Monitoring OpenSearch cluster metrics with Amazon CloudWatch
Amazon OpenSearch Service publishes data from your domains to Amazon CloudWatch. CloudWatch lets you retrieve statistics
about those data points as an ordered set of timeseries data, known as metrics. OpenSearch Service sends most metrics to CloudWatch in 60second
intervals. If you use General Purpose or Magnetic EBS volumes, the EBS volume metrics update
only every five minutes. All cumulative metrics (e.g. ThreadpoolWriteRejeceted
,
ThreadpoolSearchRejected
) are inmemory and will lose state. Metrics will
reset during a node drop, node bounce, node replacement, and blue/green deployment. For more
information about Amazon CloudWatch, see the Amazon CloudWatch User
Guide.
The OpenSearch Service console displays a series of charts based on the raw data from CloudWatch. Depending on
your needs, you might prefer to view cluster data in CloudWatch instead of the graphs in the
console. The service archives metrics for two weeks before discarding them. The metrics are
provided at no extra charge, but CloudWatch still charges for creating dashboards and alarms. For
more information, see Amazon CloudWatch
pricing
OpenSearch Service publishes the following metrics to CloudWatch:
Viewing metrics in CloudWatch
CloudWatch metrics are grouped first by the service namespace, and then by the various dimension combinations within each namespace.
To view metrics using the CloudWatch console

Open the CloudWatch console at https://console.aws.amazon.com/cloudwatch/
. 
In the left navigation pane, find Metrics and choose All metrics. Select the ES/OpenSearchService namespace.

Choose a dimension to view the corresponding metrics. Metrics for individual nodes are in the
ClientId, DomainName, NodeId
dimension. Cluster metrics are in thePerDomain, PerClient Metrics
dimension. Some node metrics are aggregated at the cluster level and thus included in both dimensions. Shard metrics are in theClientId, DomainName, NodeId, ShardRole
dimension.
To view a list of metrics using the AWS CLI
Run the following command:
aws cloudwatch listmetrics namespace "AWS/ES"
Interpreting health charts in OpenSearch Service
To view metrics in OpenSearch Service, use the Cluster health and Instance health tabs. The Instance health tab uses box charts to provide ataglance visibility into the health of each OpenSearch node:

Each colored box shows the range of values for the node over the specified time period.

Blue boxes represent values that are consistent with other nodes. Red boxes represent outliers.

The white line within each box shows the node's current value.

The “whiskers” on either side of each box show the minimum and maximum values for all nodes over the time period.
If you make configuration changes to your domain, the list of individual instances in the Cluster health and Instance health tabs often double in size for a brief period before returning to the correct number. For an explanation of this behavior, see Making configuration changes in Amazon OpenSearch Service.
Cluster metrics
Amazon OpenSearch Service provides the following metrics for clusters.
Metric  Description 

ClusterStatus.green 
A value of 1 indicates that all index shards are allocated to nodes in the cluster. Relevant statistics: Maximum 
ClusterStatus.yellow 
A value of 1 indicates that the primary shards for all indexes are
allocated to nodes in the cluster, but replica shards for at least one
index are not. For more information, see Yellow cluster
status. Relevant statistics: Maximum 
ClusterStatus.red 
A value of 1 indicates that the primary and replica shards for at least one index are not allocated to nodes in the cluster. For more information, see Red cluster status. Relevant statistics: Maximum 
Shards.active 
The total number of active primary and replica shards. Relevant statistics: Maximum, Sum 
Shards.unassigned 
The number of shards that are not allocated to nodes in the cluster. Relevant statistics: Maximum, Sum 
Shards.delayedUnassigned 
The number of shards whose node allocation has been delayed by the timeout settings. Relevant statistics: Maximum, Sum 
Shards.activePrimary 
The number of active primary shards. Relevant statistics: Maximum, Sum 
Shards.initializing 
The number of shards that are under initialization. Relevant statistics: Sum 
Shards.relocating 
The number of shards that are under relocation. Relevant statistics: Sum 
Nodes 
The number of nodes in the OpenSearch Service cluster, including dedicated master nodes and UltraWarm nodes. For more information, see Making configuration changes in Amazon OpenSearch Service. Relevant statistics: Maximum 
SearchableDocuments 
The total number of searchable documents across all data nodes in the cluster. Relevant statistics: Minimum, Maximum, Average 
DeletedDocuments 
The total number of documents marked for deletion across all data nodes in the cluster. These documents no longer appear in search results, but OpenSearch only removes deleted documents from disk during segment merges. This metric increases after delete requests and decreases after segment merges. Relevant statistics: Minimum, Maximum, Average 
CPUUtilization 
The percentage of CPU usage for data nodes in the cluster. Maximum shows the node with the highest CPU usage. Average represents all nodes in the cluster. This metric is also available for individual nodes. Relevant statistics: Maximum, Average 
FreeStorageSpace 
The free space for data nodes in the cluster. The OpenSearch Service console displays this value in GiB. The Amazon CloudWatch console displays it in MiB. Note
Relevant statistics: Minimum, Maximum, Average, Sum 
ClusterUsedSpace 
The total used space for the cluster. You must leave the period at one minute to get an accurate value. The OpenSearch Service console displays this value in GiB. The Amazon CloudWatch console displays it in MiB. Relevant statistics: Minimum, Maximum 
ClusterIndexWritesBlocked 
Indicates whether your cluster is accepting or blocking incoming write requests. A value of 0 means that the cluster is accepting requests. A value of 1 means that it is blocking requests. Some common factors include the following:
Relevant statistics: Maximum 
JVMMemoryPressure 
The maximum percentage of the Java heap used for all data nodes in the cluster. OpenSearch Service uses half of an instance's RAM for the Java heap, up to a heap size of 32 GiB. You can scale instances vertically up to 64 GiB of RAM, at which point you can scale horizontally by adding instances. See Recommended CloudWatch alarms for Amazon OpenSearch Service. Relevant statistics: Maximum NoteThe logic for this metric changed in service software R20220323. For more information, see the release notes. 
OldGenJVMMemoryPressure 
The maximum percentage of the Java heap used for the "old generation" on all data nodes in the cluster. This metric is also available at the node level. Relevant statistics: Maximum 
AutomatedSnapshotFailure 
The number of failed automated snapshots for the cluster. A value
of Relevant statistics: Minimum, Maximum 
CPUCreditBalance 
The remaining CPU credits available for data nodes in the cluster. A CPU credit provides the performance of a full CPU core for one minute. For more information, see CPU credits in the Amazon EC2 Developer Guide. This metric is available only for the T2 instance types. Relevant statistics: Minimum 
OpenSearchDashboardsHealthyNodes 
A health check for OpenSearch Dashboards. If the minimum, maximum, and average are all equal to 1, Dashboards is behaving normally. If you have 10 nodes with a maximum of 1, minimum of 0, and average of 0.7, this means 7 nodes (70%) are healthy and 3 nodes (30%) are unhealthy. Relevant statistics: Minimum, Maximum, Average 
OpensearchDashboardsReportingFailedRequestSysErrCount 
The number of requests to generate OpenSearch Dashboards reports that failed due to server problems or feature limitations. Relevant statistics: Sum 
OpensearchDashboardsReportingFailedRequestUserErrCount 
The number of requests to generate OpenSearch Dashboards reports that failed due to client issues. Relevant statistics: Sum 
OpensearchDashboardsReportingRequestCount 
The total number of requests to generate OpenSearch Dashboards reports. Relevant statistics: Sum 
OpensearchDashboardsReportingSuccessCount 
The number of successful requests to generate OpenSearch Dashboards reports. Relevant statistics: Sum 
KMSKeyError 
A value of 1 indicates that the AWS KMS key used to encrypt data at rest has been disabled. To restore the domain to normal operations, reenable the key. The console displays this metric only for domains that encrypt data at rest. Relevant statistics: Minimum, Maximum 
KMSKeyInaccessible 
A value of 1 indicates that the AWS KMS key used to encrypt data at rest has been deleted or revoked its grants to OpenSearch Service. You can't recover domains that are in this state. If you have a manual snapshot, though, you can use it to migrate the domain's data to a new domain. The console displays this metric only for domains that encrypt data at rest. Relevant statistics: Minimum, Maximum 
InvalidHostHeaderRequests 
The number of HTTP requests made to the OpenSearch cluster that included an invalid (or missing) host header. Valid requests include the domain hostname as the host header value. OpenSearch Service rejects invalid requests for public access domains that don't have a restrictive access policy. We recommend applying a restrictive access policy to all domains. If you see large values for this metric, confirm that your OpenSearch clients include the domain hostname (and not, for example, its IP address) in their requests. Relevant statistics: Sum 
OpenSearchRequests (previously
ElasticsearchRequests) 
The number of requests made to the OpenSearch cluster. Relevant statistics: Sum 
2xx, 3xx, 4xx, 5xx 
The number of requests to the domain that resulted in the given HTTP response code (2xx, 3xx, 4xx, 5xx). Relevant statistics: Sum 
ThroughputThrottle 
Indicates whether or not disks have been throttled. Throttling
occurs when the combined throughput of
For information on instance throughput, see Amazon EBS–optimized
instances. For information on volume throughput, see
Amazon EBS
volume types Relevant statistics: Minimum, Maximum 
IopsThrottle 
Indicates whether or not the number of input/output operations per second (IOPS) on the domain have been throttled. Throttling occurs when IOPS of the data node breach the maximum allowed limit of the EBS volume or the EC2 instance of the data node. For information on instance IOPS, see Amazon EBS–optimized
instances. For information on volume IOPS, see Amazon EBS volume
types Relevant statistics: Minimum, Maximum 
Dedicated master node metrics
Amazon OpenSearch Service provides the following metrics for dedicated master nodes.
Metric  Description 

MasterCPUUtilization 
The maximum percentage of CPU resources used by the dedicated master nodes. We recommend increasing the size of the instance type when this metric reaches 60 percent. Relevant statistics: Maximum 
MasterFreeStorageSpace 
This metric is not relevant and can be ignored. The service does not use master nodes as data nodes. 
MasterJVMMemoryPressure 
The maximum percentage of the Java heap used for all dedicated master nodes in the cluster. We recommend moving to a larger instance type when this metric reaches 85 percent. Relevant statistics: Maximum NoteThe logic for this metric changed in service software R20220323. For more information, see the release notes. 
MasterOldGenJVMMemoryPressure 
The maximum percentage of the Java heap used for the "old generation" per master node. Relevant statistics: Maximum 
MasterCPUCreditBalance 
The remaining CPU credits available for dedicated master nodes in the cluster. A CPU credit provides the performance of a full CPU core for one minute. For more information, see CPU credits in the Amazon EC2 Developer Guide. This metric is available only for the T2 instance types. Relevant statistics: Minimum 
MasterReachableFromNode 
A health check for Failures mean that the master node is unreachable from the source node. They're usually the result of a network connectivity issue or an AWS dependency problem. Relevant statistics: Maximum 
MasterSysMemoryUtilization 
The percentage of the master node's memory that is in use. Relevant statistics: Maximum 
EBS volume metrics
Amazon OpenSearch Service provides the following metrics for EBS volumes.
Metric  Description 

ReadLatency 
The latency, in seconds, for read operations on EBS volumes. This metric is also available for individual nodes. Relevant statistics: Minimum, Maximum, Average 
WriteLatency 
The latency, in seconds, for write operations on EBS volumes. This metric is also available for individual nodes. Relevant statistics: Minimum, Maximum, Average 
ReadThroughput 
The throughput, in bytes per second, for read operations on EBS volumes. This metric is also available for individual nodes. Relevant statistics: Minimum, Maximum, Average 
ReadThroughputMicroBursting 
The throughput, in bytes per second, for read operations on EBS
volumes when microbursting Relevant statistics: Minimum, Maximum, Average 
WriteThroughput 
The throughput, in bytes per second, for write operations on EBS volumes. This metric is also available for individual nodes. Relevant statistics: Minimum, Maximum, Average 
WriteThroughputMicroBursting 
The throughput, in bytes per second, for write operations on EBS
volumes when microbursting Relevant statistics: Minimum, Maximum, Average 
DiskQueueDepth 
The number of pending input and output (I/O) requests for an EBS volume. Relevant statistics: Minimum, Maximum, Average 
ReadIOPS 
The number of input and output (I/O) operations per second for read operations on EBS volumes. This metric is also available for individual nodes. Relevant statistics: Minimum, Maximum, Average 
ReadIOPSMicroBursting 
The number of input and output (I/O) operations per second for
read operations on EBS volumes when microbursting Relevant statistics: Minimum, Maximum, Average 
WriteIOPS 
The number of input and output (I/O) operations per second for write operations on EBS volumes. This metric is also available for individual nodes. Relevant statistics: Minimum, Maximum, Average 
WriteIOPSMicroBursting 
The number of input and output (I/O) operations per second for
write operations on EBS volumes when microbursting Relevant statistics: Minimum, Maximum, Average 
BurstBalance 
The percentage of input and output (I/O) credits remaining in the burst bucket for an EBS volume. A value of 100 means that the volume has accumulated the maximum number of credits. If this percentage falls below 70%, see Low EBS burst balance. The burst balance stays at 0 for domains with gp3 volumes types, and domains with gp2 volumes that have a volume size above 1000 GiB. Relevant statistics: Minimum, Maximum, Average 
Instance metrics
Amazon OpenSearch Service provides the following metrics for each instance in a domain. OpenSearch Service also aggregates these instance metrics to provide insight into overall cluster health. You can verify this behavior using the Sample Count statistic in the console. Note that each metric in the following table has relevant statistics for the node and the cluster.
Important
Different versions of Elasticsearch use different thread pools to process calls to
the _index
API. Elasticsearch 1.5 and 2.3 use the index thread pool.
Elasticsearch 5.x, 6.0, and 6.2 use the bulk
thread pool. OpenSearch and Elasticsearch 6.3 and later use the write thread pool.
Currently, the OpenSearch Service console doesn't include a graph for the bulk thread
pool.
Use GET _cluster/settings?include_defaults=true
to check thread pool
and queue sizes for your cluster.
Metric  Description 

ConcurrentSearchRate 
The total number of search requests using concurrent segment
search per minute for all shards on a data node. A single call to
the Relevant node statistics: Average Relevant cluster statistics: Average, Maximum, Sum 
ConcurrentSearchLatency 
The difference in total time, in milliseconds, taken by all searches using concurrent segment search in a node between minute N and minute (N1). Relevant node statistics: Average Relevant cluster statistics: Average, Maximum 
IndexingLatency 
The difference in total time, in milliseconds, taken by all indexing operations in a node between minute N and minute (N1). Relevant node statistics: Average Relevant cluster statistics: Average, Maximum 
IndexingRate 
The number of indexing operations per minute. A single call to the
Relevant node statistics: Average Relevant cluster statistics: Average, Maximum, Sum 
SearchLatency 
The difference in total time, in milliseconds, taken by all searches in a node between minute N and minute (N1). Relevant node statistics: Average Relevant cluster statistics: Average, Maximum 
SearchRate 
The total number of search requests per minute for all shards on a
data node. A single call to the Relevant node statistics: Average Relevant cluster statistics: Average, Maximum, Sum 
SegmentCount 
The number of segments on a data node. The more segments you have, the longer each search takes. OpenSearch occasionally merges smaller segments into a larger one. Relevant node statistics: Maximum, Average Relevant cluster statistics: Sum, Maximum, Average 
SysMemoryUtilization 
The percentage of the instance's memory that is in use. High
values for this metric are normal and usually do not represent a
problem with your cluster. For a better indicator of potential
performance and stability issues, see the
Relevant node statistics: Minimum, Maximum, Average Relevant cluster statistics: Minimum, Maximum, Average 
JVMGCYoungCollectionCount 
The number of times that "young generation" garbage collection has run. A large, evergrowing number of runs is a normal part of cluster operations. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
JVMGCYoungCollectionTime 
The amount of time, in milliseconds, that the cluster has spent performing "young generation" garbage collection. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
JVMGCOldCollectionCount 
The number of times that "old generation" garbage collection has run. In a cluster with sufficient resources, this number should remain small and grow infrequently. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
JVMGCOldCollectionTime 
The amount of time, in milliseconds, that the cluster has spent performing "old generation" garbage collection. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
OpenSearchDashboardsConcurrentConnections 
The number of active concurrent connections to OpenSearch Dashboards. If this number is consistently high, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
OpenSearchDashboardsHealthyNode 
A health check for the individual OpenSearch Dashboards node. A value of 1 indicates normal behavior. A value of 0 indicates that Dashboards is inaccessible. Relevant node statistics: Minimum Relevant cluster statistics: Minimum, Maximum, Average 
OpenSearchDashboardsHeapTotal 
The amount of heap memory allocated to OpenSearch Dashboards in MiB. Different EC2 instance types can impact the exact memory allocation. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
OpenSearchDashboardsHeapUsed 
The absolute amount of heap memory used by OpenSearch Dashboards in MiB. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
OpenSearchDashboardsHeapUtilization 
The maximum percentage of available heap memory used by OpenSearch Dashboards. If this value increases above 80%, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Minimum, Maximum, Average 
OpenSearchDashboardsOS1MinuteLoad 
The oneminute CPU load average for OpenSearch Dashboards. The CPU load should ideally stay below 1.00. While temporary spikes are fine, we recommend increasing the size of the instance type if this metric is consistently above 1.00. Relevant node statistics: Average Relevant cluster statistics: Average, Maximum 
OpenSearchDashboardsRequestTotal 
The total count of HTTP requests made to OpenSearch Dashboards. If your system is slow or you see high numbers of Dashboards requests, consider increasing the size of the instance type. Relevant node statistics: Sum Relevant cluster statistics: Sum 
OpenSearchDashboardsResponseTimesMaxInMillis 
The maximum amount of time, in milliseconds, that it takes for OpenSearch Dashboards to respond to a request. If requests consistently take a long time to return results, consider increasing the size of the instance type. Relevant node statistics: Maximum Relevant cluster statistics: Maximum, Average 
SearchTaskCancelled 
The number of coordinator node cancellations. Relevant node statistics: Sum Relevant cluster statistics: Sum 
SearchShardTaskCancelled 
The number of data node cancellations. Relevant node statistics: Sum Relevant cluster statistics: Sum, 
ThreadpoolForce_mergeQueue 
The number of queued tasks in the force merge thread pool. If the queue size is consistently high, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
ThreadpoolForce_mergeRejected 
The number of rejected tasks in the force merge thread pool. If this number continually grows, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Sum 
ThreadpoolForce_mergeThreads 
The size of the force merge thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum 
ThreadpoolIndexQueue 
The number of queued tasks in the index thread pool. If the queue size is consistently high, consider scaling your cluster. The maximum index queue size is 200. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
ThreadpoolIndexRejected 
The number of rejected tasks in the index thread pool. If this number continually grows, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Sum 
ThreadpoolIndexThreads 
The size of the index thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum 
ThreadpoolSearchQueue 
The number of queued tasks in the search thread pool. If the queue size is consistently high, consider scaling your cluster. The maximum search queue size is 1,000. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
ThreadpoolSearchRejected 
The number of rejected tasks in the search thread pool. If this number continually grows, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Sum 
ThreadpoolSearchThreads 
The size of the search thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum 
ThreadpoolsqlworkerQueue 
The number of queued tasks in the SQL search thread pool. If the queue size is consistently high, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
ThreadpoolsqlworkerRejected 
The number of rejected tasks in the SQL search thread pool. If this number continually grows, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Sum 
ThreadpoolsqlworkerThreads 
The size of the SQL search thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum 
ThreadpoolBulkQueue 
The number of queued tasks in the bulk thread pool. If the queue size is consistently high, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
ThreadpoolBulkRejected 
The number of rejected tasks in the bulk thread pool. If this number continually grows, consider scaling your cluster. Relevant node statistics: Maximum Relevant cluster statistics: Sum 
ThreadpoolBulkThreads 
The size of the bulk thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum 
ThreadpoolIndexSearcherQueue 
The number of queued tasks in the index searcher thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
ThreadpoolIndexSearcherRejected 
The number of rejected tasks in the index searcher thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Sum 
ThreadpoolIndexSearcherThreads 
The size of the index searcher thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum 
ThreadpoolWriteThreads 
The size of the write thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum 
ThreadpoolWriteQueue 
The number of queued tasks in the write thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum 
ThreadpoolWriteRejected 
The number of rejected tasks in the write thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum NoteBecause the default write queue size was increased from 200 to
10000 in version 7.1, this metric is no longer the only
indicator of rejections from OpenSearch Service. Use the

CoordinatingWriteRejected 
The total number of rejections happened on the coordinating node due to indexing pressure since the last OpenSearch Service process startup. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum This metric is available in version 7.1 and above. 
PrimaryWriteRejected 
The total number of rejections happened on the primary shards due to indexing pressure since the last OpenSearch Service process startup. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum This metric is available in version 7.1 and above. 
ReplicaWriteRejected 
The total number of rejections happened on the replica shards due to indexing pressure since the last OpenSearch Service process startup. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum This metric is available in version 7.1 and above. 
UltraWarm metrics
Amazon OpenSearch Service provides the following metrics for UltraWarm nodes.
Metric  Description 

WarmCPUUtilization 
The percentage of CPU usage for UltraWarm nodes in the cluster. Maximum shows the node with the highest CPU usage. Average represents all UltraWarm nodes in the cluster. This metric is also available for individual UltraWarm nodes. Relevant statistics: Maximum, Average 
WarmFreeStorageSpace 
The amount of free warm storage space in MiB. Because UltraWarm
uses Amazon S3 rather than attached disks, Relevant statistics: Sum 
WarmSearchableDocuments 
The total number of searchable documents across all warm indexes in the cluster. You must leave the period at one minute to get an accurate value. Relevant statistics: Sum 
WarmSearchLatency

The difference in total time, in milliseconds, taken by all searches in an UltraWarm between minute N and minute (N1). Relevant node statistics: Average Relevant cluster statistics: Average, Maximum 
WarmSearchRate

The total number of search requests per minute for all shards on
an UltraWarm node. A single call to the Relevant node statistics: Average Relevant cluster statistics: Average, Maximum, Sum 
WarmStorageSpaceUtilization 
The total amount of warm storage space, in MiB, that the cluster is using. Relevant statistics: Maximum 
HotStorageSpaceUtilization

The total amount of hot storage space that the cluster is using. Relevant statistics: Maximum 
WarmSysMemoryUtilization 
The percentage of the warm node's memory that is in use. Relevant statistics: Maximum 
HotToWarmMigrationQueueSize

The number of indexes currently waiting to migrate from hot to warm storage. Relevant statistics: Maximum 
WarmToHotMigrationQueueSize

The number of indexes currently waiting to migrate from warm to hot storage. Relevant statistics: Maximum 
HotToWarmMigrationFailureCount

The total number of failed hot to warm migrations. Relevant statistics: Sum 
HotToWarmMigrationForceMergeLatency

The average latency of the force merge stage of the migration
process. If this stage consistently takes too long, consider
increasing
Relevant statistics: Average 
HotToWarmMigrationSnapshotLatency

The average latency of the snapshot stage of the migration process. If this stage consistently takes too long, ensure that your shards are appropriately sized and distributed throughout the cluster. Relevant statistics: Average 
HotToWarmMigrationProcessingLatency

The average latency of successful hot to warm migrations, not including time spent in the queue. This value is the sum of the amount of time it takes to complete the force merge, snapshot, and shard relocation stages of the migration process. Relevant statistics: Average 
HotToWarmMigrationSuccessCount

The total number of successful hot to warm migrations. Relevant statistics: Sum 
HotToWarmMigrationSuccessLatency

The average latency of successful hot to warm migrations, including time spent in the queue. Relevant statistics: Average 
WarmThreadpoolSearchThreads 
The size of the UltraWarm search thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Average, Sum 
WarmThreadpoolSearchRejected 
The number of rejected tasks in the UltraWarm search thread pool. If this number continually grows, consider adding more UltraWarm nodes. Relevant node statistics: Maximum Relevant cluster statistics: Sum 
WarmThreadpoolSearchQueue 
The number of queued tasks in the UltraWarm search thread pool. If
the queue size is consistently high, consider adding more UltraWarm
nodes. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
WarmJVMMemoryPressure 
The maximum percentage of the Java heap used for the UltraWarm nodes. Relevant statistics: Maximum NoteThe logic for this metric changed in service software R20220323. For more information, see the release notes. 
WarmOldGenJVMMemoryPressure 
The maximum percentage of the Java heap used for the "old generation" per UltraWarm node. Relevant statistics: Maximum 
WarmJVMGCYoungCollectionCount 
The number of times that "young generation" garbage collection has run on UltraWarm nodes. A large, evergrowing number of runs is a normal part of cluster operations. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
WarmJVMGCYoungCollectionTime 
The amount of time, in milliseconds, that the cluster has spent performing "young generation" garbage collection on UltraWarm nodes. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
WarmJVMGCOldCollectionCount 
The number of times that "old generation" garbage collection has run on UltraWarm nodes. In a cluster with sufficient resources, this number should remain small and grow infrequently. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
WarmConcurrentSearchRate 
The total number of search requests using concurrent segment
search per minute for all shards on a UltraWarm node. A single call
to the Relevant node statistics: Average Relevant cluster statistics: Sum, Maximum, Average 
WarmConcurrentSearchLatency 
The difference in total time, in milliseconds, taken by all searches using concurrent segment search in a UltraWarm node between minute N and minute (N1). Relevant node statistics: Average Relevant cluster statistics: Maximum, Average 
WarmThreadpoolIndexSearcherQueue 
The number of queued tasks in the UltraWarm index searcher thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Maximum, Average 
WarmThreadpoolIndexSearcherRejected 
The number of rejected tasks in the UltraWarm index searcher thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Sum 
WarmThreadpoolIndexSearcherThreads 
The size of the UltraWarm index searcher thread pool. Relevant node statistics: Maximum Relevant cluster statistics: Sum, Average 
Cold storage metrics
Amazon OpenSearch Service provides the following metrics for cold storage.
Metric  Description 

ColdStorageSpaceUtilization

The total amount of cold storage space, in MiB, that the cluster is using. Relevant statistics: Max 
ColdToWarmMigrationFailureCount 
The total number of failed cold to warm migrations. Relevant statistics: Sum 
ColdToWarmMigrationLatency 
The amount of time for successful cold to warm migrations to complete. Relevant statistics: Average 
ColdToWarmMigrationQueueSize 
The number of indexes currently waiting to migrate from cold to warm storage. Relevant statistics: Maximum 
ColdToWarmMigrationSuccessCount

The total number of successful cold to warm migrations. Relevant statistics: Sum 
WarmToColdMigrationFailureCount

The total number of failed warm to cold migrations. Relevant statistics: Sum 
WarmToColdMigrationLatency 
The amount of time for successful warm to cold migrations to complete. Relevant statistics: Average 
WarmToColdMigrationQueueSize 
The number of indexes currently waiting to migrate from warm to cold storage. Relevant statistics: Maximum 
WarmToColdMigrationSuccessCount 
The total number of successful warm to cold migrations. Relevant statistics: Sum 
OR1 metrics
Amazon OpenSearch Service provides the following metrics for OR1 instances.
Metric  Description 

RemoteStorageUsedSpace

The total amount of Amazon S3 space, in MiB, that the cluster is using. Relevant statistics: Sum 
RemoteStorageWriteRejected 
The total number of requests rejected on primary shards due to remote storage and replication pressure. This is calculated starting from the last OpenSearch Service process startup. Relevant statistics: Sum 
Alerting metrics
Amazon OpenSearch Service provides the following metrics for alerting.
Metric  Description 

AlertingDegraded 
A value of 1 means that either the alerting index is red or one or more nodes is not on schedule. A value of 0 indicates normal behavior. Relevant statistics: Maximum 
AlertingIndexExists 
A value of 1 means the Relevant statistics: Maximum 
AlertingIndexStatus.green 
The health of the index. A value of 1 means green. A value of 0 means that the index either doesn't exist or isn't green. Relevant statistics: Maximum 
AlertingIndexStatus.red 
The health of the index. A value of 1 means red. A value of 0 means that the index either doesn't exist or isn't red. Relevant statistics: Maximum 
AlertingIndexStatus.yellow 
The health of the index. A value of 1 means yellow. A value of 0 means that the index either doesn't exist or isn't yellow. Relevant statistics: Maximum 
AlertingNodesNotOnSchedule 
A value of 1 means some jobs are not running on schedule. A value
of 0 means that all alerting jobs are running on schedule (or that
no alerting jobs exist). Check the OpenSearch Service console or make a
Relevant statistics: Maximum 
AlertingNodesOnSchedule 
A value of 1 means that all alerting jobs are running on schedule (or that no alerting jobs exist). A value of 0 means some jobs are not running on schedule. Relevant statistics: Maximum 
AlertingScheduledJobEnabled 
A value of 1 means that the
Relevant statistics: Maximum 
Anomaly detection metrics
Amazon OpenSearch Service provides the following metrics for anomaly detection.
Metric  Description 

ADPluginUnhealthy 
A value of 1 means that the anomaly detection plugin is not functioning properly, either because of a high number of failures or because one of the indexes that it uses is red. A value of 0 indicates the plugin is working as expected. Relevant statistics: Maximum 
ADExecuteRequestCount 
The number of requests to detect anomalies. Relevant statistics: Sum 
ADExecuteFailureCount

The number of failed requests to detect anomalies. Relevant statistics: Sum 
ADHCExecuteFailureCount 
The number of failed requests to detect anomalies for high cardinality detectors. Relevant statistics: Sum 
ADHCExecuteRequestCount 
The number of requests to detect anomalies for high cardinality detectors. Relevant statistics: Sum 
ADAnomalyResultsIndexStatusIndexExists 
A value of 1 means the index that the
Relevant statistics: Maximum 
ADAnomalyResultsIndexStatus.red 
A value of 1 means the index that the
Relevant statistics: Maximum 
ADAnomalyDetectorsIndexStatusIndexExists 
A value of 1 means that the
Relevant statistics: Maximum 
ADAnomalyDetectorsIndexStatus.red 
A value of 1 means that the
Relevant statistics: Maximum 
ADModelsCheckpointIndexStatusIndexExists 
A value of 1 means that the
Relevant statistics: Maximum 
ADModelsCheckpointIndexStatus.red 
A value of 1 means that the
Relevant statistics: Maximum 
Asynchronous search metrics
Amazon OpenSearch Service provides the following metrics for asynchronous search.
Asynchronous search coordinator node statistics (per coordinator node)
Metric  Description 

AsynchronousSearchSubmissionRate 
The number of asynchronous searches submitted in the last minute. 
AsynchronousSearchInitializedRate 
The number of asynchronous searches initialized in the last minute. 
AsynchronousSearchRunningCurrent 
The number of asynchronous searches currently running. 
AsynchronousSearchCompletionRate 
The number of asynchronous searches successfully completed in the last minute. 
AsynchronousSearchFailureRate 
The number of asynchronous searches that completed and failed in the last minute. 
AsynchronousSearchPersistRate 
The number of asynchronous searches that persisted in the last minute. 
AsynchronousSearchPersistFailedRate 
The number of asynchronous searches that failed to persist in the last minute. 
AsynchronousSearchRejected 
The total number of asynchronous searches rejected since the node up time. 
AsynchronousSearchCancelled 
The total number of asynchronous searches cancelled since the node up time. 
AsynchronousSearchMaxRunningTime 
The duration of longest running asynchronous search on a node in the last minute. 
Asynchronous search cluster statistics
Metric  Description 

AsynchronousSearchStoreHealth 
The health of the store in the persisted index (RED/nonRED) in the last minute. 
AsynchronousSearchStoreSize 
The size of the system index across all shards in the last minute. 
AsynchronousSearchStoredResponseCount 
The numbers of stored responses in the system index in the last minute. 
AutoTune metrics
Amazon OpenSearch Service provides the following metrics for AutoTune.
Metric  Description 

AutoTuneChangesHistoryHeapSize 
The change history in MiB for heap size tuning values. 
AutoTuneChangesHistoryJVMYoungGenArgs 
The change history for JVM YongGen arguments. 
AutoTuneFailed 
A boolean that indicates if the AutoTune change failed. 
AutoTuneSucceeded 
A boolean that indicates if the AutoTune change succeeded. 
AutoTuneValue 
The queue change history (count) and cache tunings change history (in MiB) for nondisruptive changes. 
MultiAZ with Standby metrics
Amazon OpenSearch Service provides the following metrics for MultiAZ with Standby.
Nodelevel metrics for data nodes in active Availability Zones
Metric  Description 

CPUUtilization 
The percentage of CPU usage for data nodes in the cluster. Maximum shows the node with the highest CPU usage. Average represents all nodes in the cluster. This metric is also available for individual nodes. 
FreeStorageSpace 
The free space for data nodes in the cluster. The OpenSearch Service console displays this value in GiB. The Amazon CloudWatch console displays it in MiB. 
JVMMemoryPressure 
The maximum percentage of the Java heap used for all data nodes in the cluster. OpenSearch Service uses half of an instance's RAM for the Java heap, up to a heap size of 32 GiB. You can scale instances vertically up to 64 GiB of RAM, at which point you can scale horizontally by adding instances. See Recommended CloudWatch alarms for Amazon OpenSearch Service. 
SysMemoryUtilization 
The percentage of the instance's memory that is in use. High
values for this metric are normal and usually do not represent a problem
with your cluster. For a better indicator of potential performance and
stability issues, see the JVMMemoryPressure metric. 
IndexingLatency 
The difference in total time, in milliseconds, taken by all indexing operations in a node between minute N and minute (N1). 
IndexingRate 
The number of indexing operations per minute. 
SearchLatency 
The difference in total time, in milliseconds, taken by all searches in a node between minute N and minute (N1). 
SearchRate 
The total number of search requests per minute for all shards on a data node. 
ThreadpoolSearchQueue 
The number of queued tasks in the search thread pool. If the queue size is consistently high, consider scaling your cluster. The maximum search queue size is 1,000. 
ThreadpoolWriteQueue 
The number of queued tasks in the write thread pool. 
ThreadpoolSearchRejected 
The number of rejected tasks in the search thread pool. If this number continually grows, consider scaling your cluster. 
ThreadpoolWriteRejected 
The number of rejected tasks in the write thread pool. 
Clusterlevel metrics for clusters in active Availability Zones
Metric  Description 

DataNodes 
The total number of active and standby shards. 
DataNodesShards.active 
The total number of active primary and replica shards. 
DataNodesShards.unassigned 
The number of shards that are not allocated to nodes in the cluster. 
DataNodesShards.initializing 
The number of shards that are under initialization. 
DataNodesShards.relocating 
The number of shards that are under relocation. 
Availability Zone rotation metrics
If ActiveReads.
, then the
zone is active. If AvailabilityZone
= 1ActiveReads.
, then the zone is in standby.AvailabilityZone
=
0
Point in time metrics
Amazon OpenSearch Service provides the following metrics for point in time (PIT) searches.
PIT coordinator node statistics (per coordinator node)
Metric  Description 

CurrentPointInTime 
The number of active PIT search contexts in the node. 
TotalPointInTime 
The number of expired PIT search contexts since the node up time. 
AvgPointInTimeAliveTime 
The average keep alive of PIT search contexts since the node up time. 
HasActivePointInTime 
A value of 1 indicates that there are active PIT contexts on nodes since the node up time. A value of 0 means there are not. 
HasUsedPointInTime 
A value of 1 indicates that there are expired PIT contexts on nodes since the node up time. A value of 0 means there are not. 
SQL metrics
Amazon OpenSearch Service provides the following metrics for SQL support.
Metric  Description 

SQLFailedRequestCountByCusErr 
The number of requests to the Relevant statistics: Sum 
SQLFailedRequestCountBySysErr 
The number of requests to the Relevant statistics: Sum 
SQLRequestCount 
The number of requests to the Relevant statistics: Sum 
SQLDefaultCursorRequestCount 
Similar to Relevant statistics: Sum 
SQLUnhealthy 
A value of 1 indicates that, in response to certain requests, the SQL plugin is returning 5xx response codes or passing invalid query DSL to OpenSearch. Other requests should continue to succeed. A value of 0 indicates no recent failures. If you see a sustained value of 1, troubleshoot the requests your clients are making to the plugin. Relevant statistics: Maximum 
kNN metrics
Amazon OpenSearch Service includes the following metrics for the knearest neighbor (kNN) plugin.
Metric  Description 

KNNCacheCapacityReached 
Pernode metric for whether cache capacity has been reached. This metric is only relevant to approximate kNN search. Relevant statistics: Maximum 
KNNCircuitBreakerTriggered 
Percluster metric for whether the circuit breaker is triggered.
If any nodes return a value of 1 for
Relevant statistics: Maximum 
KNNEvictionCount 
Pernode metric for the number of graphs that have been evicted from the cache due to memory constraints or idle time. Explicit evictions that occur because of index deletion are not counted. This metric is only relevant to approximate kNN search. Relevant statistics: Sum 
KNNGraphIndexErrors 
Pernode metric for the number of requests to add the
Relevant statistics: Sum 
KNNGraphIndexRequests 
Pernode metric for the number of requests to add the
Relevant statistics: Sum 
KNNGraphMemoryUsage 
Pernode metric for the current cache size (total size of all graphs in memory) in kilobytes. This metric is only relevant to approximate kNN search. Relevant statistics: Average 
KNNGraphQueryErrors 
Pernode metric for the number of graph queries that produced an error. Relevant statistics: Sum 
KNNGraphQueryRequests 
Pernode metric for the number of graph queries. Relevant statistics: Sum 
KNNHitCount 
Pernode metric for the number of cache hits. A cache hit occurs when a user queries a graph that is already loaded into memory. This metric is only relevant to approximate kNN search. Relevant statistics: Sum 
KNNLoadExceptionCount 
Pernode metric for the number of times an exception occurred while trying to load a graph into the cache. This metric is only relevant to approximate kNN search. Relevant statistics: Sum 
KNNLoadSuccessCount 
Pernode metric for the number of times the plugin successfully loaded a graph into the cache. This metric is only relevant to approximate kNN search. Relevant statistics: Sum 
KNNMissCount 
Pernode metric for the number of cache misses. A cache miss occurs when a user queries a graph that is not yet loaded into memory. This metric is only relevant to approximate kNN search. Relevant statistics: Sum 
KNNQueryRequests 
Pernode metric for the number of query requests the kNN plugin received. Relevant statistics: Sum 
KNNScriptCompilationErrors 
Pernode metric for the number of errors during script compilation. This statistic is only relevant to kNN score script search. Relevant statistics: Sum 
KNNScriptCompilations 
Pernode metric for the number of times the kNN script has been compiled. This value should usually be 1 or 0, but if the cache containing the compiled scripts is filled, the kNN script might be recompiled. This statistic is only relevant to kNN score script search. Relevant statistics: Sum 
KNNScriptQueryErrors 
Pernode metric for the number of errors during script queries. This statistic is only relevant to kNN score script search. Relevant statistics: Sum 
KNNScriptQueryRequests 
Pernode metric for the total number of script queries. This statistic is only relevant to kNN score script search. Relevant statistics: Sum 
KNNTotalLoadTime 
The time in nanoseconds that kNN has taken to load graphs into the cache. This metric is only relevant to approximate kNN search. Relevant statistics: Sum 
Crosscluster search metrics
Amazon OpenSearch Service provides the following metrics for crosscluster search.
Source domain metrics
Metric  Dimension  Description 

CrossClusterOutboundConnections 

Number of connected nodes. If your response includes one or more skipped domains, use this metric to trace any unhealthy connections. If this number drops to 0, then the connection is unhealthy. 
CrossClusterOutboundRequests 

Number of search requests sent to the destination domain. Use to check if the load of crosscluster search requests are overwhelming your domain, correlate any spike in this metric with any JVM/CPU spike. 
Destination domain metric
Metric  Dimension  Description 

CrossClusterInboundRequests 

Number of incoming connection requests received from the source domain. 
Add a CloudWatch alarm in the event that you lose a connection unexpectedly. For steps to create an alarm, see Create a CloudWatch Alarm Based on a Static Threshold.
Crosscluster replication metrics
Amazon OpenSearch Service provides the following metrics for crosscluster replication.
Metric  Description 

ReplicationRate 
The average rate of replication operations per second. This metric
is similar to the 
LeaderCheckPoint 
For a specific connection, the sum of leader checkpoint values across all replicating indexes. You can use this metric to measure replication latency. 
FollowerCheckPoint 
For a specific connection, the sum of follower checkpoint values across all replicating indexes. You can use this metric to measure replication latency. 
ReplicationNumSyncingIndices 
The number of indexes that have a replication status of

ReplicationNumBootstrappingIndices 
The number of indexes that have a replication status of

ReplicationNumPausedIndices 
The number of indexes that have a replication status of

ReplicationNumFailedIndices 
The number of indexes that have a replication status of


The number of replication transport requests on the follower domain. Transport requests are internal and occur each time a replication API operation is called. They also occur when the follower domain polls changes from the leader domain. 

The number of replication transport requests on the leader domain. Transport requests are internal and occur each time a replication API operation is called. 
AutoFollowNumSuccessStartReplication 
The number of follower indexes that have been successfully created by a replication rule for a specific connection. 
AutoFollowNumFailedStartReplication 
The number of follower indexes that failed to be created by a replication rule when there was a matching pattern. This problem might arise due to a network issue on the remote cluster, or a security issue (i.e. the associated role doesn't have permission to start replication). 
AutoFollowLeaderCallFailure 
Whether there have been any failed queries from the follower index
to the leader index to pull new data. A value of 
Learning to Rank metrics
Amazon OpenSearch Service provides the following metrics for Learning to Rank.
Metric  Description 

LTRRequestTotalCount 
Total count of ranking requests. 
LTRRequestErrorCount 
Total count of unsuccessful requests. 
LTRStatus.red 
Tracks if one of the indexes needed to run the plugin is red. 
LTRMemoryUsage 
Total memory used by the plugin. 
LTRFeatureMemoryUsageInBytes 
The amount of memory, in bytes, used by Learning to Rank feature fields. 
LTRFeaturesetMemoryUsageInBytes 
The amount of memory, in bytes, used by all Learning to Rank feature sets. 
LTRModelMemoryUsageInBytes 
The amount of memory, in bytes, used by all Learning to Rank models. 
Piped Processing Language metrics
Amazon OpenSearch Service provides the following metrics for Piped Processing Language.
Metric  Description 

PPLFailedRequestCountByCusErr 
The number of requests to the 
PPLFailedRequestCountBySysErr 
The number of requests to the 
PPLRequestCount 
The number of requests to the 