Add / update definition, summary metrics and golden metrics for sever…

…al AWS entity types: EFS, Elasticache, Elasticsearch, ELB, Kinesis (#73) GM & SM updates for AWS related types
newrelic · Feb 19, 2021 · 767180b · 767180b
1 parent b9c6e43
commit 767180b
Show file tree

Hide file tree

Showing 32 changed files with 476 additions and 120 deletions.
diff --git a/definitions/infra-awsefsfilesystem/definition.yml b/definitions/infra-awsefsfilesystem/definition.yml
@@ -1,5 +1,11 @@
 domain: INFRA
 type: AWSEFSFILESYSTEM
+goldenTags:
+- aws.accountId
+- aws.awsRegion
+- aws.availabilityZone
 compositeMetrics:
+  goldenMetrics:
+  - golden_metrics.yml
   summaryMetrics:
   - summary_metrics.yml
diff --git a/definitions/infra-awsefsfilesystem/golden_metrics.yml b/definitions/infra-awsefsfilesystem/golden_metrics.yml
@@ -0,0 +1,32 @@
+iops:
+  title: Total I/O (Bytes per sec)
+  query:
+    select: rate(sum(`provider.totalIOBytes.Sum`), 1 second)
+    from: BlockDeviceSample
+    where: provider='EfsFileSystem'
+    facet: entityName
+    eventId: entityGuid
+iolimit:
+  title: I/O limit (%)
+  query:
+    select: max(`provider.percentIOLimit.Maximum`)
+    from: BlockDeviceSample
+    where: provider='EfsFileSystem'
+    facet: entityName
+    eventId: entityGuid
+clientConnections:
+  title: Client Connections
+  query:
+    select: max(`provider.clientConnections.Sum`)
+    from: BlockDeviceSample
+    where: provider='EfsFileSystem'
+    facet: entityName
+    eventId: entityGuid
+burstCreditBalance:
+  title: Burst Credit Balance
+  query:
+    select: average(`provider.burstCreditBalance.Average`)
+    from: BlockDeviceSample
+    where: provider='EfsFileSystem'
+    facet: entityName
+    eventId: entityGuid
diff --git a/definitions/infra-awsefsfilesystem/summary_metrics.yml b/definitions/infra-awsefsfilesystem/summary_metrics.yml
@@ -6,14 +6,14 @@ providerAccountName:
 iops:
   query:
     eventId: entityGuid
-    select: sum(`provider.totalIOBytes.Sum`)
+    select: rate(sum(`provider.totalIOBytes.Sum`), 1 second)
     from: BlockDeviceSample
-  unit: BYTES
+  unit: BYTES_PER_SECOND
   title: Total I/O
 iolimit:
   query:
     eventId: entityGuid
-    select: 'max(`provider.percentIOLimit.Maximum`) '
+    select: 'max(`provider.percentIOLimit.Maximum`)'
     from: BlockDeviceSample
   unit: PERCENTAGE
   title: I/O limit
diff --git a/definitions/infra-awselasticachememcachedcluster/definition.yml b/definitions/infra-awselasticachememcachedcluster/definition.yml
@@ -1,8 +1,11 @@
 domain: INFRA
 type: AWSELASTICACHEMEMCACHEDCLUSTER
 goldenTags:
-- aws.availabilityZone
 - aws.accountId
+- aws.awsRegion
+- aws.customerAvailabilityZone
+- aws.cacheNodeType
+- aws.numCacheNodes
 compositeMetrics:
   goldenMetrics:
   - golden_metrics.yml

diff --git a/definitions/infra-awselasticachememcachedcluster/golden_metrics.yml b/definitions/infra-awselasticachememcachedcluster/golden_metrics.yml
@@ -1,5 +1,5 @@
-cpuUsage:
-  title: CPU usage (%)
+cpuUtilization:
+  title: CPU utilization (%)
   query:
     select: average(provider.cpuUtilization.Average)
     from: DatastoreSample
@@ -14,3 +14,27 @@ swapUsageBytes:
     where: provider='ElastiCacheMemcachedCluster'
     facet: entityName
     eventId: entityGuid
+freeableMemory:
+  title: Free memory (bytes)
+  query:
+    select: average(`provider.freeableMemory.Average`)
+    from: DatastoreSample
+    where: provider='ElastiCacheMemcachedCluster'
+    facet: entityName
+    eventId: entityGuid
+networkBytesIn:
+  title: Bytes in per sec
+  query:
+    select: rate(sum(`provider.networkBytesIn.Sum`), 1 second)
+    from: DatastoreSample
+    where: provider='ElastiCacheMemcachedCluster'
+    facet: entityName
+    eventId: entityGuid
+networkBytesOut:
+  title: Bytes out per sec
+  query:
+    select: rate(sum(`provider.networkBytesOut.Sum`), 1 second)
+    from: DatastoreSample
+    where: provider='ElastiCacheMemcachedCluster'
+    facet: entityName
+    eventId: entityGuid
diff --git a/definitions/infra-awselasticachememcachedcluster/summary_metrics.yml b/definitions/infra-awselasticachememcachedcluster/summary_metrics.yml
@@ -13,11 +13,11 @@ cpuUtilization:
 freeableMemory:
   query:
     eventId: entityGuid
-    select: sum(`provider.freeableMemory.Sum`)
+    select: average(`provider.freeableMemory.Average`)
     from: DatastoreSample
   unit: BYTES
   title: Free memory
-memoryUtilization:
+swapUsageBytes:
   query:
     eventId: entityGuid
     select: average(`provider.swapUsage.Average`)

diff --git a/definitions/infra-awselasticachememcachednode/definition.yml b/definitions/infra-awselasticachememcachednode/definition.yml
@@ -1,8 +1,11 @@
 domain: INFRA
 type: AWSELASTICACHEMEMCACHEDNODE
 goldenTags:
-- aws.availabilityZone
 - aws.accountId
+- aws.awsRegion
+- aws.customerAvailabilityZone
+- aws.cacheClusterId
+- aws.parameterGroupStatus
 compositeMetrics:
   goldenMetrics:
   - golden_metrics.yml

diff --git a/definitions/infra-awselasticachememcachednode/golden_metrics.yml b/definitions/infra-awselasticachememcachednode/golden_metrics.yml
@@ -1,23 +1,47 @@
-cpuUsage:
-  title: CPU usage (%)
+getThroughput:
+  title: Gets per sec
   query:
-    select: average(provider.cpuUtilization.Average)
+    select: rate(sum(provider.getHits.Sum)+sum(provider.getMisses.Sum), 1 second)
     from: DatastoreSample
     where: provider='ElastiCacheMemcachedNode'
     facet: entityName
     eventId: entityGuid
-currentConnections:
-  title: Current connections
+getMisses:
+  title: Get hit rate (%)
   query:
-    select: average(`provider.currConnections.Average`)
+    select: (sum(provider.getHits.Sum)/(sum(provider.getHits.Sum)+sum(provider.getMisses.Sum)))*100
     from: DatastoreSample
     where: provider='ElastiCacheMemcachedNode'
     facet: entityName
     eventId: entityGuid
 evictedItems:
-  title: Evicted items
+  title: Evicted items per sec
+  query:
+    select: rate(sum(`provider.evictions.Sum`), 1 second)
+    from: DatastoreSample
+    where: provider='ElastiCacheMemcachedNode'
+    facet: entityName
+    eventId: entityGuid
+swapUsage:
+  title: Swap Usage
   query:
-    select: average(`provider.evictions.Average`)
+    select: average(`provider.swapUsage.Average`)
+    from: DatastoreSample
+    where: provider='ElastiCacheMemcachedNode'
+    facet: entityName
+    eventId: entityGuid
+cpuUtilization:
+  title: CPU utilization (%)
+  query:
+    select: average(provider.cpuUtilization.Average)
+    from: DatastoreSample
+    where: provider='ElastiCacheMemcachedNode'
+    facet: entityName
+    eventId: entityGuid
+currentConnections:
+  title: Current connections
+  query:
+    select: average(`provider.currConnections.Average`)
     from: DatastoreSample
     where: provider='ElastiCacheMemcachedNode'
     facet: entityName

diff --git a/definitions/infra-awselasticachememcachednode/summary_metrics.yml b/definitions/infra-awselasticachememcachednode/summary_metrics.yml
@@ -3,24 +3,24 @@ providerAccountName:
     key: providerAccountName
   title: AWS Account
   unit: STRING
-cpuUtilization:
+getThroughput:
   query:
     eventId: entityGuid
-    select: average(`provider.cpuUtilization.Average`)
+    select: rate(sum(provider.getHits.Sum)+sum(provider.getMisses.Sum), 1 second)
     from: DatastoreSample
-  unit: PERCENTAGE
-  title: CPU utilization
-freeableMemory:
+  unit: REQUESTS_PER_SECOND
+  title: Get throughput
+getHitRatio:
   query:
     eventId: entityGuid
-    select: sum(`provider.freeableMemory.Sum`)
+    select: (sum(provider.getHits.Sum)/(sum(provider.getHits.Sum)+sum(provider.getMisses.Sum)))*100
     from: DatastoreSample
-  unit: BYTES
-  title: Free memory
-memoryUtilization:
+  unit: PERCENTAGE
+  title: Get Hit Rate
+swapUsage:
   query:
     eventId: entityGuid
     select: average(`provider.swapUsage.Average`)
     from: DatastoreSample
   unit: BYTES
-  title: Swap usage
+  title: Swap Usage
diff --git a/definitions/infra-awselasticacherediscluster/definition.yml b/definitions/infra-awselasticacherediscluster/definition.yml
@@ -1,8 +1,11 @@
 domain: INFRA
 type: AWSELASTICACHEREDISCLUSTER
 goldenTags:
-- aws.availabilityZone
 - aws.accountId
+- aws.awsRegion
+- aws.customerAvailabilityZone
+- aws.cacheNodeType
+- aws.numCacheNodes
 compositeMetrics:
   goldenMetrics:
   - golden_metrics.yml

diff --git a/definitions/infra-awselasticacherediscluster/golden_metrics.yml b/definitions/infra-awselasticacherediscluster/golden_metrics.yml
@@ -1,11 +1,19 @@
-cpuUsage:
-  title: CPU usage (%)
+cpuUtilization:
+  title: CPU utilizacion (%)
   query:
     select: average(provider.cpuUtilization.Average)
     from: DatastoreSample
     where: provider='ElastiCacheRedisCluster'
     facet: entityName
     eventId: entityGuid
+engineCpuUtilization:
+  title: Egine CPU utilization (%)
+  query:
+    select: average(provider.engineCpuUtilization.Average)
+    from: DatastoreSample
+    where: provider='ElastiCacheRedisCluster'
+    facet: entityName
+    eventId: entityGuid
 swapUsageBytes:
   title: Swap usage (bytes)
   query:
@@ -14,3 +22,27 @@ swapUsageBytes:
     where: provider='ElastiCacheRedisCluster'
     facet: entityName
     eventId: entityGuid
+freeableMemory:
+  title: Free memory (bytes)
+  query:
+    select: average(`provider.freeableMemory.Average`)
+    from: DatastoreSample
+    where: provider='ElastiCacheRedisCluster'
+    facet: entityName
+    eventId: entityGuid
+networkBytesIn:
+  title: Bytes in per sec
+  query:
+    select: rate(sum(`provider.networkBytesIn.Sum`), 1 second)
+    from: DatastoreSample
+    where: provider='ElastiCacheRedisCluster'
+    facet: entityName
+    eventId: entityGuid
+networkBytesOut:
+  title: Bytes out per sec
+  query:
+    select: rate(sum(`provider.networkBytesOut.Sum`), 1 second)
+    from: DatastoreSample
+    where: provider='ElastiCacheRedisCluster'
+    facet: entityName
+    eventId: entityGuid
diff --git a/definitions/infra-awselasticacherediscluster/summary_metrics.yml b/definitions/infra-awselasticacherediscluster/summary_metrics.yml
@@ -13,7 +13,7 @@ cpuUtilization:
 freeableMemory:
   query:
     eventId: entityGuid
-    select: sum(`provider.freeableMemory.Sum`)
+    select: average(`provider.freeableMemory.Average`)
     from: DatastoreSample
   unit: BYTES
   title: Free memory

diff --git a/definitions/infra-awselasticacheredisnode/definition.yml b/definitions/infra-awselasticacheredisnode/definition.yml
@@ -1,8 +1,11 @@
 domain: INFRA
 type: AWSELASTICACHEREDISNODE
 goldenTags:
-- aws.availabilityZone
 - aws.accountId
+- aws.awsRegion
+- aws.customerAvailabilityZone
+- aws.cacheClusterId
+- aws.parameterGroupStatus
 compositeMetrics:
   goldenMetrics:
   - golden_metrics.yml

diff --git a/definitions/infra-awselasticacheredisnode/golden_metrics.yml b/definitions/infra-awselasticacheredisnode/golden_metrics.yml
@@ -1,23 +1,47 @@
-cpuUsage:
-  title: CPU usage (%)
+readThroughput:
+  title: Reads per sec
   query:
-    select: average(provider.cpuUtilization.Average)
+    select: rate(sum(provider.cacheHits.Sum)+sum(provider.cacheMisses.Sum), 1 second)
     from: DatastoreSample
     where: provider='ElastiCacheRedisNode'
     facet: entityName
     eventId: entityGuid
-currentConnections:
-  title: Current connections
+cacheHitRate:
+  title: Cache hit rate (%)
   query:
-    select: average(`provider.currConnections.Average`)
+    select: (sum(provider.cacheHits.Sum)/(sum(provider.cacheHits.Sum)+sum(provider.cacheMisses.Sum)))*100
     from: DatastoreSample
     where: provider='ElastiCacheRedisNode'
     facet: entityName
     eventId: entityGuid
 evictedItems:
-  title: Evicted items
+  title: Evicted items per sec
+  query:
+    select: rate(sum(`provider.evictions.Sum`), 1 second)
+    from: DatastoreSample
+    where: provider='ElastiCacheRedisNode'
+    facet: entityName
+    eventId: entityGuid
+swapUsage:
+  title: Swap Usage
   query:
-    select: average(`provider.evictions.Average`)
+    select: average(`provider.swapUsage.Average`)
+    from: DatastoreSample
+    where: provider='ElastiCacheRedisNode'
+    facet: entityName
+    eventId: entityGuid
+cpuUtilization:
+  title: CPU utilization (%)
+  query:
+    select: average(provider.cpuUtilization.Average)
+    from: DatastoreSample
+    where: provider='ElastiCacheRedisNode'
+    facet: entityName
+    eventId: entityGuid
+currentConnections:
+  title: Current connections
+  query:
+    select: average(`provider.currConnections.Average`)
     from: DatastoreSample
     where: provider='ElastiCacheRedisNode'
     facet: entityName