db-derby-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sunitha Kambhampati <ksunitha...@gmail.com>
Subject Re: [jira] Updated: (DERBY-500) Update/Select failure when BLOB/CLOB fields updated in several rows by PreparedStatement using setBinaryStream and setCharacterStream
Date Tue, 18 Oct 2005 18:23:24 GMT
Mike Matrigali wrote:

>I am looking at commiting this, one question:
>o in ReaderToUTF8Stream.java you changed close() to not do a
>reader.close() - why is this ok?
Thanks Mike  for reviewing this patch.

E.g.: So user application calls the following api to set the stream.
ps.setCharacterStream(i, mystream, pos)   //mystream is user 's stream.

Internally, ReaderToUTF8Stream is the stream that wraps the user's 
stream as LimitReader object.    Thus calling reader.close()  will close 
the user's stream and I think that is incorrect. I think it is the 
responsibility of the application to either close the stream or do 
whatever the application wishess and derby should not close the 'user's 
stream' internally.

E.g.  The user could have passed in a Resetable stream and then after 
the statement execution can reset his stream and use the same object to 
do some other work. If derby closes the user's stream internally, any 
other call on the stream would throw an exception and user is forced to 
create another stream object. .

Also previously (before this fix), ReaderToUTF8Stream.close() was never 
exercised,but now I believe it correctly does return resources back to 
the VM.


>Sunitha Kambhampati (JIRA) wrote:
>>     [ http://issues.apache.org/jira/browse/DERBY-500?page=all ]
>>Sunitha Kambhampati updated DERBY-500:
>>    Attachment: Derby500.diff.txt
>>Background :
>>In Derby, when a stream is set as a parameter value, the wrapper stream object used
for character data is ReaderToUTF8Stream 
>>and for binary data it is RawToBinaryFormatStream.Both these stream objects on read()
return data in a format that is used to store the respective datatype value. E.g in case of
char, the characters read from the user stream are converted using utf-8 derby specific encoding
and read calls return 
>>the data as expected by store layer. Beginning 2 bytes either have the utflen or has
zeroes, or if it is a long string, then the value is ended with the special marker 0xE0 ,
0x00, 0x00. For binary data, the stream data is prepended with 4 zeroes. 
>>once,the stream has been read fully and end of file reached, further read() returns
a -1.  If a stream is re-read, it returns a -1 which is incorrect data.  E.g.in the repro
for DERBY-500, the update statement has multiple rows that qualify and since the stream parameter
is used; the first row gets updated with the correct value and the stream is drained. For
the subsequent rows, the read from the stream parameter value returns -1 and thus is updated
with incorrect data.When retrieving the row back, the format of the fields is incorrect and
thus the exception. 
>>This patch
>>1. adds changes to RawToBinaryFormatStream and ReaderToUTF8Stream to throw an EOFException
if stream is re-read.
>>If a stream value has been fully read and end of file reached, any further reads on
the stream object  will result in an EOFException. This seems reasonable and more correct
than using incorrect values.  
>>Adds a new error message - 'Stream has already been read and end-of-file reached and
cannot be re-used.'
>>2. changes to RememberBytesInputStream to keep track of the stream state and not call
read on the stream objects once eof is reached.
>>3. Fix a bug in StoredPage.logColumn related to streams. In one particular scenario,
column was not being set to RememberBytesInputStream object and thus losing the data that
would be read from stream into RememberBytesInputStream.
>>4. adds testcases to store/streamingColumn.java and lang/forbitdata.java 
>>Also note
>>- This fix affects cases when a stream is re-used in which case an exception will
be thrown. 
>>So code that reads the stream once and materializes it will not be affected. E.g.
 Currently in case of char,varchar,long varchar, streams are materialized and this will work
fine as before.
>>Ran tests ok on jdk142/win2k (using classes directory)
>>svn stat
>>M      java\engine\org\apache\derby\impl\jdbc\RawToBinaryFormatStream.java
>>M      java\engine\org\apache\derby\impl\jdbc\ReaderToUTF8Stream.java
>>M      java\engine\org\apache\derby\impl\store\raw\data\RememberBytesInputStream.java
>>M      java\engine\org\apache\derby\impl\store\raw\data\StoredPage.java
>>M      java\engine\org\apache\derby\iapi\reference\SQLState.java
>>M      java\engine\org\apache\derby\loc\messages_en.properties
>>M      java\testing\org\apache\derbyTesting\functionTests\tests\lang\forbitdata.java
>>M      java\testing\org\apache\derbyTesting\functionTests\tests\store\streamingColumn.java
>>M      java\testing\org\apache\derbyTesting\functionTests\master\streamingColumn.out
>>M      java\testing\org\apache\derbyTesting\functionTests\master\forbitdata.out
>>I'll add clarifications to the paper - JDBCImplementation.html and attach it as another
patch to this jira entry.
>>Can someone please review it. Thanks.
>>>Update/Select failure when BLOB/CLOB fields updated in several rows by PreparedStatement
using setBinaryStream and setCharacterStream
>>>        Key: DERBY-500
>>>        URL: http://issues.apache.org/jira/browse/DERBY-500
>>>    Project: Derby
>>>       Type: Bug
>>> Components: JDBC
>>>   Versions:
>>>Environment: Windows 2000, java SDK 1.4
>>>   Reporter: Peter Kovgan
>>>   Assignee: Sunitha Kambhampati
>>>    Fix For:
>>>Attachments: Derby500.diff.txt, Derby500.stat.txt
>>>I have table contained BLOB and CLOB fields:
>>>Create table string is:
>>>private static final String CREATE = "CREATE TABLE ta (" +
>>>           "ta_id INTEGER NOT NULL," +
>>>           "mname VARCHAR( 254 ) NOT NULL," +
>>>           "mvalue INT NOT NULL," +
>>>           "mdate DATE NOT NULL," +
>>>           "bytedata BLOB NOT NULL," +
>>>           "chardata CLOB NOT NULL," +
>>>           "PRIMARY KEY ( ta_id ))";
>>>Then I insert 2000 rows in the table.
>>>Then I update all 2000 rows by command:
>>>private static final String UPDATE  =  "UPDATE ta " +
>>>   		"SET bytedata=? ,chardata=? " +
>>>   		"WHERE mvalue=?";
>>>/**create blob and clob arrays**/
>>>       int len1 = 10000;//for blob length data
>>>       int len2 = 15000;//for clob length data
>>>       byte buf [] = new byte[len1];
>>>       for(int i=0;i<len1;i++){
>>>       	buf [i] = (byte)45;
>>>       }
>>>       ByteArrayInputStream bais = new ByteArrayInputStream(buf);
>>>       char[] bufc = new char[len2];
>>>       for (int i = 0; i < bufc.length; i++) {
>>>       	bufc[i] = (char)'b';
>>>		}
>>>       CharArrayReader car = new CharArrayReader(bufc);
>>>PreparedStatement pstmt = connection.prepareStatement(UPDATE);
>>>pstmt.setBinaryStream(1,bais, len1);
>>>pstmt.setCharacterStream(2,car, len2);
>>>int updated =  pstmt.executeUpdate();
>>>System.out.printlen("updated ="+updated );
>>>all 2000 rows updated , because I receive output : updated =2000
>>>But If I run select (SELECT bytedata ,chardata  FROM ta)  after update, select
failed with error:
>>>ERROR XSDA7: Restore of a serializable or SQLData object of class , attempted
>>>read more data than was originally stored
>>>       at org.apache.derby.iapi.error.StandardException.newException(StandardEx
>>>       at org.apache.derby.impl.store.raw.data.StoredPage.readRecordFromArray(S
>>>       at org.apache.derby.impl.store.raw.data.StoredPage.restoreRecordFromSlot
>>>       at org.apache.derby.impl.store.raw.data.BasePage.fetchFromSlot(BasePage.
>>>       at org.apache.derby.impl.store.access.conglomerate.GenericScanController
>>>       at org.apache.derby.impl.store.access.heap.HeapScan.fetchNextGroup(HeapS
>>>       at org.apache.derby.impl.sql.execute.BulkTableScanResultSet.reloadArray(
>>>       at org.apache.derby.impl.sql.execute.BulkTableScanResultSet.getNextRowCo
>>>       at org.apache.derby.impl.sql.execute.NestedLoopJoinResultSet.getNextRowC
>>>       at org.apache.derby.impl.sql.execute.NestedLoopLeftOuterJoinResultSet.ge
>>>       at org.apache.derby.impl.sql.execute.ProjectRestrictResultSet.getNextRow
>>>       at org.apache.derby.impl.sql.execute.SortResultSet.getRowFromResultSet(S
>>>       at org.apache.derby.impl.sql.execute.SortResultSet.getNextRowFromRS(Sort
>>>       at org.apache.derby.impl.sql.execute.SortResultSet.loadSorter(SortResult
>>>       at org.apache.derby.impl.sql.execute.SortResultSet.openCore(SortResultSe
>>>       at org.apache.derby.impl.sql.execute.BasicNoPutResultSetImpl.open(BasicN
>>>       at org.apache.derby.impl.sql.GenericPreparedStatement.execute(GenericPre
>>>       at org.apache.derby.impl.jdbc.EmbedStatement.executeStatement(EmbedState
>>>       at org.apache.derby.impl.jdbc.EmbedPreparedStatement.executeStatement(Em
>>>       at org.apache.derby.impl.jdbc.EmbedPreparedStatement.execute(EmbedPrepar
>>>       at com.beep_beep.dbtest.complex.Benchmark.testSelect(Unknown Source)
>>>       at com.beep_beep.dbtest.complex.Benchmark.executeSimplestBigTable(Unknown
>>>       at com.beep_beep.dbtest.complex.Benchmark.testBigTable(Unknown Source)
>>>       at com.beep_beep.dbtest.complex.Benchmark.executeDegradationBenchmark(Unknown
>>>       at com.beep_beep.dbtest.complex.Benchmark.main(Unknown Source)
>>>>From the stack trace and from console I see that Update passed, but error
was raised in Select after Update.
>>>When I try the same update, but with difference(I changed WHERE clause, causing
update only 1 row):
>>>private static final String UPDATE  =  "UPDATE ta " +
>>>   		"SET bytedata=? ,chardata=? " +
>>>   		"WHERE mname=?";
>>>PreparedStatement pstmt = connection.prepareStatement(UPDATE);
>>>pstmt.setBinaryStream(1,bais, len1);
>>>pstmt.setCharacterStream(2,car, len2);
>>>int updated =  pstmt.executeUpdate();
>>>System.out.printlen("updated ="+updated );
>>>Only 1 row updated , because I receive output : updated =1
>>>In this case I have NO errors in select(the same as previous) .
>>>My assumption:
>>>It seems that Update receives ByteArrayInputStream and updates correctly only
1 row, then all rows updated by some
>>>incorrect value(may be because ByteArrayInputStream reached its end in first update),
causing select failure.
>>>I tested PointBase by the same test and PointBase passed this stage without errors,
no matter how many rows was updated.
>>>So I think it is a bug.
>>>Thank you.

View raw message