Making configuration changes in Amazon OpenSearch Service

Amazon OpenSearch Service uses a blue/green deployment process when updating domains. A blue/green deployment creates an idle environment for domain updates that copies the production environment, and routes users to the new environment after those updates are complete. In a blue/green deployment, the blue environment is the current production environment. The green environment is the idle environment.

Data is migrated from the blue environment to the green environment. When the new environment is ready, OpenSearch Service switches over the environments to promote the green environment to be the new production environment. The switchover happens with no data loss. This practice minimizes downtime and maintains the original environment in the event that deployment to the new environment is unsuccessful.

Changes that usually cause blue/green deployments

The following operations cause blue/green deployments:

• Changing instance type

• Enabling fine-grained access control

• Performing service software updates

• If your domain doesn't have dedicated master nodes, changing data instance count

• Enabling or disabling dedicated master nodes

• Enabling or disabling Multi-AZ without Standby

• Changing storage type, volume type, or volume size

• Choosing different VPC subnets

• Adding or removing VPC security groups

• Enabling or disabling Amazon Cognito authentication for OpenSearch Dashboards

• Choosing a different Amazon Cognito user pool or identity pool

• Modifying advanced settings

• Upgrading to a new OpenSearch version

• Enabling encryption of data at rest or node-to-node encryption

• Enabling or disabling UltraWarm or cold storage

• Disabling Auto-Tune and rolling back its changes

• Associating an optional plugin to a domain and dissociating an optional plugin from a domain

• Increasing dedicated master node count for domains with two dedicated master nodes and zone awareness enabled

For Multi-AZ with Standby domains, you can only make one change request at a time. If a change is already in progress, the new request will be rejected. You can check the status of the current change with the DescribeDomainChangeProgress API.

When you upgrade domains, OpenSearch Dashboards might be unavailable during some or all of the upgrade. The upgrade can take from 15 minutes to several hours to complete.

When changing storage type, volume type, or volume size, a blue/green deployment might not be required. If your volume size per data node is less than 3 TiB, your instance types are not from a previous version, your previous volume update is not progressing, or if you have an update in progress, a blue/green deployment is not required.

Changes that usually don't cause blue/green deployments

In most cases, the following operations do not cause blue/green deployments:

• Changing access policy

• Modifying the custom endpoint

• Changing the Transport Layer Security (TLS) policy

• Changing the automated snapshot hour

• Changing volume size, volume type, IOPS, and throughput up to 3 TiB

• Enabling or disabling Require HTTPS

• Enabling Auto-Tune or disabling it without rolling back its changes

• If your domain has dedicated master nodes, changing data node or UltraWarm node count

• If your domain has dedicated master nodes, changing dedicated master instance type or node count (except for domains with two dedicated masters and zone awareness enabled)

• Enabling or disabling the publication of error logs, audit logs, or slow logs to CloudWatch

There are some exceptions depending on your service software version. If you want to be absolutely sure that a change will not cause a blue/green deployment, perform a dry run before updating your domain.

Determining whether a change will cause a blue/green deployment

You can test planned domain configuration changes to determine whether they will cause a blue/green deployment without having to commit to those changes. Before you initiate a configuration change, use the console or an API to run a validation check to ensure that your domain is eligible for an update.

Console

To validate a cluster configuration change, complete the following steps.

1. Navigate to the Amazon OpenSearch Service console at https://console.aws.amazon.com/aos/.

2. In the left navigation pane, choose Domains.

3. Select the domain you want to make a configuration change for. This opens the domain details page. Select the Actions dropdown menu and then choose Edit cluster configuration.

4. On the Edit cluster configuration page, you can make changes to the instance type, the number of nodes, and any other configurations. After you've confirmed your changes in the summary panel, choose Run.

5. Once your dry run is complete, the results automatically display at the bottom of the page, along with a dry run ID. These results notify you which category your change falls into:

   • Initiates a blue/green deployment

   • Doesn't require a blue/green deployment

   • Contains validation errors that you need to address before you can save your changes

   Note that each dry run overwrites the one before it. To look up the details of each dry run later on, make sure you save your dry run ID. Each dry run is available for 90 days, or until you make a configuration update.

6. To proceed with your configuration update, choose Save changes. Otherwise, choose Cancel. Either option takes you back to the Cluster configuration tab. On this tab, you can choose Dry run details to see the details of your latest dry run. This page also includes a side-by-side comparison between the configuration before the dry run and the dry run configuration.

API

You can perform a dry run validation through the configuration API. To test your changes with the API, set DryRun to true, and DryRunMode to Verbose. Verbose mode runs a validation check in addition to determining whether the change will initiate a blue/green deployment. For example, this UpdateDomainConfig request tests the deployment type that results from enabling UltraWarm:

POST https://es.us-east-1.amazonaws.com/2021-01-01/opensearch/domain/my-domain/config
{
   "ClusterConfig": {
    "WarmCount": 3,
    "WarmEnabled": true,
    "WarmType": "ultrawarm1.large.search"
   },
   "DryRun": true,
   "DryRunMode": "Verbose"
}

The request runs a validation check and returns the type of deployment the change will cause but doesn't actually perform the update:

{
   "ClusterConfig": {
     ...
    },
   "DryRunResults": {
      "DeploymentType": "Blue/Green",
      "Message": "This change will require a blue/green deployment."
    }
}

Possible deployment types are:

• Blue/Green – The change will cause a blue/green deployment.

• DynamicUpdate – The change won't cause a blue/green deployment.

• Undetermined – The domain is still in a processing state, so the deployment type can't be determined.

• None – No configuration change.

If the validation fails, it returns a list of validation failures.

{
   "ClusterConfig":{
      "..."
   },
   "DryRunProgressStatus":{
      "CreationDate":"2023-01-12T01:14:33.847Z",
      "DryRunId":"db00ca39-48b2-4774-bbd3-252cf094d205",
      "DryRunStatus":"failed",
      "UpdateDate":"2023-01-12T01:14:33.847Z",
      "ValidationFailures":[
         {
            "Code":"Cluster.Index.WriteBlock",
            "Message":"Cluster has index write blocks."
         }
      ]
   }
}

If the status is still pending, you can use the dry run ID in your UpdateDomainConfig response in subsequent DescribeDryRunProgress calls to check the status of the validation.

GET https://es.us-east-1.amazonaws.com/2021-01-01/opensearch/domain/my-domain/dryRun?dryRunId=my-dry-run-id
{
    "DryRunConfig": null,
    "DryRunProgressStatus": {
        "CreationDate": "2023-01-12T01:14:42.998Z",
        "DryRunId": "db00ca39-48b2-4774-bbd3-252cf094d205",
        "DryRunStatus": "succeeded",
        "UpdateDate": "2023-01-12T01:14:49.334Z",
        "ValidationFailures": null
    },
    "DryRunResults": {
        "DeploymentType": "Blue/Green",
        "Message": "This change will require a blue/green deployment."
    }
}

To run a dry run analysis without a validation check, set DryRunMode to Basic when you use the configuration API.

Python

The following Python code uses the UpdateDomainConfig API to perform a dry run validation check and, if the check succeeds, calls the same API without a dry run to start the update. If the check fails, the script prints out the error and stops.

import time
import boto3

client = boto3.client('opensearch')

response = client.UpdateDomainConfig(
    ClusterConfig={
        'WarmCount': 3,
        'WarmEnabled': True,
        'WarmCount': 123,
    },
    DomainName='test-domain',
    DryRun=True,
    DryRunMode='Verbose'
)

dry_run_id = response.DryRunProgressStatus.DryRunId

retry_count = 0

while True:

    if retry_count == 5:
        print('An error occured')
        break

    dry_run_progress_response = client.DescribeDryRunProgress('test-domain', dry_run_id)
    dry_run_status = dry_run_progress_response.DryRunProgressStatus.DryRunStatus

    if dry_run_status == 'succeeded':
        client.UpdateDomainConfig(
            ClusterConfig={
            'WarmCount': 3,
            'WarmEnabled': True,
            'WarmCount': 123,
        })
        break

    elif dry_run_status == 'failed':
        validation_failures_list = dry_run_progress_response.DryRunProgressStatus.ValidationFailures
        for item in validation_failures_list:
            print(f"Code: {item['Code']}, Message: {item['Message']}")
        break

    retry_count += 1
    time.sleep(30)          

Initiating a configuration change

When you initiate a configuration change, the domain state changes to Processing until OpenSearch Service has created a new environment with the latest service software, at which point it changes back to Active. During certain service software updates, the state remains Active the whole time. In both cases, you can review the cluster health and Amazon CloudWatch metrics and see that the number of nodes in the cluster temporarily increases—often doubling—while the domain update occurs. In the following illustration, you can see the number of nodes doubling from 11 to 22 during a configuration change and returning to 11 when the update is complete.

This temporary increase can strain the cluster's dedicated master nodes, which suddenly might have many more nodes to manage. It can also increase search and indexing latencies as OpenSearch Service copies data from the old cluster to the new one. It's important to maintain sufficient capacity on the cluster to handle the overhead that is associated with these blue/green deployments.

Important

You do not incur any additional charges during configuration changes and service maintenance. You're billed only for the number of nodes that you request for your cluster. For specifics, see Charges for configuration changes.

To prevent overloading dedicated master nodes, you can monitor usage with the Amazon CloudWatch metrics. For recommended maximum values, see Recommended CloudWatch alarms for Amazon OpenSearch Service.

Stages of a configuration change

After you initiate a configuration change, OpenSearch Service goes through a series of steps to update your domain. You can view the progress of the configuration change under Domain status in the console. The exact steps that an update goes through depends on the type of change you're making. You can also monitor a configuration change using the DescribeDomainChangeProgress API operation.

In some cases, such as during service software updates, you won't see progress information until the blue/green deployment actually starts. During this time the domain status is Pending Updates.

The following are possible stages an update can go through during a configuration change:

Stage name	Description
Validation	Validating that the domain is eligible for an update, and surfacing validation issues if necessary.
Creating a new environment	Completing the necessary prerequisites and creating required resources to start the blue/green deployment.
Provisioning new nodes	Creating a new set of instances in the new environment.
Traffic routing on new nodes	Redirecting traffic to the newly created data nodes.
Traffic routing on old nodes	Disabling traffic on the old data nodes.
Preparing nodes for removal	Preparing to remove nodes. This step only happens when you're downscaling your domain (for example, from 8 nodes to 6 nodes).
Copying shards to new nodes	Moving shards from the old nodes to the new nodes.
Terminating nodes	Terminating and deleting old nodes after shards are removed.
Deleting older resources	Deleting resources associated with the old environment (e.g. load balancer).
Dynamic update	Displayed when the update does not require a blue/green deployment and can be dynamically applied.
Applying dedicated master related changes	Displayed when the dedicated master instance type or count is changed.
Applying volume related changes	Displayed when volume size, type, IOPS and throughput are changed.

Troubleshooting validation errors

When you initiate a configuration change or perform an OpenSearch or Elasticsearch version upgrade, OpenSearch Service first performs a series of validation checks to ensure that your domain is eligible for an update. If any of these checks fail, you receive a notification in the console containing the specific issues that you must fix before updating your domain. The following table lists the possible domain issues that OpenSearch Service might surface, and steps to resolve them.

Issue	Error code	Troubleshooting steps
Security group not found	SecurityGroupNotFound	The security group associated with your OpenSearch Service domain does not exist. To resolve this issue, create a security group with the specified name.
Subnet not found	SubnetNotFound	The subnet associated with your OpenSearch Service domain does not exist. To resolve this issue, create a subnet in your VPC.
Service-linked role not configured	SLRNotConfigured	The service-linked role for OpenSearch Service is not configured. The service-linked role is predefined by OpenSearch Service and includes all the permissions the service requires to call other AWS services on your behalf. If the role doesn't exist, you might need to create it manually.
Not enough IP addresses	InsufficientFreeIPsForSubnets	One or more of your VPC subnets don't have enough IP addresses to update your domain. To calculate how many IP addresses you need, see Reserving IP addresses in a VPC subnet.
Cognito user pool doesn't exist	CognitoUserPoolNotFound	OpenSearch Service can't find the Amazon Cognito user pool. Confirm that you created one and have the correct ID. To find the ID, you can use the Amazon Cognito console or the following AWS CLI command: aws cognito-idp list-user-pools --max-results 60 --region us-east-1
Cognito identity pool doesn't exist	CognitoIdentityPoolNotFound	OpenSearch Service can't find the Cognito identity pool. Confirm that you created one and have the correct ID. To find the ID, you can use the Amazon Cognito console or the following AWS CLI command: aws cognito-identity list-identity-pools --max-results 60 --region us-east-1
Cognito domain not found for user pool	CognitoDomainNotFound	The user pool does not have a domain name. You can configure one using the Amazon Cognito console or the following AWS CLI command: aws cognito-idp create-user-pool-domain --domain my-domain --user-pool-id id
Cognito role not configured	CognitoRoleNotConfigured	The IAM role that grants OpenSearch Service permission to configure the Amazon Cognito user and identity pools, and use them for authentication, is not configured. Configure the role with an appropriate permission set and trust relationship. You can use the console, which creates the default CognitoAccessForAmazonOpenSearch role for you, or you can manually configure a role using the AWS CLI or the AWS SDK.
Unable to describe user pool	UserPoolNotDescribable	The specified Amazon Cognito role doesn't have permission to describe the user pool associated with your domain. Make sure the role permissions policy allows the cognito-identity:DescribeUserPool action. See About the CognitoAccessForAmazonOpenSearch role for the full permissions policy.
Unable to describe identity pool	IdentityPoolNotDescribable	The specified Amazon Cognito role doesn't have permission to describe the identity pool associated with your domain. Make sure the role permissions policy allows the cognito-identity:DescribeIdentityPool action. See About the CognitoAccessForAmazonOpenSearch role for the full permissions policy.
Unable to describe user and identity pool	CognitoPoolsNotDescribable	The specified Amazon Cognito role doesn't have permission to describe the user and identity pools associated with your domain. Make sure the role permissions policy allows the cognito-identity:DescribeIdentityPool and cognito-identity:DescribeUserPool actions. See About the CognitoAccessForAmazonOpenSearch role for the full permissions policy.
KMS key not enabled	KMSKeyNotEnabled	The AWS Key Management Service (AWS KMS) key used to encrypt your domain is disabled. Re-enable the key immediately.
Custom certificate not in ISSUED state	InvalidCertificate	If your domain uses a custom endpoint, you secure it by either generating an SSL certificate in AWS Certificate Manager (ACM) or importing one of your own. The certificate status must be Issued. If you receive this error, check the status of your certificate in the ACM console. If the status is Expired, Failed, Inactive, or Pending validation, see the ACM troubleshooting documentation to resolve the issue.
Not enough capacity to launch chosen instance type	InsufficientInstanceCapacity	The requested instance type capacity is not available. For example, you might have requested five i3.16xlarge.search nodes, but OpenSearch Service doesn't have enough i3.16xlarge.search hosts available, so the request can't be fulfilled. Check the supported instance types in OpenSearch Service and choose a different instance type.
Red indexes in cluster	RedCluster	One or more indexes in your cluster have a red status, leading to an overall red cluster status. To troubleshoot and remediate this issue, see Red cluster status.
Memory circuit breaker, too many requests	TooManyRequests	There are too many search and write requests to your domain, so OpenSearch Service can't update its configuration. You can reduce the number of requests, scale instances vertically up to 64 GiB of RAM, or scale horizontally by adding instances.
New configuration can't hold data (low disk space)	InsufficientStorageCapacity	The configured storage size can't hold all of the data on your domain. To resolve this issue, choose a larger volume, delete unused indexes, or increase the number of nodes in the cluster to immediately free up disk space.
Shards pinned to specific nodes	ShardMovementBlocked	One or more indexes in your domain are attached to specific nodes and can't be reassigned. This most likely happened because you configured shard allocation filtering, which lets you specify which nodes are allowed to host the shards of a particular index. To resolve this issue, remove shard allocation filters from all affected indexes: PUT my-index/_settings { "settings": { "index.routing.allocation.require._name": null } }
New configuration can't hold all shards (shard count)	TooManyShards	The shard count on your domain is too high, which prevents OpenSearch Service from moving them to the new configuration. To resolve this issue, scale your domain horizonally by adding nodes of the same configuration type as your current cluster nodes. Note that the maximum EBS volume size depends on the node's instance type. To prevent this issue in the future, see Choosing the number of shards and define a sharding strategy that is appropriate for your use case.

Charges for configuration changes

If you change the configuration for a domain, OpenSearch Service creates a new cluster as described in Making configuration changes in Amazon OpenSearch Service. During the migration of old to new, you incur the following charges:

• If you change the instance type, you're charged for both clusters for the first hour. After the first hour, you're only charged for the new cluster. EBS volumes aren't charged twice because they're part of your cluster, so their billing follows instance billing.

  Example: You change the configuration from three m3.xlarge instances to four m4.large instances. For the first hour, you're charged for both clusters (3 * m3.xlarge + 4 * m4.large). After the first hour, you're charged only for the new cluster (4 * m4.large).

• If you don't change the instance type, you're charged only for the largest cluster for the first hour. After the first hour, you're charged only for the new cluster.

  Example: You change the configuration from six m3.xlarge instances to three m3.xlarge instances. For the first hour, you're charged for the largest cluster (6 * m3.xlarge). After the first hour, you're charged only for the new cluster (3 * m3.xlarge).