HDFS緩存數據命令

• 查看緩存池信息

hdfs cacheadmin -listPools -stats

• 查看已緩存的數據信息

hdfs cacheadmin -listDirectives -stats

• Impala表卸載緩存數據

alter table dw_crawler.bsl_zhongda_weibo_article_hive set uncached;

• 創建緩存池

hdfs?cacheadmin?-addPool?article_pool2?-owner?impala

• 顯示表狀態

show table stats bsl_zhongda_weibo_article_hive;

• 將表加入緩存

alter table dw_crawler.bsl_zhongda_weibo_article_hive set cached in 'article_pool3';

• 指定分區將表加入緩存

alter table dw_crawler.bsl_zhongda_weibo_article_hive partition(pt_created_date=20180101) set cached in 'article_pool3';

性能測試結果：

是否緩存	數據條數	數據量	處理時間(并發1)	處理時間(并發3)	處理時間(并發5)
是	-	116.5G	8s	20s	29s
否	-	116.5G	68.17	136	240
是	-	72.7G	8.38	20.6	30.3
否	-	72.7G	80s	165s	235s

?

異常處理

1.hdfs緩存池空間不足

?

[fwqzx002.zh:21000] ods_crawler> alter table xxx partition (pt_created_date='201811') set cached in 'article_pool'; Query: alter table xxx partition (pt_created_date='201811') set cached in 'article_pool' ERROR: ImpalaRuntimeException: Caching path /user/hive/warehouse/xxxx/pt_created_date=201811 of size 24274436556 bytes at replication 1 would exceed pool article_pool's remaining capacity of 20450868109 bytes.at org.apache.hadoop.hdfs.server.namenode.CacheManager.checkLimit(CacheManager.java:405)at org.apache.hadoop.hdfs.server.namenode.CacheManager.addDirective(CacheManager.java:531)at org.apache.hadoop.hdfs.server.namenode.FSNDNCacheOp.addCacheDirective(FSNDNCacheOp.java:45)at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.addCacheDirective(FSNamesystem.java:6782)at org.apache.hadoop.hdfs.server.namenode.NameNodeRpcServer.addCacheDirective(NameNodeRpcServer.java:1883)at org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolServerSideTranslatorPB.addCacheDirective(ClientNamenodeProtocolServerSideTranslatorPB.java:1265)at org.apache.hadoop.hdfs.protocol.proto.ClientNamenodeProtocolProtos$ClientNamenodeProtocol$2.callBlockingMethod(ClientNamenodeProtocolProtos.java)at org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:523)at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:991)at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:869)at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:815)at java.security.AccessController.doPrivileged(Native Method)at javax.security.auth.Subject.doAs(Subject.java:422)at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1685)at org.apache.hadoop.ipc.Server$Handler.run(Server.java:2675)CAUSED BY: InvalidRequestException: Caching path /user/hive/warehouse/xxxx/pt_created_date=201811 of size 24274436556 bytes at replication 1 would exceed pool article_pool's remaining capacity of 20450868109 bytes.at org.apache.hadoop.hdfs.server.namenode.CacheManager.checkLimit(CacheManager.java:405)at org.apache.hadoop.hdfs.server.namenode.CacheManager.addDirective(CacheManager.java:531)at org.apache.hadoop.hdfs.server.namenode.FSNDNCacheOp.addCacheDirective(FSNDNCacheOp.java:45)at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.addCacheDirective(FSNamesystem.java:6782)at org.apache.hadoop.hdfs.server.namenode.NameNodeRpcServer.addCacheDirective(NameNodeRpcServer.java:1883)at org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolServerSideTranslatorPB.addCacheDirective(ClientNamenodeProtocolServerSideTranslatorPB.java:1265)at org.apache.hadoop.hdfs.protocol.proto.ClientNamenodeProtocolProtos$ClientNamenodeProtocol$2.callBlockingMethod(ClientNamenodeProtocolProtos.java)at org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:523)at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:991)at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:869)at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:815)at java.security.AccessController.doPrivileged(Native Method)at javax.security.auth.Subject.doAs(Subject.java:422)at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1685)at org.apache.hadoop.ipc.Server$Handler.run(Server.java:2675)CAUSED BY: RemoteException: Caching path /user/hive/warehouse/xxxx/pt_created_date=201811 of size 24274436556 bytes at replication 1 would exceed pool article_pool's remaining capacity of 20450868109 bytes.at org.apache.hadoop.hdfs.server.namenode.CacheManager.checkLimit(CacheManager.java:405)at org.apache.hadoop.hdfs.server.namenode.CacheManager.addDirective(CacheManager.java:531)at org.apache.hadoop.hdfs.server.namenode.FSNDNCacheOp.addCacheDirective(FSNDNCacheOp.java:45)at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.addCacheDirective(FSNamesystem.java:6782)at org.apache.hadoop.hdfs.server.namenode.NameNodeRpcServer.addCacheDirective(NameNodeRpcServer.java:1883)at org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolServerSideTranslatorPB.addCacheDirective(ClientNamenodeProtocolServerSideTranslatorPB.java:1265)at org.apache.hadoop.hdfs.protocol.proto.ClientNamenodeProtocolProtos$ClientNamenodeProtocol$2.callBlockingMethod(ClientNamenodeProtocolProtos.java)at org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:523)at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:991)at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:869)at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:815)at java.security.AccessController.doPrivileged(Native Method)at javax.security.auth.Subject.doAs(Subject.java:422)at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1685)at org.apache.hadoop.ipc.Server$Handler.run(Server.java:2675)

原因是創建時指定-limit 40000000000，即大小為40000000000 字節 ≈ 37.25GB，而我緩存的數據是18.21(已成功緩存分區)+22.61(緩存異常分區) 大于

之前創建緩存池命令

hdfs cacheadmin -addPool article_pool -owner impala -limit 40000000000

?解決方法：修改緩存池大小為自己可用值，默認為不限制

2.HDFS未全部緩存分區數據

明顯看出數據Size 大小與BytesCached大小不相同

[fwqzx002.zh:21000] ods_crawler> show table stats tableName ; Query: show table stats tableName +-----------------+-------+--------+---------+--------------+-------------------+---------+-------------------+-------------------------------------------------------------------------------------------------------------+ | pt_created_date | #Rows | #Files | Size | Bytes Cached | Cache Replication | Format | Incremental stats | Location | +-----------------+-------+--------+---------+--------------+-------------------+---------+-------------------+-------------------------------------------------------------------------------------------------------------+ | 201810 | -1 | 75 | 18.21GB | 5.36GB | 1 | PARQUET | false | hdfs://nameservice1/user/hive/warehouse/tableName /pt_created_date=201810 | | 201811 | -1 | 94 | 22.61GB | 55.35MB | 1 | PARQUET | false | hdfs://nameservice1/user/hive/warehouse/tableName /pt_created_date=201811 | | 201812 | -1 | 141 | 33.70GB | 33.22GB | 1 | PARQUET | false | hdfs://nameservice1/user/hive/warehouse/tableName /pt_created_date=201812 | | Total | -1 | 310 | 74.51GB | 38.63GB | | | | | +-----------------+-------+--------+---------+--------------+-------------------+---------+-------------------+-------------------------------------------------------------------------------------------------------------+

?查看hdfs，發現需要緩存大小（BYTES_NEEDED）與實際緩存（BYTES_CACHED）的大小不相同

[hdfs@fwqzx002 root]$ hdfs cacheadmin -listDirectives -stats Found 4 entriesID POOL REPL EXPIRY PATH BYTES_NEEDED BYTES_CACHED FILES_NEEDED FILES_CACHED20 article_pool3 1 never /user/hive/warehouse/tableName/pt_created_date=201812 36183104282 35666895768 141 13921 article_pool3 1 never /user/hive/warehouse/tableName 0 0 0 022 article_pool3 1 never /user/hive/warehouse/tableName/pt_created_date=201810 19549131891 5751122434 75 2223 article_pool3 1 never /user/hive/warehouse/tableName/pt_created_date=201811 24274436556 58042919 94 1

實際操作中未報異常，猜測HDFS緩存達到上限，去查看下HDFS配置

查看HDFS “dfs.datanode.max.locked.memory”參數，發現為4G，集群中有10個DataNode節點，一共加起來最多緩存40G,上面實際緩存大小≈38G左右已基本達到上限（因為數據并非絕對的平均存儲，可能部分節點數據超過4G就會達到緩存上限）

?

嘗試修改“dfs.datanode.max.locked.memory”為50G（根據個人服務器適當調整），結果全部緩存成功

?

• 注：show table stats bsl_zhongda_weibo_article_hive 與 hdfs cacheadmin -listDirectives -stats 區別在于，show table stats bsl_zhongda_weibo_article_hive 未完成Bytes Cached顯示為-1 ；后者是計算分配的空間，即后者非實際緩存完。

?

?

?

?

?

?

?

?

?

總結

以上是生活随笔為你收集整理的Impala查询 - HDFS缓存数据的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。