Overview
After graceful scale-in is enabled, if the scale-in action is triggered when a node is executing tasks, the node will not be released immediately. Instead, it will be released after completing the tasks. Graceful scale-in is available for both automatic scaling and manual scale-in.
Note:
YARN, HBase, and Presto (renamed Trino in EMR v2.7.0 and v3.40 or later) support graceful scale-in. However, Presto cannot be scaled in gracefully in Ranger, Kerberos, and OpenLDAP integration scenarios.
Directions
Auto-scaling
Auto-scaling allows you to enable or disable graceful scale-in for all scale-in rules. This feature is called global graceful scale-in and is disabled by default. When you add or edit a single scale-in rule, graceful scale-in is enabled by default, and the default duration is 60 seconds (valid range: 60–86400 seconds).
Note:
When both graceful scale-in and a single scale-in rule are enabled, the graceful scale-in takes effect.
1. Log in to the EMR console, click the ID/name of the target cluster to enter the cluster details page, and click Auto-scaling. 2. In the Scaling rule management section on the Auto-scaling page, click Add rule and add a scale-in rule.
Manual scale-in
When you try to manually remove a node, graceful scale-in is disabled by default. When you enable this feature, the default duration is 60 seconds (valid range: 60–86400 seconds).
1. Click the ID/name of the target cluster to enter the cluster details page. Then, select Cluster resources > Resources.
2. Select the node to remove and click Scale in. In this case, graceful scale-in is disabled by default. You can enable it and set a duration.
3. After finishing your settings, click Next, confirm the node information, and click Start termination.