PS:
The problem is either there in BinaryEvalCtx, or in class
public class Int4Datum extends NumericDatum.
in the method
public int compareTo(Datum datum) {
...
@Override
127 public int compareTo(Datum datum) {
128 switch (datum.type()) {
129 case INT2:
130 if (val < datum.asInt2()) {
131 return -1;
132 } else if (datum.asInt2() < val) {
133 return 1;
134 } else {
135 return 0;
136 }
137 case INT4:
138 if (val < datum.asInt4()) {
139 return -1;
140 } else if (datum.asInt4() < val) {
141 return 1;
142 } else {
143 return 0;
144 }
145 case INT8:
146 if (val < datum.asInt8()) {
147 return -1;
148 } else if (datum.asInt8() < val) {
149 return 1;
150 } else {
151 return 0;
152 }
153 case FLOAT4:
154 if (val < datum.asFloat4()) {
155 return -1;
156 } else if (datum.asFloat4() < val) {
157 return 1;
158 } else {
159 return 0;
160 }
161 case FLOAT8:
162 if (val < datum.asFloat8()) {
163 return -1;
164 } else if (datum.asFloat8() < val) {
165 return 1;
166 } else {
167 return 0;
168 }
169 default:
170 throw new InvalidOperationException(datum.type());
171 }
172 }
173
Just a hint, maybe it helps You find and solve the issue more quickly.
Camelia
________________________________
From: camelia c <[email protected]>
To: "[email protected]" <[email protected]>
Sent: Wednesday, September 11, 2013 10:18 PM
Subject: Re: [GSoc2013] - Outer Join -Possible issue about BinaryEvalCtx
Hello,
I would like to kindly ask You to take a moment and explain to me how does this
BinaryEvalCtx handle NULL values in fields. From an error I got recently it
seems that when it reads NULL values and it has to compare them to non-NULL
values, it breaks.
The portion of code leading to error in is the same as in the original
NLJoinExec and other join algorithms, where TAJO needs to verify the join
condition and then project according to the output schema:
if (joinQual != null) {
joinQual.eval(qualCtx, inSchema, frameTuple);
if (joinQual.terminate(qualCtx).asBool()) {
Here the verification of the join condition fails if one operand is a NULL
value and the other is a number, in this case number 10.
13/09/11 21:37:01 INFO physical.LeftOuter_NLJoinExec: ********leftChild.next()
=(0=>10)
13/09/11 21:37:01 INFO physical.LeftOuter_NLJoinExec: ********rightChild.next()
=(0=>333.0, 1=>10)
13/09/11 21:37:01 INFO physical.LeftOuter_NLJoinExec: ******** a result matched
padded tuple =(0=>10, 1=>333.0)
13/09/11 21:37:01 INFO physical.LeftOuter_NLJoinExec: ********rightChild.next()
=(0=>555.0, 1=>10)
13/09/11 21:37:01 INFO physical.LeftOuter_NLJoinExec: ******** a result matched
padded tuple =(0=>10, 1=>555.0)
13/09/11 21:37:01 INFO physical.LeftOuter_NLJoinExec: ********rightChild.next()
=(0=>777.0, 1=>NULL)
13/09/11 21:37:01 ERROR worker.Task:
org.apache.tajo.datum.exception.InvalidOperationException
at org.apache.tajo.datum.Int4Datum.equalsTo(Int4Datum.java:122)
at org.apache.tajo.engine.eval.BinaryEval.terminate(BinaryEval.java:158)
at
org.apache.tajo.engine.planner.physical.LeftOuter_NLJoinExec.next(LeftOuter_NLJoinExec.java:157)
at
org.apache.tajo.engine.planner.physical.StoreTableExec.next(StoreTableExec.java:85)
at org.apache.tajo.worker.Task.run(Task.java:378)
at org.apache.tajo.worker.TaskRunner$2.run(TaskRunner.java:359)
at java.lang.Thread.run(Thread.java:662)
It seems that at this moment, it takes the NULL as being Int4Datum, instead of
NullDatum.
These classes exist beyond outer join nodes' scope but maybe they weren't
tested for NULL values before. Now, some of the outer join's processing relies
on these classes' correct behavior, that is why I'm trying to signal when I
find errors.
Thank You in advance for Your answer and for the understanding!
Yours sincerely,
Camelia
