use of datawave.query.index.lookup.IndexMatch in project datawave by NationalSecurityAgency.
the class AncestorUidIntersector method reduce.
private Set<IndexMatch> reduce(Set<IndexMatch> matches, IndexMatch currentMatch) {
Set<IndexMatch> result = Sets.newHashSet();
boolean conflict = false;
for (IndexMatch match : matches) {
if (!match.getUid().startsWith(currentMatch.getUid() + UIDConstants.DEFAULT_SEPARATOR) || match.getUid().equals(currentMatch.getUid())) {
result.add(match);
}
if (currentMatch.getUid().startsWith(match.getUid() + UIDConstants.DEFAULT_SEPARATOR) || match.getUid().equals(currentMatch.getUid())) {
conflict = true;
}
}
if (!conflict) {
result.add(currentMatch);
}
return result;
}
use of datawave.query.index.lookup.IndexMatch in project datawave by NationalSecurityAgency.
the class AncestorUidIntersector method intersect.
@Override
public Set<IndexMatch> intersect(Set<IndexMatch> uids1, Set<IndexMatch> uids2, List<JexlNode> delayedNodes) {
/*
* C) Both are small, so we have an easy case where we can prune much of this sub query. Must propagate delayed nodes, though.
*/
// create a map of correlated UIDS mapped to the root uid. The values keep the two lists of uids separate
Map<String, Tuple2<ArrayList<IndexMatch>, ArrayList<IndexMatch>>> correlatedUids = new HashMap<>();
// put the first set of uids in the correlated list
for (IndexMatch match1 : uids1) {
String baseUid = TLD.parseRootPointerFromId(match1.getUid());
Tuple2<ArrayList<IndexMatch>, ArrayList<IndexMatch>> indexMatchLists = correlatedUids.get(baseUid);
if (indexMatchLists == null) {
indexMatchLists = new Tuple2<>(new ArrayList<>(), new ArrayList<>());
correlatedUids.put(baseUid, indexMatchLists);
}
indexMatchLists.first().add(match1);
}
// put the second set of uids in the correlated list
for (IndexMatch match2 : uids2) {
String baseUid = TLD.parseRootPointerFromId(match2.getUid());
Tuple2<ArrayList<IndexMatch>, ArrayList<IndexMatch>> indexMatchLists = correlatedUids.get(baseUid);
if (indexMatchLists == null) {
indexMatchLists = new Tuple2<>(new ArrayList<>(), new ArrayList<>());
correlatedUids.put(baseUid, indexMatchLists);
}
indexMatchLists.second().add(match2);
}
// now for each base uid, if we have uids in the two lists then remap them to the descendent furthest from the root
Set<IndexMatch> matches = new HashSet<>();
for (Tuple2<ArrayList<IndexMatch>, ArrayList<IndexMatch>> indexMatchLists : correlatedUids.values()) {
if (!indexMatchLists.first().isEmpty() && !indexMatchLists.second().isEmpty()) {
for (IndexMatch uid1 : indexMatchLists.first()) {
for (IndexMatch uid2 : indexMatchLists.second()) {
// if uid1 starts with uid2, then uid1 is a descendent of uid2
if (uid1.getUid().startsWith(uid2.getUid() + UIDConstants.DEFAULT_SEPARATOR) || uid1.getUid().equals(uid2.getUid())) {
JexlNodeSet nodeSet = new JexlNodeSet();
nodeSet.add(uid1.getNode());
nodeSet.add(uid2.getNode());
nodeSet.addAll(delayedNodes);
IndexMatch currentMatch = new IndexMatch(Sets.newHashSet(nodeSet.getNodes()), uid1.getUid(), IndexMatchType.AND);
matches = reduce(matches, currentMatch);
} else // if uid2 starts with uid1, then uid2 is a descendent of uid1
if (uid2.getUid().startsWith(uid1.getUid() + UIDConstants.DEFAULT_SEPARATOR)) {
JexlNodeSet nodeSet = new JexlNodeSet();
nodeSet.add(uid1.getNode());
nodeSet.add(uid2.getNode());
nodeSet.addAll(delayedNodes);
IndexMatch currentMatch = new IndexMatch(Sets.newHashSet(nodeSet), uid2.getUid(), IndexMatchType.AND);
matches = reduce(matches, currentMatch);
}
}
}
}
}
return matches;
}
use of datawave.query.index.lookup.IndexMatch in project datawave by NationalSecurityAgency.
the class AncestorUidIntersectorTest method testMultipleBranchesCommonAncestor.
@Test
public void testMultipleBranchesCommonAncestor() {
uids2.add(new IndexMatch("a.b.c", node1));
uids2.add(new IndexMatch("a.b.c.2", node1));
uids2.add(new IndexMatch("a.b.c.1", node1));
uids1.add(new IndexMatch("a.b.c.1", node2));
uids1.add(new IndexMatch("a.b.c.2.1", node2));
uids1.add(new IndexMatch("a.b.c", node2));
Set<IndexMatch> result = intersector.intersect(uids1, uids2, Collections.EMPTY_LIST);
Assert.assertNotNull(result);
Assert.assertEquals("expected size 1, got " + result.size(), 1, result.size());
Assert.assertEquals("a.b.c", result.iterator().next().getUid());
}
use of datawave.query.index.lookup.IndexMatch in project datawave by NationalSecurityAgency.
the class AncestorUidIntersectorTest method testReduceOverlapUids1.
@Test
public void testReduceOverlapUids1() {
uids1.add(new IndexMatch("a.b.c.1", node1));
uids2.add(new IndexMatch("a.b.c.1.1", node2));
uids2.add(new IndexMatch("a.b.c.1.1.2", node2));
Set<IndexMatch> result = intersector.intersect(uids1, uids2, Collections.EMPTY_LIST);
Assert.assertNotNull(result);
Assert.assertEquals("expected size 1, got " + result.size(), 1, result.size());
Assert.assertEquals("a.b.c.1.1", result.iterator().next().getUid());
}
use of datawave.query.index.lookup.IndexMatch in project datawave by NationalSecurityAgency.
the class AncestorUidIntersectorTest method testReduceSingleUids2.
@Test
public void testReduceSingleUids2() {
uids2.add(new IndexMatch("a.b.c.1", node1));
uids2.add(new IndexMatch("a.b.c.1.1", node1));
uids1.add(new IndexMatch("a.b.c.1.1.1", node2));
uids1.add(new IndexMatch("a.b.c.1.1.2", node2));
Set<IndexMatch> result = intersector.intersect(uids1, uids2, Collections.EMPTY_LIST);
Assert.assertNotNull(result);
Assert.assertEquals("expected size 2, got " + result.size(), 2, result.size());
}
Aggregations