bitflicker64 commented on code in PR #2994:
URL: https://github.com/apache/hugegraph/pull/2994#discussion_r4113117226
##########
hugegraph-server/hugegraph-core/src/main/java/org/apache/hugegraph/traversal/optimize/TraversalUtil.java:
##########
@@ -645,14 +672,544 @@ private static boolean
extractHasContainers(HugeVertexStep<?> newStep,
private static boolean canExtractHasContainers(HugeGraph graph,
HasContainerHolder holder) {
- for (HasContainer has : holder.getHasContainers()) {
+ // Keep unsafe labels and their sibling properties for local filtering.
+ if (hasUnsafeLabelPredicate(holder)) {
+ return false;
+ }
+ List<HasContainer> hasContainers = holder.getHasContainers();
+ for (HasContainer has : hasContainers) {
if (!canExtractHasContainer(graph, has)) {
return false;
}
}
return true;
}
+ private static void prepareLocalHasContainers(
+ Step<?, ?> source, Traversal.Admin<?, ?> traversal) {
+ QueryHolder query = (QueryHolder) source;
+ Step<?, ?> step = source.getNextStep();
+ while (step instanceof HasStep || step instanceof NoOpBarrierStep) {
+ Step<?, ?> next = step.getNextStep();
+ if (step instanceof HasStep) {
+ HasContainerHolder holder = (HasContainerHolder) step;
+ for (HasContainer has : new
ArrayList<>(holder.getHasContainers())) {
+ // Paging is query metadata, never an element property
filter.
+ if (QueryHolder.SYSPROP_PAGE.equals(has.getKey())) {
+ query.addHasContainer(has);
+ holder.removeHasContainer(has);
+ continue;
+ }
+ if (T.id.getAccessor().equals(has.getKey())) {
+ // ID lookup is complete across labels. Existing source
+ // IDs and unsupported predicates still filter locally.
+ if (source instanceof HugeGraphStep &&
+ GraphStep.processHasContainerIds((HugeGraphStep<?,
?>) source, has)) {
+ holder.removeHasContainer(has);
+ continue;
+ }
+ holder.removeHasContainer(has);
+ holder.addHasContainer(new
LocalIdHasContainer(has.getPredicate()));
+ continue;
+ }
+ if (T.label.getAccessor().equals(has.getKey())) {
+ holder.removeHasContainer(has);
+ HasContainer label = new
LocalLabelHasContainer(has.getPredicate());
+ if (source instanceof HugeGraphStep &&
+ canPushPositiveLabel((HugeGraphStep<?, ?>) source,
has)) {
+ // A positive label conjunct uses the label index,
+ // independent of per-label property index
coverage.
+ // Keep the runtime label matcher for source-ID
queries.
+ query.addHasContainer(label);
+ } else {
+ // Keep unsupported candidates and adjacent-vertex
+ // labels local, including their paging boundary.
+ holder.addHasContainer(label);
+ }
+ continue;
+ }
+ if (keyForContainsKey(has.getKey()) ||
keyForContainsValue(has.getKey())) {
+ // Backend CONTAINS support varies and source IDs use
+ // local filtering too. Preserve HugeGraph's map query
+ // semantics without treating "key"/"value" as names.
+ holder.removeHasContainer(has);
+ traversal.addStep(traversal.getSteps().indexOf(step),
+ new LocalContainsStep<>(traversal,
has));
+ continue;
+ }
+ List<P<Object>> predicates = new ArrayList<>();
+ collectPredicates(predicates,
ImmutableList.of(has.getPredicate()));
+ if (predicates.stream().anyMatch(p ->
+ p.getBiPredicate() ==
Condition.RelationType.TEXT_CONTAINS)) {
+ // Child traversals can be unbound during optimization;
+ // the matcher resolves the graph from each runtime
element.
+ holder.removeHasContainer(has);
+ holder.addHasContainer(new LocalSearchHasContainer(
+ has.getKey(), has.getPredicate()));
+ }
+ }
+ if (holder.getHasContainers().isEmpty()) {
+ TraversalHelper.copyLabels(step, step.getPreviousStep(),
false);
+ traversal.removeStep(step);
+ }
+ }
+ step = next;
+ }
+ if (query.queryInfo().paging() && step instanceof RangeGlobalStep) {
+ // Bound the raw backend page even when local filters prevent
normal
+ // range extraction. Keep the range step to apply offset/limit
after
+ // those filters; a filtered page may contain fewer results.
+ query.setRange(0, ((RangeGlobalStep<?>) step).getHighRange());
+ }
+ }
+
+ private static boolean canPushPositiveLabel(HugeGraphStep<?, ?> source,
+ HasContainer has) {
+ if (!isEqInLabelPredicate(has)) {
+ return false;
+ }
+ HugeGraph graph = tryGetGraph(source);
+ if (graph == null) {
+ return false;
+ }
+ List<P<Object>> predicates = new ArrayList<>();
+ collectPredicates(predicates, ImmutableList.of(has.getPredicate()));
+ for (P<Object> predicate : predicates) {
+ Object value = predicate.getValue();
+ BiPredicate<?, ?> bp = predicate.getBiPredicate();
+ if (bp == Contains.within && !(value instanceof Collection)) {
+ return false;
+ }
+ Collection<?> values = bp == Contains.within ?
+ (Collection<?>) value :
Collections.singletonList(value);
+ for (Object candidate : values) {
+ if (!hasLabelIndex(graph, source.returnsVertex(), candidate)) {
+ return false;
+ }
+ }
+ }
+ return true;
+ }
+
+ private static boolean hasLabelIndex(HugeGraph graph, boolean vertex,
Object value) {
+ if (value instanceof Number) {
+ value = IdGenerator.of(((Number) value).longValue());
+ }
+ try {
+ SchemaLabel label;
+ if (value instanceof Id) {
+ Id id = (Id) value;
+ // Nonpositive IDs include internal schema objects that aren't
+ // user labels. Preserve their local matching behavior.
+ if (!id.number() || id.asLong() <= 0L) {
+ return false;
+ }
+ label = vertex ? graph.vertexLabel(id) : graph.edgeLabel(id);
+ } else if (value instanceof String) {
+ label = vertex ? graph.vertexLabel((String) value) :
graph.edgeLabel((String) value);
+ } else {
+ return false;
+ }
+ // Do not turn a working local filter into a missing-label/index
error.
+ return label != null && label.enableLabelIndex();
+ } catch (IllegalArgumentException | NotFoundException e) {
+ return false;
+ }
+ }
+
+ private static P<?> localIdPredicate(P<?> predicate) {
+ // Keep IDs local to preserve source-ID intersections and step
ordering.
+ // UUID/Element values need the same representation as
HugeElement.id(). Leave
+ // strings unchanged so HasContainer retains its string-ID comparison.
+ P<?> copy = predicate.clone();
+ List<P<Object>> leaves = new ArrayList<>();
+ collectPredicates(leaves, ImmutableList.of(copy));
+ for (P<Object> leaf : leaves) {
+ Object value = leaf.getValue();
+ if (value instanceof Collection) {
+ List<Object> values = new ArrayList<>();
+ for (Object item : (Collection<?>) value) {
+ values.add(localIdValue(item));
+ }
+ leaf.setValue(values);
+ } else {
+ leaf.setValue(localIdValue(value));
+ }
+ }
+ return copy;
+ }
+
+ private static Object localIdValue(Object value) {
+ if (value instanceof UUID || value instanceof Element) {
+ return HugeElement.getIdValue(HugeType.VERTEX, value);
+ }
+ return value;
+ }
+
+ private static final class LocalIdHasContainer extends HasContainer {
+
+ private static final long serialVersionUID = 1L;
+
+ private LocalIdHasContainer(P<?> predicate) {
+ super(T.id.getAccessor(), localIdPredicate(predicate));
+ }
+
+ @Override
+ protected boolean testId(Element element) {
+ return testLocalIdPredicate(this.getPredicate(), element.id());
+ }
+ }
+
+ @SuppressWarnings("unchecked")
+ private static boolean testLocalIdPredicate(P<?> predicate, Object id) {
+ if (predicate instanceof ConnectiveP) {
+ boolean and = predicate instanceof AndP;
+ for (P<?> child : ((ConnectiveP<?>) predicate).getPredicates()) {
+ if (testLocalIdPredicate(child, id) != and) {
+ return !and;
+ }
+ }
+ return and;
+ }
+ Object value = predicate.getValue();
+ Object first = value;
+ if (value instanceof Collection) {
+ Collection<?> values = (Collection<?>) value;
+ first = values.isEmpty() ? null : values.iterator().next();
+ }
+ // HasContainer decides string-ID comparison from only the top-level
+ // P value. ConnectiveP has no such value; decide separately per leaf.
+ Object actual = first instanceof String ? id.toString() : id;
+ return ((BiPredicate<Object, Object>)
predicate.getBiPredicate()).test(actual, value);
+ }
+
+ private static final class LocalLabelHasContainer extends HasContainer {
+
+ private static final long serialVersionUID = 1L;
+
+ private LocalLabelHasContainer(P<?> predicate) {
+ super(T.label.getAccessor(), predicate.clone());
+ }
+
+ @Override
+ protected boolean testLabel(Element element) {
+ // Resolve against the actual element, not a graph captured while
+ // strategies run. Child traversals may be unbound or later cloned.
+ return testLabelPredicate(this.getPredicate(), ((HugeElement)
element).schemaLabel());
+ }
+ }
+
+ private static final class LocalContainsStep<S extends Element> extends
HasStep<S> {
+
+ private static final long serialVersionUID = 1L;
+
+ private LocalContainsStep(Traversal.Admin<?, ?> traversal,
HasContainer has) {
+ super(traversal, has.clone());
+ E.checkArgument(has.getPredicate().getBiPredicate() == Compare.eq,
+ "CONTAINS query with relation '%s' is not
supported",
+ has.getPredicate().getBiPredicate());
+ }
+
+ @Override
+ protected boolean filter(Traverser.Admin<S> traverser) {
+ HugeElement element = (HugeElement) traverser.get();
+ // Adjacent vertices can be ID/label-only shells. Like
properties(),
+ // load their properties before evaluating the system property map.
+ element.getFilledProperties();
+ // Keep a HasStep boundary so count/range cannot bypass this
filter.
+ // Resolve schema from the runtime element; clone/reset must not
+ // retain a graph or transaction captured during optimization.
+ for (HasContainer has : this.getHasContainers()) {
+ boolean matches = keyForContainsKey(has.getKey()) ||
keyForContainsValue(has.getKey()) ?
+ convContains2Relation(element.graph(),
has).test(element) :
+ has.test(element);
+ if (!matches) {
+ return false;
+ }
+ }
+ return true;
+ }
+ }
+
+ @SuppressWarnings("unchecked")
+ private static boolean testLabelPredicate(P<?> predicate, SchemaLabel
label) {
+ if (predicate instanceof ConnectiveP) {
+ boolean and = predicate instanceof AndP;
+ for (P<?> child : ((ConnectiveP<?>) predicate).getPredicates()) {
+ if (testLabelPredicate(child, label) != and) {
+ return !and;
+ }
+ }
+ return and;
+ }
+ BiPredicate<?, ?> bp = predicate.getBiPredicate();
+ Object value = predicate.getValue();
+ if (bp == Contains.within || bp == Contains.without) {
+ for (Object item : (Collection<?>) value) {
+ if (testLabelValue(Compare.eq, label, item)) {
+ return bp == Contains.within;
+ }
+ }
+ return bp == Contains.without;
+ }
+ return testLabelValue((BiPredicate<Object, Object>) bp, label, value);
+ }
+
+ private static boolean testLabelValue(BiPredicate<Object, Object>
predicate,
+ SchemaLabel label, Object value) {
+ if (value instanceof Number) {
+ value = IdGenerator.of(((Number) value).longValue());
+ }
+ return predicate.test(value instanceof Id ? label.id() : label.name(),
value);
+ }
+
+ private static final class LocalSearchHasContainer extends HasContainer {
+
+ private static final long serialVersionUID = 1L;
+
+ private transient HugeGraph matcherGraph;
+ private transient P<?> matcherPredicate;
+ private transient Predicate<Object> matcher;
+
+ private LocalSearchHasContainer(String key, P<?> predicate) {
+ super(key, predicate.clone());
+ }
+
+ @Override
+ protected boolean testValue(Property property) {
+ // Keep the public P tree intact for strategies, hashing and Java
+ // serialization. Resolve the analyzer from the element's graph,
+ // including after cloning, deserialization or graph rebinding.
+ HugeGraph graph = (HugeGraph) property.element().graph();
+ if (this.matcherGraph != graph ||
+ !samePredicateValues(this.getPredicate(),
this.matcherPredicate)) {
+ P<?> predicate = this.getPredicate().clone();
+ this.matcher = localSearchMatcher(predicate, graph);
+ this.matcherPredicate = predicate;
+ this.matcherGraph = graph;
+ }
+ return this.matcher.test(property.value());
+ }
+
+ @Override
+ public LocalSearchHasContainer clone() {
+ LocalSearchHasContainer clone = (LocalSearchHasContainer)
super.clone();
+ clone.matcherGraph = null;
+ clone.matcherPredicate = null;
+ clone.matcher = null;
+ return clone;
+ }
+ }
+
+ private static boolean samePredicateValues(P<?> current, P<?> cached) {
+ // P.equals() compares originalValue, not the value changed by
setValue().
+ if (cached == null || current.getClass() != cached.getClass()) {
+ return false;
+ }
+ if (current instanceof ConnectiveP) {
+ List<? extends P<?>> children = ((ConnectiveP<?>)
current).getPredicates();
+ List<? extends P<?>> oldChildren = ((ConnectiveP<?>)
cached).getPredicates();
+ if (children.size() != oldChildren.size()) {
+ return false;
+ }
+ for (int i = 0; i < children.size(); i++) {
+ if (!samePredicateValues(children.get(i), oldChildren.get(i)))
{
+ return false;
+ }
+ }
+ return true;
+ }
+ return current.getBiPredicate().equals(cached.getBiPredicate()) &&
+ Objects.equals(current.getValue(), cached.getValue());
+ }
+
+ @SuppressWarnings("unchecked")
+ private static Predicate<Object> localSearchMatcher(P<?> predicate,
HugeGraph graph) {
+ if (predicate instanceof ConnectiveP) {
+ List<Predicate<Object>> children = new ArrayList<>();
+ for (P<?> child : ((ConnectiveP<?>) predicate).getPredicates()) {
+ children.add(localSearchMatcher(child, graph));
+ }
+ boolean and = predicate instanceof AndP;
+ return value -> {
+ for (Predicate<Object> child : children) {
+ if (child.test(value) != and) {
+ return !and;
+ }
+ }
+ return and;
+ };
+ }
+ if (predicate.getBiPredicate() !=
Condition.RelationType.TEXT_CONTAINS) {
+ return (P<Object>) predicate;
+ }
+ // Match SEARCH terms in place, preserving range and side-effect
ordering.
+ return graph.searchPredicate((String) predicate.getValue());
+ }
+
+ private static boolean hasUnsafeLabelInTraversal(
+ Traversal.Admin<?, ?> traversal, Step<?, ?> sourceStep) {
+ // Partial pushdown can lose candidates before local label filtering.
+ // Scan the remaining traversal, its children and each ancestor's
+ // remaining steps for filters on child output. Arbitrary extension
+ // steps don't reliably expose element identity, so stay conservative.
+ // FIXME(#3201): Restore selective pushdown when every candidate
schema label
+ // has compatible index coverage for extracted property predicates.
+ // Outside proven root and child suffixes, negative labels can disable
+ // property pushdown even across element changes (including ancestors
+ // and unknown extension steps), potentially requiring a full scan.
+ List<Step> steps = traversal.getSteps();
+ int start = 0;
+ while (start < steps.size() && steps.get(start) != sourceStep) {
+ start++;
+ }
+ start++;
+ for (int i = start; i < steps.size(); i++) {
+ Step<?, ?> step = steps.get(i);
+ if (changesCurrentElement(step) &&
+ traversal.getParent() instanceof EmptyStep &&
+ onlyCurrentElementSuffix(steps.subList(i, steps.size()))) {
+ return false;
+ }
+ if (step instanceof HasStep) {
+ HasContainerHolder holder = (HasContainerHolder) step;
+ if (hasUnsafeLabelPredicate(holder)) {
+ return true;
+ }
+ }
+ if (hasUnsafeLabelInChildren(step)) {
+ return true;
+ }
+ }
+ TraversalParent parent = traversal.getParent();
+ if (parent instanceof Step && !(parent instanceof EmptyStep)) {
+ Step<?, ?> parentStep = (Step<?, ?>) parent;
+ // RepeatStep's until/emit siblings can filter this child's output.
+ // This helper only descends, so revisiting the owning step's
children
+ // cannot recurse back into this ancestor walk.
+ if (hasUnsafeLabelInChildren(parentStep)) {
+ return true;
+ }
+ return hasUnsafeLabelInTraversal(parentStep.getTraversal(),
parentStep);
+ }
+ return false;
+ }
+
+ private static boolean changesCurrentElement(Step<?, ?> step) {
+ // VertexStep can return the source through a self-loop or a later hop,
+ // just as edge endpoints can. Only property traversal proves that a
+ // later label predicate cannot apply to the source vertex or edge.
+ return step instanceof PropertiesStep;
Review Comment:
Important: With `VertexStep` removed here, nothing after the source ends the
scan in `hasUnsafeLabelInTraversal()` or `hasUnsafeLabelInChildTraversal()`
except `properties()`. Any later negative label, on any hop, now moves every
source property predicate into local filtering. I asked for this on 559b220.
The probe below shows it does not fix what I claimed it would fix, so I was
wrong to ask.
The gap it closes is not caused by the negative label. The source step
`g.V().has("city","Beijing")` uses the `person` city index, so it never returns
a vertex whose label has no city index. A later label filter can only remove
elements. I ran a probe on the memory backend (JDK 11) with the
`testNegativeLabelOnSelfLoopKeepsSource` data (`loop-fan`, label `returnFan`,
city Beijing, self-loop) plus a `person` in Beijing:
| query | 83ef9f3 (base) | 03a0c67 |
|---|---|---|
| `has("city","Beijing")` | `[bj-person]` | `[bj-person]` |
| `has(...).out()` | `[]` | `[]` |
| `has(...).out().hasLabel("returnFan")` | `[]` | `[]` |
| `has(...).where(__.out())` | `[]` | `[]` |
| `has(...).out().hasLabel(P.neq("person"))` | `[]` | `[loop-fan]` |
At this head, adding `hasLabel(P.neq("person"))` after `out()` returns more
rows than `out()` alone. Result sets now depend on which filter follows the
hop, not on the data.
The cost falls on common queries. After `applyStrategies()` the start step
is:
- `has("city","Beijing").out().hasLabel(P.neq("person"))`:
`HugeGraphStep(Vertex,[city.eq(Beijing)])` at the base,
`HugeGraphStep(vertex,[])` here, so a full vertex scan.
- `has("city","Beijing").outE().hasLabel(P.neq("knows"))`:
`HugeGraphStep(vertex,[])` here. That label filters edges, which can never be
the source vertex, so even the self-loop argument does not apply.
- `has("unindexedProp","x").out().hasLabel(P.neq("person"))` and the
`outE()` form return `[]` after a scan instead of raising `NoIndexException`.
The test change at VertexCoreTest.java:10195-10200 records the same switch for
`where(__.out()...)` and `not(__.out()...)`.
Requested change: restore `VertexStep` in `changesCurrentElement()` as in
559b220, and restore the plan assertions this commit inverted
(`testVertexAdjacencyKeepsSourcePropertyLocal` and
`testChildVertexAdjacencyKeepsSourcePropertyLocal` in
TraversalUtilOptimizeTest, the `out()` cases added to
`testAdjacencyAndEdgeEndpointsKeepSourceLabelFallback`, and
VertexCoreTest.java:9472, 9603, 10139 and 10195-10200). Drop
`testNegativeLabelAfterAdjacentStepsReturnToSource` and
`testNegativeLabelOnSelfLoopKeepsSource`, or change them to assert the same
result as the matching query without the negative label. Leave source vertices
missing from a property query without full index coverage to #3201, which
covers every downstream filter and not only negative labels. If you keep the
conservative rule for `out().in()` style paths, at least treat `outE()`,
`inE()` and `bothE()` followed by the allowed suffix as leaving the source,
since an edge label cannot match a vertex.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
