[CASSANDRA-716] bootstrapping does not work properly using multiple DataFileDirectory

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 0.5

Environment

storage-conf.xml:
<CommitLogDirectory>/home/store1/commitlog</CommitLogDirectory>
<DataFileDirectories>
<DataFileDirectory>/home/store0/data</DataFileDirectory>
<DataFileDirectory>/home/store1/data</DataFileDirectory>
</DataFileDirectories>
<CalloutLocation>/home/store1/cassandra/callouts</CalloutLocation>
<StagingFileDirectory>/home/store1/cassandra/staging</StagingFileDirectory>

• Severity: Normal
• Assignee: Gary Dusbabek
• Reporter: david.pan

Authors

Gary Dusbabek

• Created: 19/Jan/10 06:55
• Updated: 16/Apr/19 09:33
• Resolved: 21/Jan/10 23:13

Description

I was adding a new machine A which has 2 DataFileDirectories into the ring. The A will throw exception while bootstrapping.
DEBUG [MESSAGING-SERVICE-POOL:4] 2010-01-19 11:43:32,837 ContentStreamState.java (line 88) Removing stream context /home/store0/data/pic/raw_data-tmp-1-Data.db:209833142
WARN [MESSAGING-SERVICE-POOL:4] 2010-01-19 11:43:32,837 TcpConnection.java (line 484) Problem reading from socket connected to : java.nio.channels.SocketChannel[connected local
=/10.81.37.65:7000 remote=/10.81.42.26:10418]
WARN [MESSAGING-SERVICE-POOL:4] 2010-01-19 11:43:32,837 TcpConnection.java (line 485) Exception was generated at : 01/19/2010 11:43:32 on thread MESSAGING-SERVICE-POOL:4
java.io.IOException: rename failed of /home/store0/data/pic/raw_data-1-Filter.db
java.io.IOError: java.io.IOException: rename failed of /home/store0/data/pic/raw_data-1-Filter.db
at org.apache.cassandra.io.SSTableWriter.rename(SSTableWriter.java:154)
at org.apache.cassandra.io.SSTableWriter.renameAndOpen(SSTableWriter.java:162)
at org.apache.cassandra.io.Streaming$StreamCompletionHandler.onStreamCompletion(Streaming.java:284)
at org.apache.cassandra.net.io.ContentStreamState.handleStreamCompletion(ContentStreamState.java:108)
at org.apache.cassandra.net.io.ContentStreamState.read(ContentStreamState.java:90)
at org.apache.cassandra.net.io.TcpReader.read(TcpReader.java:96)
at org.apache.cassandra.net.TcpConnection$ReadWorkItem.run(TcpConnection.java:445)
at java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:885)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:907)
at java.lang.Thread.run(Thread.java:619)
Caused by: java.io.IOException: rename failed of /home/store0/data/pic/raw_data-1-Filter.db
at org.apache.cassandra.utils.FBUtilities.renameWithConfirm(FBUtilities.java:306)
at org.apache.cassandra.io.SSTableWriter.rename(SSTableWriter.java:150)
... 9 more

I traced the exception and maybe found the reason.
StreamInitiateVerbHandler::doVerb() will create 3 temporary files(index, filter, data) for each ssTable. The name for each file is generated by getNewFileNameFromOldContextAndNames(). This method will generate a file name and a path for each ssTable, but the path is generated with DatabaseDescriptor.getDataFileLocationForTable() which will return different path for ech call when we configure multi-DataFileDirectory.
eg: the ssTable raw_data-1 may have 3 temporary files :
/home/store0/data/pic/raw_data-tmp-1-Index.db
/home/store1/data/pic/raw_data-tmp-1-Filter.db
/home/store0/data/pic/raw_data-tmp-1-Data.db

After receiving all data, StreamCompletionHandler::onStreamCompletion() will rename all temporary files and this method think all ssTable files will have the same path as data.db file.
if (streamContext.getTargetFile().contains("-Data.db"))
{
......
try

{ SSTableReader sstable = SSTableWriter.renameAndOpen(streamContext.getTargetFile()); ...... }

......
}
Then the renameAndOpen() will throw that exception.