widgets.json and metrics.json
files. (jaimin)
Repository: ambari
Updated Branches:
refs/heads/trunk 48edffff0 -> bb4633918
AMBARI-10893. Update widgets.json and metrics.json files. (jaimin)
Project: http://git-wip-us.apache.org/repos/asf/ambari/repo
Commit: http://git-wip-us.apache.org/repos/asf/ambari/commit/bb463391
Tree: http://git-wip-us.apache.org/repos/asf/ambari/tree/bb463391
Diff: http://git-wip-us.apache.org/repos/asf/ambari/diff/bb463391
Branch: refs/heads/trunk
Commit: bb4633918eb4ee18359b870ff681a619bd6839f3
Parents: 48edfff
Author: Jaimin Jetly <ja...@hortonworks.com>
Authored: Fri May 1 18:49:30 2015 -0700
Committer: Jaimin Jetly <ja...@hortonworks.com>
Committed: Fri May 1 18:49:40 2015 -0700
----------------------------------------------------------------------
.../HBASE/0.96.0.2.0/widgets.json | 173 ++-----
.../common-services/HDFS/2.1.0.2.0/metrics.json | 40 ++
.../common-services/HDFS/2.1.0.2.0/widgets.json | 187 ++++++-
.../common-services/YARN/2.1.0.2.0/metrics.json | 230 ++++++---
.../common-services/YARN/2.1.0.2.0/widgets.json | 513 ++++++++++++++++---
.../stacks/HDP/2.3/services/HBASE/widgets.json | 199 +++----
ambari-web/app/config.js | 2 +-
.../views/common/widget/graph_widget_view.js | 3 +-
8 files changed, 915 insertions(+), 432 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/ambari/blob/bb463391/ambari-server/src/main/resources/common-services/HBASE/0.96.0.2.0/widgets.json
----------------------------------------------------------------------
diff --git a/ambari-server/src/main/resources/common-services/HBASE/0.96.0.2.0/widgets.json b/ambari-server/src/main/resources/common-services/HBASE/0.96.0.2.0/widgets.json
index 949f273..42087b4 100644
--- a/ambari-server/src/main/resources/common-services/HBASE/0.96.0.2.0/widgets.json
+++ b/ambari-server/src/main/resources/common-services/HBASE/0.96.0.2.0/widgets.json
@@ -65,18 +65,6 @@
"is_visible": true,
"metrics": [
{
- "name": "regionserver.Server.Get_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Get_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Get_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Get_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
"name": "regionserver.Server.Get_95th_percentile._max",
"metric_path": "metrics/hbase/regionserver/Server/Get_95th_percentile._max",
"service_name": "HBASE",
@@ -85,15 +73,7 @@
],
"values": [
{
- "name": "Cluster wide minimum of 95% Read Latency",
- "value": "${regionserver.Server.Get_95th_percentile._min}"
- },
- {
- "name": "Weighted average of 95% read Latency",
- "value": "${regionserver.Server.Get_95th_percentile._avg}"
- },
- {
- "name": "Cluster wide maximum of 95% read Latency",
+ "name": "Cluster wide maximum of 95% Get Latency",
"value": "${regionserver.Server.Get_95th_percentile._max}"
}
],
@@ -110,54 +90,6 @@
"is_visible": true,
"metrics": [
{
- "name": "regionserver.Server.Mutate_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Mutate_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Increment_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Increment_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Append_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Append_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Delete_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Delete_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Mutate_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Mutate_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Increment_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Increment_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Append_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Append_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Delete_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Delete_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
"name": "regionserver.Server.Mutate_95th_percentile._max",
"metric_path": "metrics/hbase/regionserver/Server/Mutate_95th_percentile._max",
"service_name": "HBASE",
@@ -180,50 +112,24 @@
"metric_path": "metrics/hbase/regionserver/Server/Delete_95th_percentile._max",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Get_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Get_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Append_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Append_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Delete_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Delete_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Increment_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Increment_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Mutate_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Mutate_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
}
],
"values": [
{
- "name": "Cluster wide minimum of 95% write Latency",
- "value": "${regionserver.Server.Mutate_95th_percentile._min + regionserver.Server.Increment_95th_percentile._min + regionserver.Server.Append_95th_percentile._min + regionserver.Server.Delete_95th_percentile._min}"
+ "name": "Cluster wide maximum of 95% Mutate Latency",
+ "value": "${regionserver.Server.Mutate_95th_percentile._max}"
+ },
+ {
+ "name": "Cluster wide maximum of 95% Increment Latency",
+ "value": "${regionserver.Server.Increment_95th_percentile._max}"
},
{
- "name": "Weighted average of 95% write Latency",
- "value": "${((regionserver.Server.Mutate_95th_percentile._avg * regionserver.Server.Mutate_num_ops._sum) + (regionserver.Server.Increment_95th_percentile._avg * regionserver.Server.Increment_num_ops._sum) + (regionserver.Server.Append_95th_percentile._avg * regionserver.Server.Append_num_ops._sum) + (regionserver.Server.Delete_95th_percentile._avg * regionserver.Server.Delete_num_ops._sum))/(regionserver.Server.Append_num_ops._sum + regionserver.Server.Delete_num_ops._sum + regionserver.Server.Increment_num_ops._sum + regionserver.Server.Mutate_num_ops._sum)}"
+ "name": "Cluster wide maximum of 95% Append Latency",
+ "value": "${regionserver.Server.Append_95th_percentile._max}"
},
{
- "name": "Cluster wide maximum of 95% write Latency",
- "value": "${regionserver.Server.Mutate_95th_percentile._max + regionserver.Server.Increment_95th_percentile._max + regionserver.Server.Append_95th_percentile._max + regionserver.Server.Delete_95th_percentile._max}"
+ "name": "Cluster wide maximum of 95% Delete Latency",
+ "value": "${regionserver.Server.Delete_95th_percentile._max}"
}
],
"properties": {
@@ -298,8 +204,8 @@
"is_visible": true,
"metrics": [
{
- "name": "regionserver.Server.percentFilesLocal._avg",
- "metric_path": "metrics/hbase/regionserver/Server/percentFilesLocal._avg",
+ "name": "regionserver.Server.percentFilesLocal",
+ "metric_path": "metrics/hbase/regionserver/Server/percentFilesLocal",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
}
@@ -307,7 +213,7 @@
"values": [
{
"name": "Files Local",
- "value": "${regionserver.Server.percentFilesLocal._avg}"
+ "value": "${regionserver.Server.percentFilesLocal}"
}
],
"properties": {
@@ -348,21 +254,46 @@
"is_visible": true,
"metrics": [
{
- "name": "cpu_idle._avg",
- "metric_path": "metrics/cpu/cpu_idle._avg",
+ "name": "cpu_system._sum",
+ "metric_path": "metrics/cpu/cpu_system._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "cpu_user._sum",
+ "metric_path": "metrics/cpu/cpu_user._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "cpu_nice._sum",
+ "metric_path": "metrics/cpu/cpu_nice._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "cpu_idle._sum",
+ "metric_path": "metrics/cpu/cpu_idle._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "cpu_wio._sum",
+ "metric_path": "metrics/cpu/cpu_wio._sum",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
}
],
"values": [
{
- "name": "CPU Idle",
- "value": "${cpu_idle._avg}"
+ "name": "ReqionServer widget for CPU utilization",
+ "value": "${(cpu_system._sum + cpu_user._sum + cpu_nice._sum)/(cpu_system._sum + cpu_user._sum + cpu_nice._sum + cpu_idle._sum + cpu_wio._sum)}"
}
],
"properties": {
"graph_type": "LINE",
- "time_range": "1"
+ "time_range": "1",
+ "display_unit": "%"
}
},
{
@@ -401,21 +332,27 @@
},
{
"widget_name": "Cluster Disk",
- "description": "ReqionServer widget for disk utilization",
+ "description": "ReqionServer widget for Disk utilization",
"widget_type": "GRAPH",
"is_visible": true,
"metrics": [
{
- "name": "disk_free._avg",
- "metric_path": "metrics/disk/disk_free._avg",
+ "name": "disk_free._sum",
+ "metric_path": "metrics/disk/disk_free._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "disk_total._sum",
+ "metric_path": "metrics/disk/disk_total._sum",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
}
],
"values": [
{
- "name": "Disk Free",
- "value": "${disk_free._avg}"
+ "name": "Disk Utlization",
+ "value": "${(disk_total._sum - disk_free._sum)/disk_total._sum}"
}
],
"properties": {
http://git-wip-us.apache.org/repos/asf/ambari/blob/bb463391/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/metrics.json
----------------------------------------------------------------------
diff --git a/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/metrics.json b/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/metrics.json
index a96d45c..19ba890 100644
--- a/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/metrics.json
+++ b/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/metrics.json
@@ -45,6 +45,11 @@
"pointInTime": true,
"temporal": true
},
+ "metrics/dfs/namenode/TotalFileOps": {
+ "metric": "dfs.namenode.TotalFileOps",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/rpcdetailed/getProtocolVersion_num_ops": {
"metric": "rpcdetailed.rpcdetailed.getProtocolVersion_num_ops",
"pointInTime": false,
@@ -1574,6 +1579,11 @@
"pointInTime": true,
"temporal": true
},
+ "metrics/dfs/namenode/TotalFileOps": {
+ "metric": "dfs.namenode.TotalFileOps",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/rpcdetailed/getProtocolVersion_num_ops": {
"metric": "rpcdetailed.rpcdetailed.getProtocolVersion_num_ops",
"pointInTime": false,
@@ -2921,6 +2931,21 @@
"pointInTime": true,
"temporal": true
},
+ "metrics/dfs/datanode/NumFailedVolumes": {
+ "metric": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.NumFailedVolumes",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/dfs/datanode/DfsUsed": {
+ "metric": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.DfsUsed",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/dfs/datanode/Capacity": {
+ "metric": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.Capacity",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/rpc/closeRegion_num_ops": {
"metric": "rpc.rpc.closeRegion_num_ops",
"pointInTime": true,
@@ -4617,6 +4642,21 @@
"pointInTime": true,
"temporal": true
},
+ "metrics/dfs/datanode/NumFailedVolumes": {
+ "metric": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.NumFailedVolumes",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/dfs/datanode/DfsUsed": {
+ "metric": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.DfsUsed",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/dfs/datanode/Capacity": {
+ "metric": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.Capacity",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/rpc/closeRegion_num_ops": {
"metric": "rpc.rpc.closeRegion_num_ops",
"pointInTime": true,
http://git-wip-us.apache.org/repos/asf/ambari/blob/bb463391/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/widgets.json
----------------------------------------------------------------------
diff --git a/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/widgets.json b/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/widgets.json
index 5e910ae..6e52d03 100644
--- a/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/widgets.json
+++ b/ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/widgets.json
@@ -6,7 +6,7 @@
"section_name": "HDFS_SUMMARY",
"widgetLayoutInfo": [
{
- "widget_name": "GC Statistics",
+ "widget_name": "NameNode GC",
"description": "This widget shows JVM Garbage collection statistics for Active NameNode",
"widget_type": "GRAPH",
"is_visible": true,
@@ -54,7 +54,7 @@
}
},
{
- "widget_name": "Client connections",
+ "widget_name": "NN Connection Load",
"description": "Number of open connections for Active NameNode.",
"widget_type": "GRAPH",
"is_visible": true,
@@ -70,7 +70,7 @@
],
"values": [
{
- "name": "Client connections",
+ "name": "NN Connection Load",
"value": "${rpc.rpc.NumOpenConnections}"
}
],
@@ -81,7 +81,7 @@
}
},
{
- "widget_name": "NameNode JVM Heap Statistics",
+ "widget_name": "NameNode Heap",
"widget_type": "GRAPH",
"is_visible": true,
"metrics": [
@@ -116,11 +116,32 @@
}
},
{
- "widget_name": "NameNode CPU and RAM Utilization",
+ "widget_name": "NameNode Host Load",
"widget_type": "GRAPH",
"is_visible": true,
"metrics": [
{
+ "name": "cpu_system",
+ "metric_path": "metrics/cpu/cpu_system",
+ "service_name": "HDFS",
+ "component_name": "NAMENODE",
+ "host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
+ },
+ {
+ "name": "cpu_user",
+ "metric_path": "metrics/cpu/cpu_user",
+ "service_name": "HDFS",
+ "component_name": "NAMENODE",
+ "host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
+ },
+ {
+ "name": "cpu_nice",
+ "metric_path": "metrics/cpu/cpu_nice",
+ "service_name": "HDFS",
+ "component_name": "NAMENODE",
+ "host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
+ },
+ {
"name": "cpu_idle",
"metric_path": "metrics/cpu/cpu_idle",
"service_name": "HDFS",
@@ -128,8 +149,29 @@
"host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
},
{
- "name": "proc_total",
- "metric_path": "metrics/process/proc_total",
+ "name": "cpu_wio",
+ "metric_path": "metrics/cpu/cpu_wio",
+ "service_name": "HDFS",
+ "component_name": "NAMENODE",
+ "host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
+ },
+ {
+ "name": "mem_total",
+ "metric_path": "metrics/memory/mem_total",
+ "service_name": "HDFS",
+ "component_name": "NAMENODE",
+ "host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
+ },
+ {
+ "name": "mem_free",
+ "metric_path": "metrics/memory/mem_free",
+ "service_name": "HDFS",
+ "component_name": "NAMENODE",
+ "host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
+ },
+ {
+ "name": "mem_cached",
+ "metric_path": "metrics/memory/mem_cached",
"service_name": "HDFS",
"component_name": "NAMENODE",
"host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
@@ -137,50 +179,64 @@
],
"values": [
{
- "name": "CPU Idle on NameNode host",
- "value": "${cpu_idle}"
+ "name": "CPU utilization on NameNode host",
+ "value": "${(cpu_system + cpu_user + cpu_nice)/(cpu_system + cpu_user + cpu_nice + cpu_idle + cpu_wio)}"
},
{
- "name": "Total Processes on NameNode host",
- "value": "${proc_total}"
+ "name": "Memory utilization on NameNode host",
+ "value": "${(mem_total - mem_free - mem_cached)/mem_total}"
}
],
"properties": {
"graph_type": "LINE",
- "time_range": "1"
+ "time_range": "1",
+ "display_unit": "%"
}
},
{
- "widget_name": "RPC Queue Wait time",
- "widget_type": "NUMBER",
+ "widget_name": "NameNode RPC",
+ "widget_type": "GRAPH",
"is_visible": true,
"metrics": [
{
- "name": "Hadoop:service=NameNode,name=RpcActivity.RpcQueueTimeAvgTime",
+ "name": "rpc.rpc.RpcQueueTimeAvgTime",
"metric_path": "metrics/rpc/RpcQueueTime_avg_time",
"service_name": "HDFS",
"component_name": "NAMENODE",
"host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
+ },
+ {
+ "name": "rpc.rpc.RpcProcessingTimeAvgTime",
+ "metric_path": "metrics/rpc/RpcProcessingTime_avg_time",
+ "service_name": "HDFS",
+ "component_name": "NAMENODE",
+ "host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
}
],
"values": [
{
"name": "RPC Queue Wait time",
- "value": "${Hadoop:service=NameNode,name=RpcActivity.RpcQueueTimeAvgTime}"
+ "value": "${rpc.rpc.RpcQueueTimeAvgTime}"
+ },
+ {
+ "name": "RPC Processing time",
+ "value": "${rpc.rpc.RpcProcessingTimeAvgTime}"
}
],
"properties": {
+ "graph_type": "LINE",
+ "time_range": "1",
"display_unit": "ms"
}
},
{
- "widget_name": "RPC Processing time",
- "widget_type": "NUMBER",
+ "widget_name": "NameNode Operations",
+ "widget_type": "GRAPH",
"is_visible": true,
"metrics": [
{
- "name": "Hadoop:service=DataNode,name=RpcActivity.RpcProcessingTimeAvgTime",
- "metric_path": "metrics/rpc/RpcProcessingTime_avg_time",
+ "name": "dfs.namenode.TotalFileOps",
+ "metric_path": "metrics/dfs/namenode/TotalFileOps",
"service_name": "HDFS",
"component_name": "NAMENODE",
"host_component_criteria": "host_components/metrics/dfs/FSNamesystem/HAState=active"
@@ -188,12 +244,35 @@
],
"values": [
{
- "name": "RPC Processing time",
- "value": "${Hadoop:service=DataNode,name=RpcActivity.RpcProcessingTimeAvgTime}"
+ "name": "NameNode File Operations",
+ "value": "${dfs.namenode.TotalFileOps}"
}
],
"properties": {
- "display_unit": "ms"
+ "graph_type": "LINE",
+ "time_range": "1"
+ }
+ },
+ {
+ "widget_name": "Failed disk volumes",
+ "widget_type": "NUMBER",
+ "is_visible": true,
+ "metrics": [
+ {
+ "name": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.NumFailedVolumes._sum",
+ "metric_path": "metrics/dfs/datanode/NumFailedVolumes",
+ "service_name": "HDFS",
+ "component_name": "DATANODE"
+ }
+ ],
+ "values": [
+ {
+ "name": "Failed disk volumes",
+ "value": "${FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.NumFailedVolumes._sum}"
+ }
+ ],
+ "properties": {
+ "display_unit": ""
}
},
{
@@ -215,6 +294,8 @@
}
],
"properties": {
+ "warning_threshold": "0",
+ "error_threshold": "50"
}
},
{
@@ -236,6 +317,37 @@
}
],
"properties": {
+ "warning_threshold": "0",
+ "error_threshold": "50"
+ }
+ },
+ {
+ "widget_name": "HDFS Space Utilization",
+ "widget_type": "GAUGE",
+ "is_visible": true,
+ "metrics": [
+ {
+ "name": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.DfsUsed",
+ "metric_path": "metrics/dfs/datanode/DfsUsed",
+ "service_name": "HDFS",
+ "component_name": "DATANODE"
+ },
+ {
+ "name": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.Capacity",
+ "metric_path": "metrics/dfs/datanode/Capacity",
+ "service_name": "HDFS",
+ "component_name": "DATANODE"
+ }
+ ],
+ "values": [
+ {
+ "name": "HDFS Space Utilization",
+ "value": "${FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.DfsUsed/FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.Capacity}"
+ }
+ ],
+ "properties": {
+ "warning_threshold": "0.75",
+ "error_threshold": "0.9"
}
}
]
@@ -450,6 +562,35 @@
"display_unit": "%",
"max_limit": "100"
}
+ },
+ {
+ "widget_name": "HDFS Space Utilization",
+ "widget_type": "HEATMAP",
+ "is_visible": false,
+ "metrics": [
+ {
+ "name": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.DfsUsed",
+ "metric_path": "metrics/dfs/datanode/DfsUsed",
+ "service_name": "HDFS",
+ "component_name": "DATANODE"
+ },
+ {
+ "name": "FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.Capacity",
+ "metric_path": "metrics/dfs/datanode/Capacity",
+ "service_name": "HDFS",
+ "component_name": "DATANODE"
+ }
+ ],
+ "values": [
+ {
+ "name": "HDFS Space Utilization",
+ "value": "${(FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.DfsUsed/FSDatasetState.org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.Capacity) * 100}"
+ }
+ ],
+ "properties": {
+ "display_unit": "%",
+ "max_limit": "100"
+ }
}
]
}
http://git-wip-us.apache.org/repos/asf/ambari/blob/bb463391/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/metrics.json
----------------------------------------------------------------------
diff --git a/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/metrics.json b/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/metrics.json
index dc7dcbd..b85b610 100644
--- a/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/metrics.json
+++ b/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/metrics.json
@@ -187,11 +187,6 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/ContainersCompleted": {
- "metric": "yarn.NodeManagerMetrics.ContainersCompleted",
- "pointInTime": true,
- "temporal": true
- },
"metrics/rpc/SentBytes": {
"metric": "rpc.rpc.SentBytes",
"pointInTime": false,
@@ -202,11 +197,6 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/ContainersKilled": {
- "metric": "yarn.NodeManagerMetrics.ContainersKilled",
- "pointInTime": true,
- "temporal": true
- },
"metrics/jvm/logWarn": {
"metric": "jvm.JvmMetrics.LogWarn",
"pointInTime": false,
@@ -237,16 +227,91 @@
"pointInTime": false,
"temporal": true
},
+ "metrics/yarn/AllocatedContainers": {
+ "metric": "yarn.NodeManagerMetrics.AllocatedContainers",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/yarn/AllocatedGB": {
"metric": "yarn.NodeManagerMetrics.AllocatedGB",
"pointInTime": true,
"temporal": true
},
+ "metrics/yarn/AvailableGB": {
+ "metric": "yarn.NodeManagerMetrics.AvailableGB",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/yarn/AllocatedVCores": {
"metric": "yarn.NodeManagerMetrics.AllocatedVCores",
"pointInTime": true,
"temporal": true
},
+ "metrics/yarn/AvailableVCores": {
+ "metric": "yarn.NodeManagerMetrics.AvailableVCores",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/BadLocalDirs": {
+ "metric": "yarn.NodeManagerMetrics.BadLocalDirs",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/BadLogDirs": {
+ "metric": "yarn.NodeManagerMetrics.BadLogDirs",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainerLaunchDurationAvgTime": {
+ "metric": "yarn.NodeManagerMetrics.ContainerLaunchDurationAvgTime",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainerLaunchDurationNumOps": {
+ "metric": "yarn.NodeManagerMetrics.ContainerLaunchDurationNumOps",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersCompleted": {
+ "metric": "yarn.NodeManagerMetrics.ContainersCompleted",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersFailed": {
+ "metric": "yarn.NodeManagerMetrics.ContainersFailed",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersIniting": {
+ "metric": "yarn.NodeManagerMetrics.ContainersIniting",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersKilled": {
+ "metric": "yarn.NodeManagerMetrics.ContainersKilled",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersLaunched": {
+ "metric": "yarn.NodeManagerMetrics.ContainersLaunched",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersRunning": {
+ "metric": "yarn.NodeManagerMetrics.ContainersRunning",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/GoodLocalDirsDiskUtilizationPerc": {
+ "metric": "yarn.NodeManagerMetrics.GoodLocalDirsDiskUtilizationPerc",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/GoodLogDirsDiskUtilizationPerc": {
+ "metric": "yarn.NodeManagerMetrics.GoodLogDirsDiskUtilizationPerc",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/disk/part_max_used": {
"metric": "part_max_used",
"pointInTime": true,
@@ -272,11 +337,6 @@
"pointInTime": true,
"temporal": true
},
- "metrics/yarn/ContainersFailed": {
- "metric": "yarn.NodeManagerMetrics.ContainersFailed",
- "pointInTime": true,
- "temporal": true
- },
"metrics/ugi/loginSuccess_num_ops": {
"metric": "ugi.ugi.LoginSuccessNumOps",
"pointInTime": false,
@@ -287,11 +347,6 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/AllocatedContainers": {
- "metric": "yarn.NodeManagerMetrics.AllocatedContainers",
- "pointInTime": true,
- "temporal": true
- },
"metrics/jvm/threadsTerminated": {
"metric": "jvm.JvmMetrics.ThreadsTerminated",
"pointInTime": false,
@@ -317,21 +372,11 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/ContainersRunning": {
- "metric": "yarn.NodeManagerMetrics.ContainersRunning",
- "pointInTime": true,
- "temporal": true
- },
"metrics/jvm/logInfo": {
"metric": "jvm.JvmMetrics.LogInfo",
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/ContainersLaunched": {
- "metric": "yarn.NodeManagerMetrics.ContainersLaunched",
- "pointInTime": true,
- "temporal": true
- },
"metrics/rpc/RpcProcessingTime_num_ops": {
"metric": "rpc.rpc.RpcProcessingTimeNumOps",
"pointInTime": false,
@@ -342,21 +387,11 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/AvailableGB": {
- "metric": "yarn.NodeManagerMetrics.AvailableGB",
- "pointInTime": true,
- "temporal": true
- },
"metrics/mapred/ShuffleConnections": {
"metric": "mapred.ShuffleConnections",
"pointInTime": true,
"temporal": true
},
- "metrics/yarn/ContainersIniting": {
- "metric": "yarn.NodeManagerMetrics.ContainersIniting",
- "pointInTime": true,
- "temporal": true
- },
"metrics/jvm/logError": {
"metric": "jvm.JvmMetrics.LogError",
"pointInTime": false,
@@ -615,11 +650,6 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/ContainersCompleted": {
- "metric": "yarn.NodeManagerMetrics.ContainersCompleted",
- "pointInTime": true,
- "temporal": true
- },
"metrics/rpc/SentBytes": {
"metric": "rpc.rpc.SentBytes",
"pointInTime": false,
@@ -630,11 +660,6 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/ContainersKilled": {
- "metric": "yarn.NodeManagerMetrics.ContainersKilled",
- "pointInTime": true,
- "temporal": true
- },
"metrics/jvm/logWarn": {
"metric": "jvm.JvmMetrics.LogWarn",
"pointInTime": false,
@@ -665,16 +690,91 @@
"pointInTime": false,
"temporal": true
},
+ "metrics/yarn/AllocatedContainers": {
+ "metric": "yarn.NodeManagerMetrics.AllocatedContainers",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/yarn/AllocatedVCores": {
"metric": "yarn.NodeManagerMetrics.AllocatedVCores",
"pointInTime": true,
"temporal": true
},
+ "metrics/yarn/AvailableVCores": {
+ "metric": "yarn.NodeManagerMetrics.AvailableVCores",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/AvailableGB": {
+ "metric": "yarn.NodeManagerMetrics.AvailableGB",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/yarn/AllocatedGB": {
"metric": "yarn.NodeManagerMetrics.AllocatedGB",
"pointInTime": true,
"temporal": true
},
+ "metrics/yarn/BadLocalDirs": {
+ "metric": "yarn.NodeManagerMetrics.BadLocalDirs",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/BadLogDirs": {
+ "metric": "yarn.NodeManagerMetrics.BadLogDirs",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainerLaunchDurationAvgTime": {
+ "metric": "yarn.NodeManagerMetrics.ContainerLaunchDurationAvgTime",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainerLaunchDurationNumOps": {
+ "metric": "yarn.NodeManagerMetrics.ContainerLaunchDurationNumOps",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersCompleted": {
+ "metric": "yarn.NodeManagerMetrics.ContainersCompleted",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersFailed": {
+ "metric": "yarn.NodeManagerMetrics.ContainersFailed",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersIniting": {
+ "metric": "yarn.NodeManagerMetrics.ContainersIniting",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersKilled": {
+ "metric": "yarn.NodeManagerMetrics.ContainersKilled",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersLaunched": {
+ "metric": "yarn.NodeManagerMetrics.ContainersLaunched",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/ContainersRunning": {
+ "metric": "yarn.NodeManagerMetrics.ContainersRunning",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/GoodLocalDirsDiskUtilizationPerc": {
+ "metric": "yarn.NodeManagerMetrics.GoodLocalDirsDiskUtilizationPerc",
+ "pointInTime": true,
+ "temporal": true
+ },
+ "metrics/yarn/GoodLogDirsDiskUtilizationPerc": {
+ "metric": "yarn.NodeManagerMetrics.GoodLogDirsDiskUtilizationPerc",
+ "pointInTime": true,
+ "temporal": true
+ },
"metrics/disk/part_max_used": {
"metric": "part_max_used",
"pointInTime": true,
@@ -700,11 +800,6 @@
"pointInTime": true,
"temporal": true
},
- "metrics/yarn/ContainersFailed": {
- "metric": "yarn.NodeManagerMetrics.ContainersFailed",
- "pointInTime": true,
- "temporal": true
- },
"metrics/ugi/loginSuccess_num_ops": {
"metric": "ugi.ugi.LoginSuccessNumOps",
"pointInTime": false,
@@ -715,11 +810,6 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/AllocatedContainers": {
- "metric": "yarn.NodeManagerMetrics.AllocatedContainers",
- "pointInTime": true,
- "temporal": true
- },
"metrics/jvm/threadsTerminated": {
"metric": "jvm.JvmMetrics.ThreadsTerminated",
"pointInTime": false,
@@ -745,21 +835,11 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/ContainersRunning": {
- "metric": "yarn.NodeManagerMetrics.ContainersRunning",
- "pointInTime": true,
- "temporal": true
- },
"metrics/jvm/logInfo": {
"metric": "jvm.JvmMetrics.LogInfo",
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/ContainersLaunched": {
- "metric": "yarn.NodeManagerMetrics.ContainersLaunched",
- "pointInTime": true,
- "temporal": true
- },
"metrics/rpc/RpcProcessingTime_num_ops": {
"metric": "rpc.rpc.RpcProcessingTimeNumOps",
"pointInTime": false,
@@ -770,21 +850,11 @@
"pointInTime": false,
"temporal": true
},
- "metrics/yarn/AvailableGB": {
- "metric": "yarn.NodeManagerMetrics.AvailableGB",
- "pointInTime": true,
- "temporal": true
- },
"metrics/mapred/ShuffleConnections": {
"metric": "mapred.ShuffleConnections",
"pointInTime": true,
"temporal": true
},
- "metrics/yarn/ContainersIniting": {
- "metric": "yarn.NodeManagerMetrics.ContainersIniting",
- "pointInTime": true,
- "temporal": true
- },
"metrics/jvm/logError": {
"metric": "jvm.JvmMetrics.LogError",
"pointInTime": false,
http://git-wip-us.apache.org/repos/asf/ambari/blob/bb463391/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/widgets.json
----------------------------------------------------------------------
diff --git a/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/widgets.json b/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/widgets.json
index 6b62b80..a4f35b9 100644
--- a/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/widgets.json
+++ b/ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/widgets.json
@@ -6,145 +6,378 @@
"section_name": "YARN_SUMMARY",
"widgetLayoutInfo": [
{
- "widget_name": "NodeManager RAM Utilization",
- "description": "",
+ "widget_name": "Memory Utilization",
+ "description": "Memory Utilization Ratio",
"widget_type": "GRAPH",
"is_visible": true,
"metrics": [
{
- "name": "yarn.NodeManagerMetrics.AllocatedGB",
- "metric_path": "metrics/yarn/AllocatedGB",
+ "name": "yarn.QueueMetrics.Queue=root.AllocatedMB._sum",
+ "metric_path": "metrics/yarn/Queue/root/AllocatedMB._sum",
"service_name": "YARN",
- "component_name": "NODEMANAGER"
+ "component_name": "RESOURCEMANAGER"
},
{
- "name": "proc_total",
- "metric_path": "metrics/process/proc_total",
+ "name": "yarn.QueueMetrics.Queue=root.AvailableMB._sum",
+ "metric_path": "metrics/yarn/Queue/root/AvailableMB._sum",
"service_name": "YARN",
- "component_name": "NODEMANAGER"
+ "component_name": "RESOURCEMANAGER"
}
],
"values": [
{
- "name": "Total Allocatable RAM Utilized across NodeManager",
- "value": "${yarn.NodeManagerMetrics.AllocatedGB}"
+ "name": "Memory Utilization",
+ "value": "${yarn.QueueMetrics.Queue=root.AllocatedMB._sum / (yarn.QueueMetrics.Queue=root.AvailableMB._sum + yarn.QueueMetrics.Queue=root.AvailableMB._sum)}"
+ }
+ ],
+ "properties": {
+ "display_unit": "%",
+ "graph_type": "LINE",
+ "time_range": "1"
+ }
+ },
+ {
+ "widget_name": "CPU Utilization",
+ "description": "CPU Utilization Ratio",
+ "widget_type": "GRAPH",
+ "is_visible": true,
+ "metrics": [
+ {
+ "name": "yarn.QueueMetrics.Queue=root.default.AllocatedVCores._sum",
+ "metric_path": "metrics/yarn/Queue/root/AllocatedVCores._sum",
+ "service_name": "YARN",
+ "component_name": "RESOURCEMANAGER"
},
{
- "name": "NodeManager total processes",
- "value": "${proc_total}"
+ "name": "yarn.QueueMetrics.Queue=root.default.AvailableVCores._sum",
+ "metric_path": "metrics/yarn/Queue/root/AvailableVCores._sum",
+ "service_name": "YARN",
+ "component_name": "RESOURCEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "Total Allocatable CPU Utilized across NodeManager",
+ "value": "${yarn.QueueMetrics.Queue=root.default.AllocatedVCores._sum / (yarn.QueueMetrics.Queue=root.default.AllocatedVCores._sum + yarn.QueueMetrics.Queue=root.default.AvailableVCores._sum)}"
}
],
"properties": {
- "display_unit": "",
+ "display_unit": "%",
"graph_type": "LINE",
"time_range": "1"
}
},
{
- "widget_name": "NodeManager CPU Utilization",
- "description": "",
+ "widget_name": "Bad Local Disks",
+ "description": "Number of unhealthy local disks accross all NodeManagers",
+ "widget_type": "NUMBER",
+ "is_visible": true,
+ "metrics": [
+ {
+ "name": "yarn.NodeManagerMetrics.BadLocalDirs",
+ "metric_path": "metrics/yarn/BadLocalDirs",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.BadLogDirs",
+ "metric_path": "metrics/yarn/BadLogDirs",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "Number of unhealthy local disks for NodeManager",
+ "value": "${yarn.NodeManagerMetrics.BadLocalDirs + yarn.NodeManagerMetrics.BadLogDirs}"
+ }
+ ],
+ "properties": {
+ "display_unit": ""
+ }
+ },
+ {
+ "widget_name": "Container Failures",
+ "description": "Ratio of Container Failures",
"widget_type": "GRAPH",
"is_visible": true,
"metrics": [
{
- "name": "yarn.NodeManagerMetrics.AllocatedVCores",
- "metric_path": "metrics/yarn/AllocatedVCores",
+ "name": "yarn.NodeManagerMetrics.ContainersFailed._sum",
+ "metric_path": "metrics/yarn/ContainersFailed._sum",
"service_name": "YARN",
"component_name": "NODEMANAGER"
},
{
- "name": "cpu_idle",
- "metric_path": "metrics/cpu/cpu_idle",
+ "name": "yarn.NodeManagerMetrics.ContainersCompleted._sum",
+ "metric_path": "metrics/yarn/ContainersCompleted._sum",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersLaunched._sum",
+ "metric_path": "metrics/yarn/ContainersLaunched._sum",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersIniting._sum",
+ "metric_path": "metrics/yarn/ContainersIniting._sum",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersKilled._sum",
+ "metric_path": "metrics/yarn/ContainersKilled._sum",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersRunning._sum",
+ "metric_path": "metrics/yarn/ContainersRunning._sum",
"service_name": "YARN",
"component_name": "NODEMANAGER"
}
],
"values": [
{
- "name": "Total Allocatable CPU Utilized across NodeManager",
- "value": "${yarn.NodeManagerMetrics.AllocatedVCores}"
+ "name": "Container Failures",
+ "value": "${yarn.NodeManagerMetrics.ContainersFailed._sum/(yarn.NodeManagerMetrics.ContainersFailed._sum + yarn.NodeManagerMetrics.ContainersCompleted._sum + yarn.NodeManagerMetrics.ContainersLaunched._sum + yarn.NodeManagerMetrics.ContainersIniting._sum + yarn.NodeManagerMetrics.ContainersKilled._sum + yarn.NodeManagerMetrics.ContainersRunning._sum)}"
+ }
+ ],
+ "properties": {
+ "display_unit": "%",
+ "graph_type": "LINE",
+ "time_range": "1"
+ }
+ },
+ {
+ "widget_name": "App Failures",
+ "description": "App Failures ratio",
+ "widget_type": "GRAPH",
+ "is_visible": true,
+ "metrics": [
+ {
+ "name": "yarn.QueueMetrics.Queue=root.AppsFailed",
+ "metric_path": "metrics/yarn/Queue/root/AppsFailed",
+ "service_name": "YARN",
+ "component_name": "RESOURCEMANAGER"
},
{
- "name": "NodeManager CPU Idle",
- "value": "${cpu_idle}"
+ "name": "yarn.QueueMetrics.Queue=root.AppsKilled",
+ "metric_path": "metrics/yarn/Queue/root/AppsKilled",
+ "service_name": "YARN",
+ "component_name": "RESOURCEMANAGER"
+ },
+ {
+ "name": "yarn.QueueMetrics.Queue=root.AppsPending",
+ "metric_path": "metrics/yarn/Queue/root/AppsPending",
+ "service_name": "YARN",
+ "component_name": "RESOURCEMANAGER"
+ },
+ {
+ "name": "yarn.QueueMetrics.Queue=root.AppsRunning",
+ "metric_path": "metrics/yarn/Queue/root/AppsRunning",
+ "service_name": "YARN",
+ "component_name": "RESOURCEMANAGER"
+ },
+ {
+ "name": "yarn.QueueMetrics.Queue=root.AppsSubmitted",
+ "metric_path": "metrics/yarn/Queue/root/AppsSubmitted",
+ "service_name": "YARN",
+ "component_name": "RESOURCEMANAGER"
+ },
+ {
+ "name": "yarn.QueueMetrics.Queue=root.AppsCompleted",
+ "metric_path": "metrics/yarn/Queue/root/AppsCompleted",
+ "service_name": "YARN",
+ "component_name": "RESOURCEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "App Failures",
+ "value": "${yarn.QueueMetrics.Queue=root.AppsFailed/(yarn.QueueMetrics.Queue=root.AppsFailed + yarn.QueueMetrics.Queue=root.AppsKilled + yarn.QueueMetrics.Queue=root.AppsPending + yarn.QueueMetrics.Queue=root.AppsRunning + yarn.QueueMetrics.Queue=root.AppsSubmitted + yarn.QueueMetrics.Queue=root.AppsCompleted)}"
}
],
"properties": {
- "display_unit": "",
+ "display_unit": "%",
"graph_type": "LINE",
"time_range": "1"
}
},
{
- "widget_name": "NodeManager Disk and Network Utilization",
- "description": "",
+ "widget_name": "Pending Apps",
+ "description": "Applications in pending scheduling state for cluster",
"widget_type": "GRAPH",
"is_visible": true,
"metrics": [
{
- "name": "read_bytes",
- "metric_path": "metrics/disk/read_bytes",
+ "name": "yarn.QueueMetrics.Queue=root.AppsPending._sum",
+ "metric_path": "metrics/yarn/Queue/root/AppsPending._sum",
+ "service_name": "YARN",
+ "component_name": "RESOURCEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "Pending Apps",
+ "value": "${yarn.QueueMetrics.Queue=root.AppsPending._sum}"
+ }
+ ],
+ "properties": {
+ "display_unit": "Apps",
+ "graph_type": "LINE",
+ "time_range": "1"
+ }
+ },
+ {
+ "widget_name": "Cluster(NodeManager) Memory",
+ "description": "Memory utilization on NodeManager hosts",
+ "widget_type": "GRAPH",
+ "is_visible": true,
+ "metrics": [
+ {
+ "name": "mem_total._sum",
+ "metric_path": "metrics/memory/mem_total._avg",
"service_name": "YARN",
"component_name": "NODEMANAGER"
},
{
- "name": "write_bytes",
- "metric_path": "metrics/disk/write_bytes",
+ "name": "mem_free._sum",
+ "metric_path": "metrics/memory/mem_free._avg",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "mem_cached._sum",
+ "metric_path": "metrics/memory/mem_cached._avg",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "Memory utilization on NameNode host",
+ "value": "${(mem_total._sum - mem_free._sum - mem_cached._sum)/mem_total._sum}"
+ }
+ ],
+ "properties": {
+ "graph_type": "LINE",
+ "time_range": "1"
+ }
+ },
+ {
+ "widget_name": "Cluster(NodeManager) Disk",
+ "description": "NodeManager widget for disk utilization",
+ "widget_type": "GRAPH",
+ "is_visible": true,
+ "metrics": [
+ {
+ "name": "disk_free._sum",
+ "metric_path": "metrics/disk/disk_free._sum",
"service_name": "YARN",
"component_name": "NODEMANAGER"
},
{
- "name": "bytes_in",
- "metric_path": "metrics/network/bytes_in",
+ "name": "disk_total._sum",
+ "metric_path": "metrics/disk/disk_total._sum",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "Disk Utlization",
+ "value": "${(disk_total._sum - disk_free._sum)/disk_total._sum}"
+ }
+ ],
+ "properties": {
+ "graph_type": "LINE",
+ "time_range": "1"
+ }
+ },
+ {
+ "widget_name": "Cluster(NodeManager) Network",
+ "description": "NodeManager widget for network utilization",
+ "widget_type": "GRAPH",
+ "is_visible": true,
+ "metrics": [
+ {
+ "name": "pkts_in._avg",
+ "metric_path": "metrics/network/pkts_in._avg",
"service_name": "YARN",
"component_name": "NODEMANAGER"
},
{
- "name": "bytes_out",
- "metric_path": "metrics/network/bytes_out",
+ "name": "pkts_out._avg",
+ "metric_path": "metrics/network/pkts_out._avg",
"service_name": "YARN",
"component_name": "NODEMANAGER"
}
],
"values": [
{
- "name": "DISK IO",
- "value": "${read_bytes + write_bytes}"
+ "name": "Packets In",
+ "value": "${pkts_in._avg}"
},
{
- "name": "Network IO",
- "value": "${bytes_in + bytes_out}"
+ "name": "Packets Out",
+ "value": "${pkts_out._avg}"
}
],
"properties": {
- "display_unit": "",
"graph_type": "LINE",
"time_range": "1"
}
},
{
- "widget_name": "NodeManager Container Failures",
- "description": "",
+ "widget_name": "Cluster(NodeManager) CPU",
+ "description": "NodeManager widget for CPU utilization",
"widget_type": "GRAPH",
"is_visible": true,
"metrics": [
{
- "name": "yarn.NodeManagerMetrics.ContainersFailed",
- "metric_path": "metrics/yarn/ContainersFailed",
+ "name": "cpu_system._sum",
+ "metric_path": "metrics/cpu/cpu_system._sum",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "cpu_user._sum",
+ "metric_path": "metrics/cpu/cpu_user._sum",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "cpu_nice._sum",
+ "metric_path": "metrics/cpu/cpu_nice._sum",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "cpu_idle._sum",
+ "metric_path": "metrics/cpu/cpu_idle._sum",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "cpu_wio._sum",
+ "metric_path": "metrics/cpu/cpu_wio._sum",
"service_name": "YARN",
"component_name": "NODEMANAGER"
}
],
"values": [
{
- "name": "NodeManager Container Failures",
- "value": "${yarn.NodeManagerMetrics.ContainersFailed}"
+ "name": "CPU utilization on NameNode host",
+ "value": "${(cpu_system._sum + cpu_user._sum + cpu_nice._sum)/(cpu_system._sum + cpu_user._sum + cpu_nice._sum + cpu_idle._sum + cpu_wio._sum)}"
}
],
"properties": {
- "display_unit": "",
"graph_type": "LINE",
- "time_range": "1"
+ "time_range": "1",
+ "display_unit": "%"
}
}
]
@@ -155,76 +388,172 @@
"section_name": "YARN_HEATMAPS",
"widgetLayoutInfo": [
{
- "widget_name": "NodeManager GC Time",
+ "widget_name": "YARN local disk space utilization per NodeManager",
"description": "",
"widget_type": "HEATMAP",
"is_visible": true,
"metrics": [
{
- "name": "Hadoop:service=NodeManager,name=JvmMetrics.GcTimeMillis",
- "metric_path": "metrics/jvm/gcTimeMillis",
+ "name": "yarn.NodeManagerMetrics.GoodLocalDirsDiskUtilizationPerc",
+ "metric_path": "metrics/yarn/GoodLocalDirsDiskUtilizationPerc",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.GoodLogDirsDiskUtilizationPerc",
+ "metric_path": "metrics/yarn/GoodLogDirsDiskUtilizationPerc",
"service_name": "YARN",
"component_name": "NODEMANAGER"
}
],
"values": [
{
- "name": "NodeManager Garbage Collection Time",
- "value": "${Hadoop:service=NodeManager,name=JvmMetrics.GcTimeMillis}"
+ "name": "YARN local disk space utilization per NodeManager",
+ "value": "${(yarn.NodeManagerMetrics.GoodLocalDirsDiskUtilizationPerc + yarn.NodeManagerMetrics.GoodLogDirsDiskUtilizationPerc)/2}"
}
],
"properties": {
- "display_unit": "ms",
- "max_limit": "10000"
+ "display_unit": "%",
+ "max_limit": "100"
}
},
{
- "widget_name": "NodeManager JVM Heap Memory Used",
+ "widget_name": "Total Allocatable RAM Utilized per NodeManager",
"description": "",
"widget_type": "HEATMAP",
"is_visible": false,
"metrics": [
{
- "name": "Hadoop:service=NodeManager,name=JvmMetrics.MemHeapUsedM",
- "metric_path": "metrics/jvm/memHeapUsedM",
+ "name": "yarn.NodeManagerMetrics.AllocatedGB",
+ "metric_path": "metrics/yarn/AllocatedGB",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.AvailableGB",
+ "metric_path": "metrics/yarn/AvailableGB",
"service_name": "YARN",
"component_name": "NODEMANAGER"
}
],
"values": [
{
- "name": "NodeManager JVM Heap Memory Used",
- "value": "${Hadoop:service=NodeManager,name=JvmMetrics.MemHeapUsedM}"
+ "name": "Total Allocatable RAM Utilized per NodeManager",
+ "value": "${(yarn.NodeManagerMetrics.AllocatedGB/(yarn.NodeManagerMetrics.AvailableGB + yarn.NodeManagerMetrics.AllocatedGB)) * 100}"
}
],
"properties": {
- "display_unit": "MB",
- "max_limit": "512"
+ "display_unit": "%",
+ "max_limit": "100"
}
},
{
- "widget_name": "NodeManager Memory used %",
+ "widget_name": "Total Allocatable CPU Utilized per NodeManager",
"description": "",
"widget_type": "HEATMAP",
"is_visible": false,
"metrics": [
{
- "name": "yarn.NodeManagerMetrics.AllocatedGB",
- "metric_path": "metrics/yarn/AllocatedGB",
+ "name": "yarn.NodeManagerMetrics.AllocatedVCores",
+ "metric_path": "metrics/yarn/AllocatedVCores",
"service_name": "YARN",
"component_name": "NODEMANAGER"
},
{
- "name": "yarn.NodeManagerMetrics.AvailableGB",
- "metric_path": "metrics/yarn/AvailableGB",
+ "name": "yarn.NodeManagerMetrics.AvailableVCores",
+ "metric_path": "metrics/yarn/AvailableVCores",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "Total Allocatable CPU Utilized per NodeManager",
+ "value": "${(yarn.NodeManagerMetrics.AllocatedVCores/(yarn.NodeManagerMetrics.AllocatedVCores + yarn.NodeManagerMetrics.AvailableVCores)) * 100}"
+ }
+ ],
+ "properties": {
+ "display_unit": "%",
+ "max_limit": "100"
+ }
+ },
+ {
+ "widget_name": "Number of unhealthy local disks for NodeManager",
+ "description": "",
+ "widget_type": "HEATMAP",
+ "is_visible": false,
+ "metrics": [
+ {
+ "name": "yarn.NodeManagerMetrics.BadLocalDirs",
+ "metric_path": "metrics/yarn/BadLocalDirs",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.BadLogDirs",
+ "metric_path": "metrics/yarn/BadLogDirs",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "Number of unhealthy local disks for NodeManager",
+ "value": "${yarn.NodeManagerMetrics.BadLocalDirs + yarn.NodeManagerMetrics.BadLogDirs}"
+ }
+ ],
+ "properties": {
+ "display_unit": "",
+ "max_limit": "100"
+ }
+ },
+ {
+ "widget_name": "Container Failures",
+ "description": "",
+ "widget_type": "HEATMAP",
+ "is_visible": false,
+ "metrics": [
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersFailed",
+ "metric_path": "metrics/yarn/ContainersFailed",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersCompleted",
+ "metric_path": "metrics/yarn/ContainersCompleted",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersLaunched",
+ "metric_path": "metrics/yarn/ContainersLaunched",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersIniting",
+ "metric_path": "metrics/yarn/ContainersIniting",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersKilled",
+ "metric_path": "metrics/yarn/ContainersKilled",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ },
+ {
+ "name": "yarn.NodeManagerMetrics.ContainersRunning",
+ "metric_path": "metrics/yarn/ContainersRunning",
"service_name": "YARN",
"component_name": "NODEMANAGER"
}
],
"values": [
{
- "name": "NodeManager Memory used %",
- "value": "${yarn.NodeManagerMetrics.AllocatedGB/(yarn.NodeManagerMetrics.AvailableGB + yarn.NodeManagerMetrics.AllocatedGB)}"
+ "name": "Container Failures",
+ "value": "${(yarn.NodeManagerMetrics.ContainersFailed/(yarn.NodeManagerMetrics.ContainersFailed + yarn.NodeManagerMetrics.ContainersCompleted + yarn.NodeManagerMetrics.ContainersLaunched + yarn.NodeManagerMetrics.ContainersIniting + yarn.NodeManagerMetrics.ContainersKilled + yarn.NodeManagerMetrics.ContainersRunning)) * 100}"
}
],
"properties": {
@@ -233,6 +562,54 @@
}
},
{
+ "widget_name": "NodeManager GC Time",
+ "description": "",
+ "widget_type": "HEATMAP",
+ "is_visible": false,
+ "metrics": [
+ {
+ "name": "Hadoop:service=NodeManager,name=JvmMetrics.GcTimeMillis",
+ "metric_path": "metrics/jvm/gcTimeMillis",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "NodeManager Garbage Collection Time",
+ "value": "${Hadoop:service=NodeManager,name=JvmMetrics.GcTimeMillis}"
+ }
+ ],
+ "properties": {
+ "display_unit": "ms",
+ "max_limit": "10000"
+ }
+ },
+ {
+ "widget_name": "NodeManager JVM Heap Memory Used",
+ "description": "",
+ "widget_type": "HEATMAP",
+ "is_visible": false,
+ "metrics": [
+ {
+ "name": "Hadoop:service=NodeManager,name=JvmMetrics.MemHeapUsedM",
+ "metric_path": "metrics/jvm/memHeapUsedM",
+ "service_name": "YARN",
+ "component_name": "NODEMANAGER"
+ }
+ ],
+ "values": [
+ {
+ "name": "NodeManager JVM Heap Memory Used",
+ "value": "${Hadoop:service=NodeManager,name=JvmMetrics.MemHeapUsedM}"
+ }
+ ],
+ "properties": {
+ "display_unit": "MB",
+ "max_limit": "512"
+ }
+ },
+ {
"widget_name": "Allocated Containers",
"description": "",
"widget_type": "HEATMAP",
http://git-wip-us.apache.org/repos/asf/ambari/blob/bb463391/ambari-server/src/main/resources/stacks/HDP/2.3/services/HBASE/widgets.json
----------------------------------------------------------------------
diff --git a/ambari-server/src/main/resources/stacks/HDP/2.3/services/HBASE/widgets.json b/ambari-server/src/main/resources/stacks/HDP/2.3/services/HBASE/widgets.json
index 2f92e15..06f7a1a 100644
--- a/ambari-server/src/main/resources/stacks/HDP/2.3/services/HBASE/widgets.json
+++ b/ambari-server/src/main/resources/stacks/HDP/2.3/services/HBASE/widgets.json
@@ -71,66 +71,26 @@
"is_visible": true,
"metrics": [
{
- "name": "regionserver.Server.Get_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Get_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Get_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Get_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
"name": "regionserver.Server.Get_95th_percentile._max",
"metric_path": "metrics/hbase/regionserver/Server/Get_95th_percentile._max",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
},
{
- "name": "regionserver.Server.ScanNext_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/ScanNext_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.ScanNext_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/ScanNext_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
"name": "regionserver.Server.ScanNext_95th_percentile._max",
"metric_path": "metrics/hbase/regionserver/Server/ScanNext_95th_percentile._max",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Get_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Get_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.ScanNext_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/ScanNext_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
}
],
"values": [
{
- "name": "Cluster wide minimum of 95% Read Latency",
- "value": "${regionserver.Server.Get_95th_percentile._min + regionserver.Server.ScanNext_95th_percentile._min}"
- },
- {
- "name": "Weighted average of 95% Read Latency",
- "value": "${((regionserver.Server.Get_95th_percentile._avg * regionserver.Server.Get_num_ops._sum) + (regionserver.Server.ScanNext_95th_percentile._avg * regionserver.Server.ScanNext_num_ops._sum))/(regionserver.Server.Get_num_ops._sum + regionserver.Server.ScanNext_num_ops._sum)}"
+ "name": "Cluster wide maximum of 95% Get Latency",
+ "value": "${regionserver.Server.Get_95th_percentile._max}"
},
{
- "name": "Cluster wide maximum of 95% Read Latency",
- "value": "${regionserver.Server.Get_95th_percentile._max + regionserver.Server.ScanNext_95th_percentile._max}"
+ "name": "Cluster wide maximum of 95% ScanNext Latency",
+ "value": "${regionserver.Server.ScanNext_95th_percentile._max}"
}
],
"properties": {
@@ -146,54 +106,6 @@
"is_visible": true,
"metrics": [
{
- "name": "regionserver.Server.Mutate_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Mutate_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Increment_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Increment_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Append_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Append_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Delete_95th_percentile._min",
- "metric_path": "metrics/hbase/regionserver/Server/Delete_95th_percentile._min",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Mutate_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Mutate_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Increment_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Increment_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Append_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Append_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Delete_95th_percentile._avg",
- "metric_path": "metrics/hbase/regionserver/Server/Delete_95th_percentile._avg",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
"name": "regionserver.Server.Mutate_95th_percentile._max",
"metric_path": "metrics/hbase/regionserver/Server/Mutate_95th_percentile._max",
"service_name": "HBASE",
@@ -216,50 +128,24 @@
"metric_path": "metrics/hbase/regionserver/Server/Delete_95th_percentile._max",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Get_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Get_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Append_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Append_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Delete_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Delete_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Increment_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Increment_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
- },
- {
- "name": "regionserver.Server.Mutate_num_ops._sum",
- "metric_path": "metrics/hbase/regionserver/Server/Mutate_num_ops._sum",
- "service_name": "HBASE",
- "component_name": "HBASE_REGIONSERVER"
}
],
"values": [
{
- "name": "Cluster wide minimum of 95% write Latency",
- "value": "${regionserver.Server.Mutate_95th_percentile._min + regionserver.Server.Increment_95th_percentile._min + regionserver.Server.Append_95th_percentile._min + regionserver.Server.Delete_95th_percentile._min}"
+ "name": "Cluster wide maximum of 95% Mutate Latency",
+ "value": "${regionserver.Server.Mutate_95th_percentile._max}"
+ },
+ {
+ "name": "Cluster wide maximum of 95% Increment Latency",
+ "value": "${regionserver.Server.Increment_95th_percentile._max}"
},
{
- "name": "Weighted average of 95% write Latency",
- "value": "${((regionserver.Server.Mutate_95th_percentile._avg * regionserver.Server.Mutate_num_ops._sum) + (regionserver.Server.Increment_95th_percentile._avg * regionserver.Server.Increment_num_ops._sum) + (regionserver.Server.Append_95th_percentile._avg * regionserver.Server.Append_num_ops._sum) + (regionserver.Server.Delete_95th_percentile._avg * regionserver.Server.Delete_num_ops._sum))/(regionserver.Server.Append_num_ops._sum + regionserver.Server.Delete_num_ops._sum + regionserver.Server.Increment_num_ops._sum + regionserver.Server.Mutate_num_ops._sum)}"
+ "name": "Cluster wide maximum of 95% Append Latency",
+ "value": "${regionserver.Server.Append_95th_percentile._max}"
},
{
- "name": "Cluster wide maximum of 95% write Latency",
- "value": "${regionserver.Server.Mutate_95th_percentile._max + regionserver.Server.Increment_95th_percentile._max + regionserver.Server.Append_95th_percentile._max + regionserver.Server.Delete_95th_percentile._max}"
+ "name": "Cluster wide maximum of 95% Delete Latency",
+ "value": "${regionserver.Server.Delete_95th_percentile._max}"
}
],
"properties": {
@@ -334,8 +220,8 @@
"is_visible": true,
"metrics": [
{
- "name": "regionserver.Server.percentFilesLocal._avg",
- "metric_path": "metrics/hbase/regionserver/Server/percentFilesLocal._avg",
+ "name": "regionserver.Server.percentFilesLocal",
+ "metric_path": "metrics/hbase/regionserver/Server/percentFilesLocal",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
}
@@ -343,7 +229,7 @@
"values": [
{
"name": "Files Local",
- "value": "${regionserver.Server.percentFilesLocal._avg}"
+ "value": "${regionserver.Server.percentFilesLocal}"
}
],
"properties": {
@@ -384,21 +270,46 @@
"is_visible": true,
"metrics": [
{
- "name": "cpu_idle._avg",
- "metric_path": "metrics/cpu/cpu_idle._avg",
+ "name": "cpu_system._sum",
+ "metric_path": "metrics/cpu/cpu_system._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "cpu_user._sum",
+ "metric_path": "metrics/cpu/cpu_user._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "cpu_nice._sum",
+ "metric_path": "metrics/cpu/cpu_nice._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "cpu_idle._sum",
+ "metric_path": "metrics/cpu/cpu_idle._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "cpu_wio._sum",
+ "metric_path": "metrics/cpu/cpu_wio._sum",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
}
],
"values": [
{
- "name": "CPU Idle",
- "value": "${cpu_idle._avg}"
+ "name": "ReqionServer widget for CPU utilization",
+ "value": "${(cpu_system._sum + cpu_user._sum + cpu_nice._sum)/(cpu_system._sum + cpu_user._sum + cpu_nice._sum + cpu_idle._sum + cpu_wio._sum)}"
}
],
"properties": {
"graph_type": "LINE",
- "time_range": "1"
+ "time_range": "1",
+ "display_unit": "%"
}
},
{
@@ -437,21 +348,27 @@
},
{
"widget_name": "Cluster Disk",
- "description": "ReqionServer widget for disk utilization",
+ "description": "ReqionServer widget for Disk utilization",
"widget_type": "GRAPH",
"is_visible": true,
"metrics": [
{
- "name": "disk_free._avg",
- "metric_path": "metrics/disk/disk_free._avg",
+ "name": "disk_free._sum",
+ "metric_path": "metrics/disk/disk_free._sum",
+ "service_name": "HBASE",
+ "component_name": "HBASE_REGIONSERVER"
+ },
+ {
+ "name": "disk_total._sum",
+ "metric_path": "metrics/disk/disk_total._sum",
"service_name": "HBASE",
"component_name": "HBASE_REGIONSERVER"
}
],
"values": [
{
- "name": "Disk Free",
- "value": "${disk_free._avg}"
+ "name": "Disk Utlization",
+ "value": "${(disk_total._sum - disk_free._sum)/disk_total._sum}"
}
],
"properties": {
http://git-wip-us.apache.org/repos/asf/ambari/blob/bb463391/ambari-web/app/config.js
----------------------------------------------------------------------
diff --git a/ambari-web/app/config.js b/ambari-web/app/config.js
index 440a1ea..f44294a 100644
--- a/ambari-web/app/config.js
+++ b/ambari-web/app/config.js
@@ -71,7 +71,7 @@ App.supports = {
customizeAgentUserAccount: false,
installGanglia: false,
opsDuringRollingUpgrade: false,
- customizedWidgets: false,
+ customizedWidgets: true,
customizedWidgetLayout: false,
enhancedConfigs: false
};
http://git-wip-us.apache.org/repos/asf/ambari/blob/bb463391/ambari-web/app/views/common/widget/graph_widget_view.js
----------------------------------------------------------------------
diff --git a/ambari-web/app/views/common/widget/graph_widget_view.js b/ambari-web/app/views/common/widget/graph_widget_view.js
index e6d854e..9388f4c 100644
--- a/ambari-web/app/views/common/widget/graph_widget_view.js
+++ b/ambari-web/app/views/common/widget/graph_widget_view.js
@@ -120,7 +120,8 @@ App.GraphWidgetView = Em.View.extend(App.WidgetMixin, {
timestamp = dataLinks[match][i][1];
return dataLinks[match][i][0];
});
- value.push([Number(window.eval(beforeCompute)), timestamp]);
+ var dataLinkPointValue = window.isNaN(Number(window.eval(beforeCompute))) ? 0 : Number(window.eval(beforeCompute));
+ value.push([dataLinkPointValue, timestamp]);
}
}