You are viewing a plain text version of this content. The canonical link for it is here.
Posted to hdfs-issues@hadoop.apache.org by "Konstantin Shvachko (Jira)" <ji...@apache.org> on 2019/12/06 21:45:00 UTC

[jira] [Updated] (HDFS-15037) Encryption Zone operations should not block other RPC calls while retreivingencryption keys.

     [ https://issues.apache.org/jira/browse/HDFS-15037?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Konstantin Shvachko updated HDFS-15037:
---------------------------------------
    Description: 
I believe it was an intention to avoid blocking other operations while retrieving keys with holding {{FSDirectory.dirLock}}. But in reality all other operations enter first {{FSNamesystemLock}} then {{dirLock}}. So they are all blocked waiting for the key.
We see substantial increase in RPC wait time ({{RpcQueueTimeAvgTime}}) on NameNode when encryption operations are intermixed with regular workloads.

  was:
I believe it was an intention to avoid blocking other operations while retrieving keys with holding {{[FSDirectory.dirLock}}. But in reality all other operations enter first {{FSNamesystemLock}} then {{dirLock}}. So they are all blocked waiting for the key.
We see substantial increase in RPC wait time ({{RpcQueueTimeAvgTime}}) on NameNode when encryption operations are intermixed with regular workloads.


Here are the three methods, which hold only {{FSDirectory.dirLock}}, but not {{FSNamesystemLock}}:
* {{ReencryptionHandler.run()}}
* {{FSDirEncryptionZoneOp.getKeyNameForZone()}}
* {{EncryptionZoneManager.pauseForTestingAfterNthCheckpoint()}}

Looks to me the code need to be rearranged using some other lock for key retrieval.

> Encryption Zone operations should not block other RPC calls while retreivingencryption keys.
> --------------------------------------------------------------------------------------------
>
>                 Key: HDFS-15037
>                 URL: https://issues.apache.org/jira/browse/HDFS-15037
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: encryption, namenode
>    Affects Versions: 2.10.0
>            Reporter: Konstantin Shvachko
>            Priority: Major
>
> I believe it was an intention to avoid blocking other operations while retrieving keys with holding {{FSDirectory.dirLock}}. But in reality all other operations enter first {{FSNamesystemLock}} then {{dirLock}}. So they are all blocked waiting for the key.
> We see substantial increase in RPC wait time ({{RpcQueueTimeAvgTime}}) on NameNode when encryption operations are intermixed with regular workloads.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org