[ 
https://issues.apache.org/jira/browse/HBASE-5922?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13267200#comment-13267200
 ] 

Anoop Sam John commented on HBASE-5922:
---------------------------------------

After reading the code for HalfStoreFileReader and the javadoc for the 
HFileScanner.seekBefore(), got more confusion with the code...

{code}
/**
* Consider the key stream of all the keys in the file,
* <code>k[0] .. k[n]</code>, where there are n keys in the file.
* @param key Key to find
* @return false if key <= k[0] or true with scanner in position 'i' such
* that: k[i] < key.  Furthermore: there may be a k[i+1], such that
* k[i] < key <= k[i+1] but there may also NOT be a k[i+1], and next() will
* return false (EOF).
* @throws IOException
*/
public boolean seekBefore(byte [] key) throws IOException;
{code}

Here is the code for seekBefore in HalfStoreFileReader
For top half file
{code}
if (top) {
  if (getComparator().compare(key, offset, length, splitkey, 0,
          splitkey.length) < 0) {
        return false;
  }
}
{code}
The top file is from [splitkey,endkey). When the passed key is less than or 
equal to the splitkey, we wont be able to find a point i in the file such that 
k[i]<key..  This should be a case of return false. So here we should have if 
condition with <= 0 ?


Code for bottom file
{code}
else{
        if (getComparator().compare(key, offset, length, splitkey, 0,
          splitkey.length) >= 0) {
        return seekBefore(splitkey, 0, splitkey.length);
        }
}
{code}
The bottom file is from [startkey,splitkey). When the passed key is >= the 
split key we can point to the last key in the file. I think for this only 
seekBefore splitKey was done. But as it checks =0 also this will result in 
infinite calls to seekBefore  :( 

So may be we can change the condition from >=0 to >0 only...  Then the actual 
seekBefore will be done by the delegate.seekBefore() call...   But return false 
for getComparator().compare(key, offset, length, splitkey, 0, splitkey.length) 
>= 0  will be wrong I  feel.....

@Stack, @Nate What do u say?

Also this bug is applicable for other versions too...

                
> HalfStoreFileReader seekBefore causes StackOverflowError
> --------------------------------------------------------
>
>                 Key: HBASE-5922
>                 URL: https://issues.apache.org/jira/browse/HBASE-5922
>             Project: HBase
>          Issue Type: Bug
>          Components: client, io
>    Affects Versions: 0.90.0
>         Environment: HBase 0.90.4
>            Reporter: Nate Putnam
>            Assignee: Nate Putnam
>             Fix For: 0.90.0
>
>         Attachments: HBASE-5922.patch, HBASE-5922.patch
>
>
> Calling HRegionServer.getClosestRowBefore() can cause a stack overflow if the 
> underlying store file is a reference and the row key is in the bottom.
> java.io.IOException: java.io.IOException: java.lang.StackOverflowError
>         at 
> org.apache.hadoop.hbase.regionserver.HRegionServer.convertThrowableToIOE(HRegionServer.java:990)
>         at 
> org.apache.hadoop.hbase.regionserver.HRegionServer.convertThrowableToIOE(HRegionServer.java:978)
>         at 
> org.apache.hadoop.hbase.regionserver.HRegionServer.getClosestRowBefore(HRegionServer.java:1651)
>         at sun.reflect.GeneratedMethodAccessor174.invoke(Unknown Source)
>         at 
> sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
>         at java.lang.reflect.Method.invoke(Method.java:597)
>         at org.apache.hadoop.hbase.ipc.HBaseRPC$Server.call(HBaseRPC.java:570)
>         at 
> org.apache.hadoop.hbase.ipc.HBaseServer$Handler.run(HBaseServer.java:1039)
> Caused by: java.lang.StackOverflowError
>         at 
> org.apache.hadoop.hbase.io.HalfStoreFileReader$1.seekBefore(HalfStoreFileReader.java:147)
>         at 
> org.apache.hadoop.hbase.io.HalfStoreFileReader$1.seekBefore(HalfStoreFileReader.java:149)
>         at 
> org.apache.hadoop.hbase.io.HalfStoreFileReader$1.seekBefore(HalfStoreFileReader.java:149)
>         at 
> org.apache.hadoop.hbase.io.HalfStoreFileReader$1.seekBefore(HalfStoreFileReader.java:149)
>         at 
> org.apache.hadoop.hbase.io.HalfStoreFileReader$1.seekBefore(HalfStoreFileReader.java:149)

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Reply via email to