If Yarn is restarted in either of the following scenarios, local logs will not be deleted as scheduled and will be retained permanently:
NodeManager has a restart recovery mechanism (for details, see https://hadoop.apache.org/docs/r3.1.1/hadoop-yarn/hadoop-yarn-site/NodeManager.html#NodeManager_Restart). Go to the All Configurations page of Yarn by referring to Modifying Cluster Service Configuration Parameters. Set yarn.nodemanager.recovery.enabled of NodeManager to true to make the configuration take effect. The default value is true. In this way, redundant local logs are periodically deleted when the YARN is restarted.