一、問題現象
某一個nodemanager退出後，導致 application master中出現大量的如下日誌，並且持續很長時間，application master才成功退出。
2016-06-24 09:32:35,596 INFO [ContainerLauncher #3] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 1 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:35,596 INFO [ContainerLauncher #9] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 1 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:35,597 INFO [ContainerLauncher #7] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 1 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:36,455 INFO [ContainerLauncher #8] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 7 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:36,539 INFO [ContainerLauncher #5] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 4 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:36,539 INFO [ContainerLauncher #1] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 4 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:36,539 INFO [ContainerLauncher #6] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 3 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:36,539 INFO [ContainerLauncher #2] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 3 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:36,539 INFO [ContainerLauncher #0] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 4 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:36,596 INFO [ContainerLauncher #4] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 2 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:36,597 INFO [ContainerLauncher #3] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 2 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 09:32:36,597 INFO [ContainerLauncher #9] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 2 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) ....... 2016-06-24 12:57:52,328 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 8 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:57:53,339 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 9 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:58:04,357 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 0 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:58:05,367 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 1 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:58:06,378 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 2 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:58:07,392 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 3 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:58:08,399 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 4 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:58:09,408 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 5 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:58:10,417 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 6 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:58:11,425 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 7 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 2016-06-24 12:58:12,434 INFO [Thread-1835] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 8 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS)
1）dchadoop206上的nodemanager退出後（由於重啓），導致application master持續的去連接之前nodemanager上的container。顯然這些container是已經連接不上了。
2）最終經過非常長的時間大概3-4小時後，連接不上的異常才拋出，application master正常結束。
二、問題分析
這個問題主要涉及hadoop的rpc機制。首先看下面兩個配置參數
#定義client連接到nodemanager的最大超時時間,不是單次連接，而是經過多少時間連接不上nodemanager，則認爲操作失敗 <property> <name>yarn.client.nodemanager-connect.max-wait-ms</name> <value>15*60*1000</value> </property> # 定義每次嘗試去連接nodemanager的時間間隔 <property> <name>yarn.client.nodemanager-connect.retry-interval-ms</name> <value>10*1000</value> </property>
根據這兩個參數的定義，ApplicationMaster經過15分鐘仍然連不上nodemanager的container，會取消try connect。但觀察的情況是Application Master 需要等大約30分鐘，才取消try connect。主要原因在於hadoop 的rpc機制如下，首先ApplicationMaster 會根據上面的兩個參數，構造一個RetryUpToMaximumCountWithFixedSleep的重連策略，這個重連策略會通過以下方式計算
MaximumCount：yarn.client.nodemanager-connect.max-wait-ms/yarn.client.nodemanager-connect.retry-interval-ms=90次
而每次的RPC請求中，Client也有自己的重連策略，就是類似這樣的東東:
2016-06-24 09:32:36,455 INFO [ContainerLauncher #8] org.apache.hadoop.ipc.Client: Retrying connect to server: dchadoop206/192.168.1.199:32951. Already tried 7 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS) 由兩個rpc參數控制，ipc.client.connect.max.retries=10和ipc.client.connect.retry.interval=1000ms控制
所以最終ApplicationMaster 放棄try connect的等待時間是：90*(10+10)=1800s
三、解決辦法
1）在提交map-reduce/hive sql/hive server2等客戶端機器修改yarn-site.xml的以下參數
2）hadoop命令行中通過-D設置該參數
<property> <name>yarn.client.nodemanager-connect.max-wait-ms</name> <value>180000</value> </property>
這樣總的等待時間就是6分鐘。
注意事項
這個修改是不需要做任何重啓yarn組件操作的，是一個客戶端相關的操作！