Parameter | Description |
Minimum Number of Nodes | Minimum number of task nodes retained for auto scaling in the cluster when the automatic scale-in policy is triggered. |
Maximum Number of Nodes | Maximum number of task nodes retained for auto scaling in the cluster when the automatic scale-out policy is triggered. The cumulative number of nodes scaled out by one or more specifications cannot exceed the maximum number of nodes. |
Release All | One-click removal of all nodes scaled out by auto scaling, without affecting nodes not scaled out by auto scaling. |
Release Spot Instances | One-click removal of only the spot instance nodes scaled out by auto scaling, without affecting non-spot instance resource nodes. |
Release Pay-As-You-Go Instances | One-click removal of the pay-as-you-go instance nodes scaled out by auto scaling, without affecting pay-as-you-go nodes not scaled out by auto scaling. |
Global Switch of Graceful Scale-In | Disabled by default, which means the graceful scale-in policy is disabled in any scale-in rule. After it is enabled, the graceful scale-in policy takes effect when graceful scale-in and a single rule are enabled at the same time. |
Resource Type | The HOST resource type supports pay-as-you-go and spot instance billing modes, while POD resources support the pay-as-you-go billing mode only and can only be used to deploy the NodeManager role of YARN. |
Configuration Item | Description |
Rule Type | Scale-out and scale-in. |
Policy Type | By Load |
Rule Name | Name of the scaling rule. In a cluster, scaling rule names (including scale-out and scale-in rule names) must be unique. |
Validity Period | The load-based scaling rule is triggered only within the validity period. No Limit is selected for the time range by default, and custom time periods are supported for configuring load-based scaling rules. |
Load Type | YARN or Trino load metrics are supported. Trino load-based scaling is supported only by clusters of EMR-V2.7.0 and EMR-V3.40 or later that have the Trino component deployed. |
Statistical Rule | Set trigger thresholds for single or multiple rules based on the selected cluster load metrics. Up to five statistical rules can be set, and rule statistics by subqueue is supported. Rule: specifies the queue and load metrics, and sets conditional rules for the trigger threshold. Statistical period: Within a statistical period, it counts as one trigger when the selected load metric reaches the trigger threshold according to the selected aggregation dimension (average, maximum, or minimum). Three statistical periods are currently supported: 300 seconds, 600 seconds, and 900 seconds. Repetition count: The number of times the load metric reaches the threshold after aggregation. The elastic scaling action of the cluster is triggered after this count is reached. |
Scale-Out/Scale-In Method | Three methods are supported: node, memory, and number of cores. Only non-zero integers are accepted as input values for all three methods. When the method is number of cores or memory, the system will calculate the number of nodes to be scaled out based on maximum computing power, and the system will calculate the minimum number of nodes to be removed while ensuring business continuity, scaling in in reverse chronological order and ensuring that at least one node is scaled in. |
Scale-Out Service | Scale-out components inherit the cluster-level configuration by default, and scaled-out nodes belong to the default configuration group of the node type. To adjust the scale-out component configuration, you can specify configuration settings. |
Node Label | By default, when this parameter is left empty, scaled-out resources are marked with the default label. Once configured, scaled-out resources will be marked with the specified label. |
Resource Replenishment Retry | During peak order placement, automatic scale-out may result in the actual number of scale-out machines failing to reach the elastic target quantity due to resource contention. When the resource replenishment retry policy is enabled and the configured scaling specification resources are sufficient, the system will automatically retry resource requests until the target quantity is achieved or approached. If automatic scale-out frequently falls short of expectations due to insufficient resources, you can enable this configuration. Note that enabling retries may extend the automatic scale-out time. Pay attention to the impact of the policy adjustment on your business. |
Cooldown Period | Interval between the successful execution of the current rule and the start of the next auto scaling action (the cooldown period ranges from 0 to 43200 seconds). |
Graceful Scale-In | After graceful scale-in mode is enabled, if a node is running a task when the scale-in action is triggered, the node will not be released immediately but waits for the task to complete within a custom time period before being scaled in. If the task is not completed when the custom time period ends, the node will still be scaled in. |
Configuration Item | Description |
Rule Type | Scale-out and scale-in. |
Policy Type | By Time |
Rule Name | Name of the scaling rule. In a cluster, scaling rule names (including scale-out and scale-in rule names) must be unique. |
Execution Type | Execute once: The scaling action is triggered at a specific time, accurate to the minute. Repeat: The scaling action is triggered at each specified time period or at a specific time. Daily, Weekly, and Monthly are supported. Execution time: The specific time to perform the scaling action each day. Rule validity period: The validity period range for triggering a single repeated rule. |
Scale-Out/Scale-In Method | Three methods are supported: node, memory, and number of cores. Only non-zero integers are accepted as input values for all three methods. When the method is number of cores or memory, the system will calculate the number of nodes to be scaled out based on maximum computing power, and the system will calculate the minimum number of nodes to be removed while ensuring business continuity, scaling in in reverse chronological order and ensuring that at least one node is scaled in. |
Scale-Out Service | Scale-out components inherit the cluster-level configuration by default, and scaled-out nodes belong to the default configuration group of the node type. To adjust the scale-out component configuration, you can specify configuration settings. |
Node Label | By default, when this parameter is left empty, scaled-out resources are marked with the default label. Once configured, scaled-out resources will be marked with the specified label. |
Resource Replenishment Retry | During peak order placement, automatic scale-out may result in the actual number of scale-out machines failing to reach the elastic target quantity due to resource contention. When the resource replenishment retry policy is enabled and the configured scaling specification resources are sufficient, the system will automatically retry resource requests until the target quantity is achieved or approached. If automatic scale-out frequently falls short of expectations due to insufficient resources, you can enable this configuration. Note that enabling retries may extend the automatic scale-out time. Pay attention to the impact of the policy adjustment on your business. |
Retry Expiration Time | Elastic scaling may fail to execute at the specified time for various reasons. By setting the retry expiration time, the system retries execution at regular intervals within the time range until the scaling is executed when the conditions are met. |
Cooldown Period | Interval between the successful execution of the current rule and the start of the next auto scaling action (the cooldown period ranges from 0 to 43200 seconds). |
Scheduled Termination | Specifies the usage duration of scaled-out resources, so that the current batch of nodes is not affected when scale-in rules are triggered. No Limit is selected by default, and a custom termination duration is supported. Enter an integer in the range of 1 to 24 hours. Usage scenario: Suitable when computing power needs to be supplemented during fixed periods and maintained within one day, and other scale-in rules do not affect this batch of resources. |
Graceful Scale-In | After graceful scale-in mode is enabled, if a node is running a task when the scale-in action is triggered, the node will not be released immediately but waits for the task to complete within a custom time period before being scaled in. If the task is not completed when the custom time period ends, the node will still be scaled in. |
フィードバック