Receive `ArrayIndexOutOfBoundsException` when inserting to type with full text index with multiple properties defined.

[alan-strickland-red]

ArcadeDB Version:

ArcadeDB Server v23.12.1 (build df4e3d5/1704296563159/main)

OS and JDK Version:

Linux 5.10.0-26-amd64 - OpenJDK 64-Bit Server VM 11.0.21 (Temurin-11.0.21+9)

Expected behavior

No errors when inserting new records to a type that has a full text index containing multiple fields.

Actual behavior

For the most part this works as expected, however now and again we find when inserting to a document type we starting receiving ArrayIndexOutOfBoundsException errors.

We then drop and create the full text index which resolves the error.

I don't seem to get the same issue with full text indexes with a single property however because it feels quite random it might be that I haven't come across it yet.

Steps to reproduce

I have not yet found a way to consistently reproduce the issue, I will keep trying though.

[lvca]

@alan-strickland-red are you inserting with the local or remote interface?
Also, are you executing operations in parallel or single thread?

[alan-strickland-red]

I am inserting values via the http client from nodejs, and I also tried it via the http client in studio and it gives the same error.

Also, are you executing operations in parallel or single thread?

Single thread I believe, asyncWorkerThreads is set to 1, typeDefaultBuckets is set to 1

[lvca]

I'd love to reproduce the issue. Do you have a stack trace of the exception to understand where it's thrown?

[alan-strickland-red]

@lvca Sorry for the delay in getting back to you, been working some short weeks recently and had other commitments.

This issue has cropped up again this morning, on a little used database.

I have attached a backup of the database and the command that is causing the error.

I can't see a stack trace anywhere, the error in the client does not have one and I can't see the error in the logs.

Command causing the error:

INSERT INTO `authentication-token` CONTENT {"id":"613b7796-87f2-4bc6-8491-f11215995aff","issued":"2024-01-29T14:42:36.465Z","expires":"2024-01-29T14:57:36.466Z","ip":"::ffff:10.99.2.3","authenticated":false,"mfaCompleted":false,"revoked":false,"userAction":"","_sys":{"rev":1,"hash":"authentication-token!613b7796-87f2-4bc6-8491-f11215995aff","createdon":"2024-01-29T14:42:36.467Z","updatedon":"2024-01-29T14:42:36.467Z"},"$$id":"613b7796-87f2-4bc6-8491-f11215995aff","@type":"authentication-token"}

Database backup with error

[lvca]

Reproduced, thanks!

[alan-strickland-red]

Now you have reproduced the issue you might not need this but I noticed that if I create a database with a full text index then restart the docker container the reported type of the index in studio switches from FULL_TEXT to LSM_TREE.

Before the restart

                {
                    "name": "authentication-token[ip,userAction]",
                    "typeName": "authentication-token",
                    "type": "FULL_TEXT",
                    "unique": false,
                    "properties": [
                        "ip",
                        "userAction"
                    ],
                    "automatic": true
                }

After the restart

                {
                    "name": "authentication-token[ip,userAction]",
                    "typeName": "authentication-token",
                    "type": "LSM_TREE",
                    "unique": false,
                    "properties": [
                        "ip",
                        "userAction"
                    ],
                    "automatic": true
                }

[lvca]

We have zeros test case with composite full text indexes. It was easy to reproduce your issue in a test case. I'll look also into the type issue after restart

[lvca]

the composite index is not supported in the full text index: it always creates a sub LSMTree index with 1 string property:

arcadedb/engine/src/main/java/com/arcadedb/index/lsm/LSMTreeFullTextIndex.java

Line 97 in 5ba0571

underlyingIndex = new LSMTreeIndex(database, name, false, filePath, mode, new Type[] { Type.STRING }, pageSize, nullStrategy);

.

The creation should throw an exception, but instead it completely ignores the passed key types (assuming it's always one string).

About the type showing LSM_TREE it's because it's a LSM_TREE under the hood, but this should be hidden for the end user that should see FULL_TEXT instead.

What's your use case where you have a composite full text index? Can you have 2 indexes?

[alan-strickland-red]

I think originally I did have multiple indexes and then I went through a period where I was trying to reduce the overall disk/memory footprint of the database and realised that changing it to one index helped to reduce the disk usage.

Although maybe my recollection is wrong because if I try to create a type that has multiple full text indexes I can't query using CONTAINSTEXT I get an error, so perhaps that's why I changed it.

Multiple Indexes Example

CREATE DOCUMENT TYPE MultipleIndexes IF NOT EXISTS;
CREATE PROPERTY MultipleIndexes.String1 IF NOT EXISTS string ;
CREATE PROPERTY MultipleIndexes.String2 IF NOT EXISTS string ;
CREATE INDEX ON MultipleIndexes (String1) FULL_TEXT;
CREATE INDEX ON MultipleIndexes (String2) FULL_TEXT;

INSERT INTO MultipleIndexes CONTENT {String1: 'Joe', String2: 'Bloggs'};

SELECT * FROM MultipleIndexes WHERE `String1` CONTAINSTEXT 'Joe';
SELECT * FROM MultipleIndexes WHERE `String2` CONTAINSTEXT 'Bloggs';

Multiple Indexes Query Result

{
    "error": "Internal error",
    "detail": "Cannot execute index query with `String1` CONTAINSTEXT \u0027Joe\u0027",
    "exception": "java.lang.UnsupportedOperationException"
}

Single Index Example

CREATE DOCUMENT TYPE SingleIndex IF NOT EXISTS;
CREATE PROPERTY SingleIndex.String1 string;
CREATE PROPERTY SingleIndex.String2 string;
CREATE INDEX ON SingleIndex (String1, String2) FULL_TEXT;

INSERT INTO SingleIndex CONTENT {String1: 'Joe', String2: 'Bloggs'};

SELECT * FROM SingleIndex WHERE `String1` CONTAINSTEXT 'Joe';
SELECT * FROM SingleIndex WHERE `String2` CONTAINSTEXT 'Bloggs';

Single Index Query Result - before a restart of the server

{
    "user": "root",
    "version": "23.12.1 (build df4e3d56e9063a245b46bca14001e393dc7f8001/1704296563159/main)",
    "serverName": "ArcadeDB_0",
    "result": {
        "vertices": [],
        "edges": [],
        "records": [
            {
                "@rid": "#25:0",
                "@type": "SingleIndex",
                "@cat": "d",
                "String2": "Bloggs",
                "String1": "Joe"
            }
        ]
    },
    "explain": "+ DDL\n  create document type SingleIndex IF NOT EXISTS\n+ DDL\n  CREATE PROPERTY SingleIndex.String1 string\n+ DDL\n  CREATE PROPERTY SingleIndex.String2 string\n+ DDL\n  CREATE INDEX SingleIndex[String1,String2] ON SingleIndex (String1, String2) FULL_TEXT NULL_STRATEGY SKIP\n+ CREATE EMPTY RECORDS\n  1 record\n+ SET TYPE\n  SingleIndex\n+ UPDATE CONTENT\n  {\"String1\": \u0027Joe\u0027, \"String2\": \u0027Bloggs\u0027}\n+ SAVE RECORD\n+ FETCH FROM TYPE SingleIndex\n  + FETCH FROM BUCKET 25 (SingleIndex_0) ASC \u003d 1 RECORDS\n  + FETCH FROM BUCKET 26 (SingleIndex_1) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 27 (SingleIndex_2) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 28 (SingleIndex_3) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 29 (SingleIndex_4) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 30 (SingleIndex_5) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 31 (SingleIndex_6) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 32 (SingleIndex_7) ASC \u003d 0 RECORDS\n+ FILTER ITEMS WHERE \n  `String1` CONTAINSTEXT \u0027Joe\u0027\n+ CALCULATE PROJECTIONS\n  *\n+ FETCH FROM TYPE SingleIndex\n  + FETCH FROM BUCKET 25 (SingleIndex_0) ASC \u003d 1 RECORDS\n  + FETCH FROM BUCKET 26 (SingleIndex_1) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 27 (SingleIndex_2) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 28 (SingleIndex_3) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 29 (SingleIndex_4) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 30 (SingleIndex_5) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 31 (SingleIndex_6) ASC \u003d 0 RECORDS\n  + FETCH FROM BUCKET 32 (SingleIndex_7) ASC \u003d 0 RECORDS\n+ FILTER ITEMS WHERE \n  `String2` CONTAINSTEXT \u0027Bloggs\u0027\n+ CALCULATE PROJECTIONS\n  *"
}

[lvca]

@alan-strickland-red I just pushed a check to return an error in case you create a full-text composite index.

Try to create 2 indexes, one per property.