[
https://issues.apache.org/jira/browse/ZOOKEEPER-2249?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14699791#comment-14699791
]
Benjamin Jaton commented on ZOOKEEPER-2249:
-------------------------------------------
Something to consider is adding a way to bypass the CRC check to be able to
start ZK even if there is a transaction that failed, so that part of the rest
of the data can be recovered.
I'm thinking about a magical system property like -Dskip.crc.check=true
> CRC check failed when preAllocSize smaller than node data
> ---------------------------------------------------------
>
> Key: ZOOKEEPER-2249
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-2249
> Project: ZooKeeper
> Issue Type: Bug
> Reporter: Benjamin Jaton
> Labels: server
>
> Unexpected exception, exiting abnormally
> java.io.IOException: CRC check failed
> org.apache.zookeeper.server.persistence.FileTxnLog$FileTxnIterator.next(FileTxnLog.java:612)
>
> org.apache.zookeeper.server.persistence.FileTxnSnapLog.restore(FileTxnSnapLog.java:157)
>
> org.apache.zookeeper.server.ZKDatabase.loadDataBase(ZKDatabase.java:223)
> org.apache.zookeeper.server.ZooKeeperServer.loadData(ZooKeeperServer.java:272)
>
> org.apache.zookeeper.server.ZooKeeperServer.startdata(ZooKeeperServer.java:399)
> To reproduce, set the preAllocSize to 8MB, the jute.maxbuffer to 20MB and try
> saving a 15MB node several times.
> In my case the erroneous CRC appears after the second save. I use the
> LogFormatter class to detect it.
> I suspect that the CRC error happens when the new transaction log is created,
> the code probably expects to have enough room to save the transaction when
> creating a new file, but it's too small.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
