@Override
public void processOp(Object row, int tag) throws HiveException {
try {
reportProgress();
// get alias
alias = (byte) tag;
if ((lastAlias == null) || (!lastAlias.equals(alias))) {
nextSz = joinEmitInterval;
}
ArrayList<Object> nr =
JoinUtil.computeValues(
row,
joinValues.get(alias),
joinValuesObjectInspectors.get(alias),
joinFilters.get(alias),
joinFilterObjectInspectors.get(alias),
noOuterJoin);
if (handleSkewJoin) {
skewJoinKeyContext.handleSkew(tag);
}
// number of rows for the key in the given table
int sz = storage.get(alias).size();
StructObjectInspector soi = (StructObjectInspector) inputObjInspectors[tag];
StructField sf = soi.getStructFieldRef(Utilities.ReduceField.KEY.toString());
Object keyObject = soi.getStructFieldData(row, sf);
// Are we consuming too much memory
if (alias == numAliases - 1 && !(handleSkewJoin && skewJoinKeyContext.currBigKeyTag >= 0)) {
if (sz == joinEmitInterval) {
// The input is sorted by alias, so if we are already in the last join
// operand,
// we can emit some results now.
// Note this has to be done before adding the current row to the
// storage,
// to preserve the correctness for outer joins.
checkAndGenObject();
storage.get(alias).clear();
}
} else {
if (sz == nextSz) {
// Output a warning if we reached at least 1000 rows for a join
// operand
// We won't output a warning for the last join operand since the size
// will never goes to joinEmitInterval.
LOG.warn("table " + alias + " has " + sz + " rows for join key " + keyObject);
nextSz = getNextSize(nextSz);
}
}
// Add the value to the vector
storage.get(alias).add(nr);
// if join-key is null, process each row in different group.
if (SerDeUtils.hasAnyNullObject(keyObject, sf.getFieldObjectInspector())) {
endGroup();
startGroup();
}
} catch (Exception e) {
e.printStackTrace();
throw new HiveException(e);
}
}
