12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244 |
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions?fields=*",
- "items" : [
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/1",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "RESOURCEMANAGER",
- "enabled" : true,
- "id" : 1,
- "ignore_host" : false,
- "interval" : 5,
- "label" : "ResourceManager RPC Latency",
- "name" : "yarn_resourcemanager_rpc_latency",
- "scope" : "ANY",
- "service_name" : "YARN",
- "source" : {
- "jmx" : {
- "property_list" : [
- "Hadoop:service=ResourceManager,name=RpcActivityForPort*/RpcQueueTimeAvgTime",
- "Hadoop:service=ResourceManager,name=RpcActivityForPort*/RpcProcessingTimeAvgTime"
- ],
- "value" : "{0}"
- },
- "reporting" : {
- "ok" : {
- "text" : "Average Queue Time:[{0}], Average Processing Time:[{1}]"
- },
- "warning" : {
- "text" : "Average Queue Time:[{0}], Average Processing Time:[{1}]",
- "value" : 3000.0
- },
- "critical" : {
- "text" : "Average Queue Time:[{0}], Average Processing Time:[{1}]",
- "value" : 5000.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{yarn-site/yarn.resourcemanager.webapp.address}}",
- "https" : "{{yarn-site/yarn.resourcemanager.webapp.https.address}}",
- "https_property" : "{{yarn-site/yarn.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/2",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "RESOURCEMANAGER",
- "enabled" : true,
- "id" : 2,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "ResourceManager Web UI",
- "name" : "yarn_resourcemanager_webui",
- "scope" : "ANY",
- "service_name" : "YARN",
- "source" : {
- "reporting" : {
- "ok" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "warning" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "critical" : {
- "text" : "Connection failed to {1}"
- }
- },
- "type" : "WEB",
- "uri" : {
- "http" : "{{yarn-site/yarn.resourcemanager.webapp.address}}",
- "https" : "{{yarn-site/yarn.resourcemanager.webapp.https.address}}",
- "https_property" : "{{yarn-site/yarn.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/3",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "APP_TIMELINE_SERVER",
- "enabled" : true,
- "id" : 3,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "App Timeline Web UI",
- "name" : "yarn_app_timeline_server_webui",
- "scope" : "ANY",
- "service_name" : "YARN",
- "source" : {
- "reporting" : {
- "ok" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "warning" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "critical" : {
- "text" : "Connection failed to {1}"
- }
- },
- "type" : "WEB",
- "uri" : {
- "http" : "{{yarn-site/yarn.timeline-service.webapp.address}}",
- "https" : "{{yarn-site/yarn.timeline-service.webapp.https.address}}",
- "https_property" : "{{yarn-site/yarn.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/4",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : null,
- "description" : null,
- "enabled" : true,
- "id" : 4,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Percent NodeManagers Available",
- "name" : "yarn_nodemanager_webui_percent",
- "scope" : "SERVICE",
- "service_name" : "YARN",
- "source" : {
- "alert_name" : "yarn_nodemanager_webui",
- "reporting" : {
- "ok" : {
- "text" : "affected: [{1}], total: [{0}]"
- },
- "warning" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.1
- },
- "critical" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.3
- }
- },
- "type" : "AGGREGATE"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/5",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "RESOURCEMANAGER",
- "enabled" : true,
- "id" : 5,
- "ignore_host" : false,
- "interval" : 5,
- "label" : "ResourceManager CPU Utilization",
- "name" : "yarn_resourcemanager_cpu",
- "scope" : "ANY",
- "service_name" : "YARN",
- "source" : {
- "jmx" : {
- "property_list" : [
- "java.lang:type=OperatingSystem/SystemCpuLoad",
- "java.lang:type=OperatingSystem/AvailableProcessors"
- ],
- "value" : "{0} * 100"
- },
- "reporting" : {
- "ok" : {
- "text" : "{1} CPU, load {0:.1%}"
- },
- "warning" : {
- "text" : "{1} CPU, load {0:.1%}",
- "value" : 200.0
- },
- "critical" : {
- "text" : "{1} CPU, load {0:.1%}",
- "value" : 250.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{yarn-site/yarn.resourcemanager.webapp.address}}",
- "https" : "{{yarn-site/yarn.resourcemanager.webapp.https.address}}",
- "https_property" : "{{yarn-site/yarn.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/6",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NODEMANAGER",
- "enabled" : true,
- "id" : 6,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "NodeManager Health",
- "name" : "yarn_nodemanager_health",
- "scope" : "HOST",
- "service_name" : "YARN",
- "source" : {
- "path" : "HDP/2.0.6/services/YARN/package/files/alert_nodemanager_health.py",
- "type" : "SCRIPT"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/7",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NODEMANAGER",
- "enabled" : true,
- "id" : 7,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "NodeManager Web UI",
- "name" : "yarn_nodemanager_webui",
- "scope" : "HOST",
- "service_name" : "YARN",
- "source" : {
- "reporting" : {
- "ok" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "warning" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "critical" : {
- "text" : "Connection failed to {1}"
- }
- },
- "type" : "WEB",
- "uri" : {
- "http" : "{{yarn-site/yarn.nodemanager.webapp.address}}",
- "https" : "{{yarn-site/yarn.nodemanager.webapp.https.address}}",
- "https_property" : "{{yarn-site/yarn.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 8042.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/8",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "ZOOKEEPER_SERVER",
- "enabled" : true,
- "id" : 8,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "ZooKeeper Server Process",
- "name" : "zookeeper_server_process",
- "scope" : "ANY",
- "service_name" : "ZOOKEEPER",
- "source" : {
- "default_port" : 2181.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "{{zookeeper-env/clientPort}}"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/9",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : null,
- "enabled" : true,
- "id" : 9,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Percent ZooKeeper Servers Available",
- "name" : "zookeeper_server_process_percent",
- "scope" : "SERVICE",
- "service_name" : "ZOOKEEPER",
- "source" : {
- "alert_name" : "zookeeper_server_process",
- "reporting" : {
- "ok" : {
- "text" : "affected: [{1}], total: [{0}]"
- },
- "warning" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.35
- },
- "critical" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.7
- }
- },
- "type" : "AGGREGATE"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/10",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "HISTORYSERVER",
- "enabled" : true,
- "id" : 10,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "History Server Web UI",
- "name" : "mapreduce_history_server_webui",
- "scope" : "ANY",
- "service_name" : "MAPREDUCE2",
- "source" : {
- "reporting" : {
- "ok" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "warning" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "critical" : {
- "text" : "Connection failed to {1}"
- }
- },
- "type" : "WEB",
- "uri" : {
- "http" : "{{mapred-site/mapreduce.jobhistory.webapp.address}}",
- "https" : "{{mapred-site/mapreduce.jobhistory.webapp.https.address}}",
- "https_property" : "{{mapred-site/mapreduce.jobhistory.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/11",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "HISTORYSERVER",
- "enabled" : true,
- "id" : 11,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "History Server Process",
- "name" : "mapreduce_history_server_process",
- "scope" : "ANY",
- "service_name" : "MAPREDUCE2",
- "source" : {
- "default_port" : 19888.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "{{mapred-site/mapreduce.jobhistory.webapp.address}}"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/12",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "HISTORYSERVER",
- "enabled" : true,
- "id" : 12,
- "ignore_host" : false,
- "interval" : 5,
- "label" : "History Server RPC Latency",
- "name" : "mapreduce_history_server_rpc_latency",
- "scope" : "ANY",
- "service_name" : "MAPREDUCE2",
- "source" : {
- "jmx" : {
- "property_list" : [
- "Hadoop:service=JobHistoryServer,name=RpcActivityForPort*/RpcQueueTimeAvgTime",
- "Hadoop:service=JobHistoryServer,name=RpcActivityForPort*/RpcProcessingTimeAvgTime"
- ],
- "value" : "{0}"
- },
- "reporting" : {
- "ok" : {
- "text" : "Average Queue Time:[{0}], Average Processing Time:[{1}]"
- },
- "warning" : {
- "text" : "Average Queue Time:[{0}], Average Processing Time:[{1}]",
- "value" : 3000.0
- },
- "critical" : {
- "text" : "Average Queue Time:[{0}], Average Processing Time:[{1}]",
- "value" : 5000.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{mapred-site/mapreduce.jobhistory.webapp.address}}",
- "https" : "{{mapred-site/mapreduce.jobhistory.webapp.https.address}}",
- "https_property" : "{{mapred-site/mapreduce.jobhistory.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/13",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "HISTORYSERVER",
- "enabled" : true,
- "id" : 13,
- "ignore_host" : false,
- "interval" : 5,
- "label" : "History Server CPU Utilization",
- "name" : "mapreduce_history_server_cpu",
- "scope" : "ANY",
- "service_name" : "MAPREDUCE2",
- "source" : {
- "jmx" : {
- "property_list" : [
- "java.lang:type=OperatingSystem/SystemCpuLoad",
- "java.lang:type=OperatingSystem/AvailableProcessors"
- ],
- "value" : "{0} * 100"
- },
- "reporting" : {
- "ok" : {
- "text" : "{1} CPU, load {0:.1%}"
- },
- "warning" : {
- "text" : "{1} CPU, load {0:.1%}",
- "value" : 200.0
- },
- "critical" : {
- "text" : "{1} CPU, load {0:.1%}",
- "value" : 250.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{mapred-site/mapreduce.jobhistory.webapp.address}}",
- "https" : "{{mapred-site/mapreduce.jobhistory.webapp.https.address}}",
- "https_property" : "{{mapred-site/mapreduce.jobhistory.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/14",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "GANGLIA_SERVER",
- "enabled" : true,
- "id" : 14,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Ganglia History Server Process Monitor",
- "name" : "ganglia_monitor_mapreduce_history_server",
- "scope" : "ANY",
- "service_name" : "GANGLIA",
- "source" : {
- "default_port" : 8666.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "8666"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/15",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "GANGLIA_SERVER",
- "enabled" : true,
- "id" : 15,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Ganglia ResourceManager Process Monitor",
- "name" : "ganglia_monitor_yarn_resourcemanager",
- "scope" : "ANY",
- "service_name" : "GANGLIA",
- "source" : {
- "default_port" : 8664.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "8664"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/16",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "GANGLIA_SERVER",
- "enabled" : true,
- "id" : 16,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Ganglia NameNode Process Monitor",
- "name" : "ganglia_monitor_hdfs_namenode",
- "scope" : "ANY",
- "service_name" : "GANGLIA",
- "source" : {
- "default_port" : 8661.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "8661"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/17",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "GANGLIA_SERVER",
- "enabled" : true,
- "id" : 17,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Ganglia HBase Master Process Monitor",
- "name" : "ganglia_monitor_hbase_master",
- "scope" : "ANY",
- "service_name" : "GANGLIA",
- "source" : {
- "default_port" : 8663.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "8663"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/18",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "GANGLIA_SERVER",
- "enabled" : true,
- "id" : 18,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Ganglia Server Process",
- "name" : "ganglia_server_process",
- "scope" : "ANY",
- "service_name" : "GANGLIA",
- "source" : {
- "default_port" : 8651.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "8651"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/19",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "SECONDARY_NAMENODE",
- "enabled" : true,
- "id" : 19,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Secondary NameNode Process",
- "name" : "secondary_namenode_process",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "default_port" : 50071.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "{{hdfs-site/dfs.namenode.secondary.http-address}}"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/20",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NAMENODE",
- "enabled" : true,
- "id" : 20,
- "ignore_host" : true,
- "interval" : 1,
- "label" : "NameNode High Availability Health",
- "name" : "namenode_ha_health",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "path" : "HDP/2.0.6/services/HDFS/package/files/alert_ha_namenode_health.py",
- "type" : "SCRIPT"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/21",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "DATANODE",
- "enabled" : true,
- "id" : 21,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "DataNode Web UI",
- "name" : "datanode_webui",
- "scope" : "HOST",
- "service_name" : "HDFS",
- "source" : {
- "reporting" : {
- "ok" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "warning" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "critical" : {
- "text" : "Connection failed to {1}"
- }
- },
- "type" : "WEB",
- "uri" : {
- "http" : "{{hdfs-site/dfs.datanode.http.address}}",
- "https" : "{{hdfs-site/dfs.datanode.https.address}}",
- "https_property" : "{{hdfs-site/dfs.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/22",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NAMENODE",
- "enabled" : true,
- "id" : 22,
- "ignore_host" : false,
- "interval" : 5,
- "label" : "NameNode Host CPU Utilization",
- "name" : "namenode_cpu",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "jmx" : {
- "property_list" : [
- "java.lang:type=OperatingSystem/SystemCpuLoad",
- "java.lang:type=OperatingSystem/AvailableProcessors"
- ],
- "value" : "{0} * 100"
- },
- "reporting" : {
- "ok" : {
- "text" : "{1} CPU, load {0:.1%}"
- },
- "warning" : {
- "text" : "{1} CPU, load {0:.1%}",
- "value" : 200.0
- },
- "critical" : {
- "text" : "{1} CPU, load {0:.1%}",
- "value" : 250.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{hdfs-site/dfs.namenode.http-address}}",
- "https" : "{{hdfs-site/dfs.namenode.https-address}}",
- "https_property" : "{{hdfs-site/dfs.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/23",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NAMENODE",
- "enabled" : true,
- "id" : 23,
- "ignore_host" : false,
- "interval" : 2,
- "label" : "NameNode RPC Latency",
- "name" : "namenode_rpc_latency",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "jmx" : {
- "property_list" : [
- "Hadoop:service=NameNode,name=RpcActivityForPort*/RpcQueueTimeAvgTime",
- "Hadoop:service=NameNode,name=RpcActivityForPort*/RpcProcessingTimeAvgTime"
- ],
- "value" : "{0}"
- },
- "reporting" : {
- "ok" : {
- "text" : "Average Queue Time:[{0}], Average Processing Time:[{1}]"
- },
- "warning" : {
- "text" : "Average Queue Time:[{0}], Average Processing Time:[{1}]",
- "value" : 3000.0
- },
- "critical" : {
- "text" : "Average Queue Time:[{0}], Average Processing Time:[{1}]",
- "value" : 5000.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{hdfs-site/dfs.namenode.http-address}}",
- "https" : "{{hdfs-site/dfs.namenode.https-address}}",
- "https_property" : "{{hdfs-site/dfs.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/24",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NAMENODE",
- "enabled" : true,
- "id" : 24,
- "ignore_host" : false,
- "interval" : 2,
- "label" : "NameNode Blocks Health",
- "name" : "namenode_hdfs_blocks_health",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "jmx" : {
- "property_list" : [
- "Hadoop:service=NameNode,name=FSNamesystem/MissingBlocks",
- "Hadoop:service=NameNode,name=FSNamesystem/BlocksTotal"
- ],
- "value" : "{0}"
- },
- "reporting" : {
- "ok" : {
- "text" : "Total Blocks:[{1}], Missing Blocks:[{0}]"
- },
- "warning" : {
- "text" : "Total Blocks:[{1}], Missing Blocks:[{0}]",
- "value" : 1.0
- },
- "critical" : {
- "text" : "Total Blocks:[{1}], Missing Blocks:[{0}]",
- "value" : 1.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{hdfs-site/dfs.namenode.http-address}}",
- "https" : "{{hdfs-site/dfs.namenode.https-address}}",
- "https_property" : "{{hdfs-site/dfs.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/25",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NAMENODE",
- "enabled" : true,
- "id" : 25,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "NameNode Web UI",
- "name" : "namenode_webui",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "reporting" : {
- "ok" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "warning" : {
- "text" : "HTTP {0} response in {2:.4f} seconds"
- },
- "critical" : {
- "text" : "Connection failed to {1}"
- }
- },
- "type" : "WEB",
- "uri" : {
- "http" : "{{hdfs-site/dfs.namenode.http-address}}",
- "https" : "{{hdfs-site/dfs.namenode.https-address}}",
- "https_property" : "{{hdfs-site/dfs.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/26",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "DATANODE",
- "enabled" : true,
- "id" : 26,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "DateNode Process",
- "name" : "datanode_process",
- "scope" : "HOST",
- "service_name" : "HDFS",
- "source" : {
- "default_port" : 50010.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "{{hdfs-site/dfs.datanode.address}}"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/27",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : null,
- "enabled" : true,
- "id" : 27,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Percent DataNodes Available",
- "name" : "datanode_process_percent",
- "scope" : "SERVICE",
- "service_name" : "HDFS",
- "source" : {
- "alert_name" : "datanode_process",
- "reporting" : {
- "ok" : {
- "text" : "affected: [{1}], total: [{0}]"
- },
- "warning" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.1
- },
- "critical" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.3
- }
- },
- "type" : "AGGREGATE"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/28",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NAMENODE",
- "enabled" : true,
- "id" : 28,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "NameNode Process",
- "name" : "namenode_process",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "default_port" : 50070.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "{{hdfs-site/dfs.namenode.http-address}}"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/29",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NAMENODE",
- "enabled" : true,
- "id" : 29,
- "ignore_host" : false,
- "interval" : 2,
- "label" : "HDFS Capacity Utilization",
- "name" : "namenode_hdfs_capacity_utilization",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "jmx" : {
- "property_list" : [
- "Hadoop:service=NameNode,name=FSNamesystemState/CapacityUsed",
- "Hadoop:service=NameNode,name=FSNamesystemState/CapacityRemaining"
- ],
- "value" : "{0}/({0} + {1}) * 100"
- },
- "reporting" : {
- "ok" : {
- "text" : "Capacity Used:[{2:d}%, {0}], Capacity Remaining:[{1}]"
- },
- "warning" : {
- "text" : "Capacity Used:[{2:d}%, {0}], Capacity Remaining:[{1}]",
- "value" : 80.0
- },
- "critical" : {
- "text" : "Capacity Used:[{2:d}%, {0}], Capacity Remaining:[{1}]",
- "value" : 90.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{hdfs-site/dfs.namenode.http-address}}",
- "https" : "{{hdfs-site/dfs.namenode.https-address}}",
- "https_property" : "{{hdfs-site/dfs.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/30",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NAMENODE",
- "enabled" : true,
- "id" : 30,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "NameNode Last Checkpoint",
- "name" : "namenode_last_checkpoint",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "path" : "HDP/2.0.6/services/HDFS/package/files/alert_checkpoint_time.py",
- "type" : "SCRIPT"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/31",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "JOURNALNODE",
- "enabled" : true,
- "id" : 31,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "JournalNode Process",
- "name" : "journalnode_process",
- "scope" : "HOST",
- "service_name" : "HDFS",
- "source" : {
- "default_port" : 8480.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} to {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "{{hdfs-site/dfs.journalnode.http-address}}"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/32",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "ZKFC",
- "enabled" : true,
- "id" : 32,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "ZooKeeper Failover Controller Process",
- "name" : "hdfs_zookeeper_failover_controller_process",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "default_port" : 2181.0,
- "reporting" : {
- "ok" : {
- "text" : "TCP OK - {0:.4f} response on port {1}"
- },
- "critical" : {
- "text" : "Connection failed: {0} on host {1}:{2}"
- }
- },
- "type" : "PORT",
- "uri" : "{{core-site/ha.zookeeper.quorum}}"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/33",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "NAMENODE",
- "enabled" : true,
- "id" : 33,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "NameNode Directory Status",
- "name" : "namenode_directory_status",
- "scope" : "ANY",
- "service_name" : "HDFS",
- "source" : {
- "jmx" : {
- "property_list" : [
- "Hadoop:service=NameNode,name=NameNodeInfo/NameDirStatuses"
- ],
- "value" : "calculate(args)\ndef calculate(args):\n import json\n json_statuses = json.loads({0})\n return len(json_statuses['failed']) if 'failed' in json_statuses else 0"
- },
- "reporting" : {
- "ok" : {
- "text" : "Directories are healthy"
- },
- "warning" : {
- "text" : "Failed directory count: {1}",
- "value" : 1.0
- },
- "critical" : {
- "text" : "Failed directory count: {1}",
- "value" : 1.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{hdfs-site/dfs.namenode.http-address}}",
- "https" : "{{hdfs-site/dfs.namenode.https-address}}",
- "https_property" : "{{hdfs-site/dfs.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/34",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : "DATANODE",
- "enabled" : true,
- "id" : 34,
- "ignore_host" : false,
- "interval" : 2,
- "label" : "DataNode Storage",
- "name" : "datanode_storage",
- "scope" : "HOST",
- "service_name" : "HDFS",
- "source" : {
- "jmx" : {
- "property_list" : [
- "Hadoop:service=DataNode,name=FSDatasetState-*/Remaining",
- "Hadoop:service=DataNode,name=FSDatasetState-*/Capacity"
- ],
- "value" : "({1} - {0})/{1} * 100"
- },
- "reporting" : {
- "ok" : {
- "text" : "Remaining Capacity:[{0}], Total Capacity:[{2:d}% Used, {1}]"
- },
- "warning" : {
- "text" : "Remaining Capacity:[{0}], Total Capacity:[{2:d}% Used, {1}]",
- "value" : 80.0
- },
- "critical" : {
- "text" : "Remaining Capacity:[{0}], Total Capacity:[{2:d}% Used, {1}]",
- "value" : 90.0
- }
- },
- "type" : "METRIC",
- "uri" : {
- "http" : "{{hdfs-site/dfs.datanode.http.address}}",
- "https" : "{{hdfs-site/dfs.datanode.https.address}}",
- "https_property" : "{{hdfs-site/dfs.http.policy}}",
- "https_property_value" : "HTTPS_ONLY",
- "default_port" : 0.0
- }
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/35",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : null,
- "enabled" : true,
- "id" : 35,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Percent DataNodes With Available Space",
- "name" : "datanode_storage_percent",
- "scope" : "SERVICE",
- "service_name" : "HDFS",
- "source" : {
- "alert_name" : "datanode_storage",
- "reporting" : {
- "ok" : {
- "text" : "affected: [{1}], total: [{0}]"
- },
- "warning" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.1
- },
- "critical" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.3
- }
- },
- "type" : "AGGREGATE"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/tdk/alert_definitions/36",
- "AlertDefinition" : {
- "cluster_name" : "tdk",
- "component_name" : null,
- "enabled" : true,
- "id" : 36,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Percent JournalNodes Available",
- "name" : "journalnode_process_percent",
- "scope" : "SERVICE",
- "service_name" : "HDFS",
- "source" : {
- "alert_name" : "journalnode_process",
- "reporting" : {
- "ok" : {
- "text" : "affected: [{1}], total: [{0}]"
- },
- "warning" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.33
- },
- "critical" : {
- "text" : "affected: [{1}], total: [{0}]",
- "value" : 0.5
- }
- },
- "type" : "AGGREGATE"
- }
- }
- },
- {
- "href" : "http://host:8080/api/v1/clusters/c1/alert_definitions/37",
- "AlertDefinition" : {
- "cluster_name" : "c1",
- "component_name" : "AMBARI_AGENT",
- "description" : null,
- "enabled" : true,
- "id" : 37,
- "ignore_host" : false,
- "interval" : 1,
- "label" : "Ambari Agent Disk Usage",
- "name" : "ambari_agent_disk_usage",
- "scope" : "HOST",
- "service_name" : "AMBARI",
- "source" : {
- "path" : "alert_disk_space.py",
- "type" : "SCRIPT"
- }
- }
- }
- ]
- }
|