wangzhihui created YARN-11626: --------------------------------- Summary: Optimization of the safeDelete operation in ZKRMStateStore Key: YARN-11626 URL: https://issues.apache.org/jira/browse/YARN-11626 Project: Hadoop YARN Issue Type: Improvement Components: resourcemanager Affects Versions: 3.3.0, 3.1.1, 3.0.0-alpha4 Reporter: wangzhihui
h1. Description * We can be observed that removing app info started at 06:17:20, but the NoNodeException was received at 06:17:35. * During the 15s interval, Curator was retrying the metadata operation. Due to the non-idempotent nature of the Zookeeper deletion operation, in one of the retry attempts, the metadata operation was successful but no response was received. In the next retry it resulted in a NoNodeException, triggering the STATE_STORE_FENCED event and ultimately causing the current ResourceManager to switch to standby . {code:java} 2023-10-28 06:17:20,359 INFO recovery.RMStateStore (RMStateStore.java:transition(333)) - Removing info for app: application_1697410508608_140368 2023-10-28 06:17:20,359 INFO resourcemanager.RMAppManager (RMAppManager.java:checkAppNumCompletedLimit(303)) - Application should be expired, max number of completed apps kept in memory met: maxCompletedAppsInMemory = 1000, removing app application_1697410508608_140368 from memory: 2023-10-28 06:17:35,665 ERROR recovery.RMStateStore (RMStateStore.java:transition(337)) - Error removing app: application_1697410508608_140368 org.apache.zookeeper.KeeperException$NoNodeException: KeeperErrorCode = NoNode at org.apache.zookeeper.KeeperException.create(KeeperException.java:111) 2023-10-28 06:17:35,666 INFO recovery.RMStateStore (RMStateStore.java:handleStoreEvent(1147)) - RMStateStore state change from ACTIVE to FENCED 2023-10-28 06:17:35,666 ERROR resourcemanager.ResourceManager (ResourceManager.java:handle(898)) - Received RMFatalEvent of type STATE_STORE_FENCED, caused by org.apache.zookeeper.KeeperException$NoNodeException: KeeperErrorCode = NoNode 2023-10-28 06:17:35,666 INFO resourcemanager.ResourceManager (ResourceManager.java:transitionToStandby(1309)) - Transitioning to standby state {code} h1. Solution The NoNodeException clearly indicates that the Znode no longer exists, so we can safely ignore this exception to avoid triggering a larger impact on the cluster caused by ResourceManager failover. h1. Other We also need to discuss and optimize the same issues in safeCreate. -- This message was sent by Atlassian Jira (v8.20.10#820010) --------------------------------------------------------------------- To unsubscribe, e-mail: yarn-dev-unsubscr...@hadoop.apache.org For additional commands, e-mail: yarn-dev-h...@hadoop.apache.org