What index is needed to answer a disjunct (__.or) query?

Question

I want to find a set of User vertices that match a given search string. These vertices have four String properties to match against - FIRST_NAME, LAST_NAME, DISPLAY_NAME and EMAIL. My query is constructed like so:

Query

GraphTraversal<Vertex, Vertex> query = GraphFactory.getDefault().traversal().V()
    .hasLabel(VertexLabel.USER.name())
    .or(    __.has(PropertyKey.EMAIL_LC.name(), Text.textRegex(regex)),
            __.has(PropertyKey.DISPLAY_NAME_LC.name(), Text.textRegex(regex)),
            __.has(PropertyKey.FIRST_NAME.name(), Text.textRegex(regex)),
            __.has(PropertyKey.LAST_NAME.name(), Text.textRegex(regex)));

To answer it, there is one main, MIXED index in place (commented "user search"), amongst others to ensure uniqueness / match against a single property.

Index

//Unique Email Addresses - COMPOSITE
mgmt.buildIndex("byEmailComp", Vertex.class)
    .addKey(emailLowercaseKey)
    .indexOnly(userLabel)
    .unique()
    .buildCompositeIndex();

//Unique Email Addresses - MIXED
mgmt.buildIndex("byEmailMixed", Vertex.class)
    .addKey(emailLowercaseKey, Mapping.TEXTSTRING.asParameter())
    .indexOnly(userLabel)
    .buildMixedIndex("search");

//Unique Display Name - COMPOSITE
mgmt.buildIndex("byDisplayNameComp", Vertex.class)
    .addKey(displayNameLowercaseKey)
    .indexOnly(userLabel)
    .unique()
    .buildCompositeIndex();

//Unique Display Name - MIXED
mgmt.buildIndex("byDisplayNameMixed", Vertex.class)
    .indexOnly(userLabel)
    .addKey(displayNameLowercaseKey, Mapping.TEXTSTRING.asParameter())
    .buildMixedIndex("search");

//User search - MIXED
mgmt.buildIndex("userSearch", Vertex.class)
    .indexOnly(userLabel)
    .addKey(displayNameLowercaseKey, Mapping.TEXTSTRING.asParameter())
    .addKey(emailLowercaseKey, Mapping.TEXTSTRING.asParameter())
    .addKey(firstNameKey, Mapping.TEXTSTRING.asParameter())
    .addKey(lastNameKey, Mapping.TEXTSTRING.asParameter())
    .buildMixedIndex("search");

When running the query, an exception is thrown.

Query needs suitable index to be answered [(~label = USER)]:VERTEX

However, if I match against a property at a time, no exception is thrown. Like so:

GraphFactory.getDefault().traversal().V()
    .hasLabel(VertexLabel.USER.name())
    .has(PropertyKey.EMAIL_LC.name(), Text.textRegex(regex));

// or

GraphFactory.getDefault().traversal().V()
    .hasLabel(VertexLabel.USER.name())
    .has(PropertyKey.DISPLAY_NAME_LC.name(), Text.textRegex(regex));

// or...

How could I restructure the query and/or index for this to work?

score 3 · Accepted Answer · answered Jan 16 '18 at 14:50

3

Try to use a direct index query:

final String searchTerm =
    "v." + PropertyKey.EMAIL_LC.name() + ":(" + regex + ") OR " +
    "v." + PropertyKey.DISPLAY_NAME_LC.name() + ":(" + regex + ") OR " +
    "v." + PropertyKey.FIRST_NAME.name() + ":(" + regex + ") OR " +
    "v." + PropertyKey.LAST_NAME.name() + ":(" + regex + ")";

GraphFactory.getDefault().indexQuery("userSearch", searchTerm).vertices()

The great thing about direct index queries is, that you can also use things like query boosting, which comes handy if you search through multiple fields.

answered Jan 16 '18 at 14:50

Daniel Kuppitz

10,846
1
25
34

1

Great, it works! Took me several hours though to figure out that the index needs about a second after definition to be available to a direct index query; Thus automatic tests kept failing to find expected nodes. May I suggest adding this as a note to your answer? Thanks for introducing me to this kind of queries. – Double M Jan 18 '18 at 12:46
If the index the query is executed on is restricted to a set of vertex labels, only results with such labels will be returned, right? – Double M Feb 25 '18 at 14:11
1

Obviously yes, the index will not contain any data of other vertices. – Daniel Kuppitz Feb 26 '18 at 15:11

What index is needed to answer a disjunct (__.or) query?

1 Answers1