Hi All,
I was looking into Derby-2017
(https://issues.apache.org/jira/browse/DERBY-2017) the following were my
findings :
- when error occurs, then a dummyDSS is built and the partial data
(which is *'this' *in the repro ) is sent across to NetworkServer. Now
when the next select statement is executed, it commits the transaction
and the partial data gets committed and select returns the incorrect
committed data.
- I modified the repro to rollback the transaction when error occurs,
then it behaves correctly.
code snippet....................
StringReader reader = new StringReader("this string is way too
long");
ps.setCharacterStream(1, reader, 5);
try {
ps.executeUpdate();
} catch (SQLException e) {
*c.rollback(); //see here*
System.out.println(e);
}
---------------------------------------------------
- DRDA Header specifications provides a flag which can notify that error
had occurred & do not continue.
/**
* Read DSS header
* DSS Header format is
* 2 bytes - length
* 1 byte - 'D0' - indicates DDM data
* 1 byte - DSS format
* |---|---------|----------|
* | 0 | flags | type |
* |---|---------|----------|
* | 0 | 1 2 3 | 4 5 6 7 |
* |---|---------|----------|
* bit 0 - '0'
* bit 1 - '0' - unchained, '1' - chained
* *bit 2 - '0' - do not continue on error, '1' - continue
on error //See Here*
* bit 3 - '0' - next DSS has different correlator, '1' -
next DSS has
* same correlator
..........
*/
- What I observed is, when error occurs (i.e. accumulateReadError) and
writeCommitSubstitute is called where a dummy DSS is built
(buildDummyEXCSAT()).
- buildDss() method write the header information and here we should set
the CONTINUE_ON_ERROR flag to '0' thus the data shold not be send to
NetworkServer.
Is my understanding correct? I am new to this part of derby and not
familiar with the DRDA implementation.
Comments/Suggestions please.
I tried to follow Army's work on DERBY-35, but that is much target for
chaining and I am not aware about Tomohito 's work in Layer B streaming.
Can anyone help me out in this.
Thanks,
Saurabh
