Cruddl compatibility with multi-node ArangoDB clusters

Hi, is Cruddl expected to generate AQL that is compatible with anonymous graph traversals on multi-node clusters? It seems as if the generated AQL is missing the necessary WITH <collections> stanza, resulting in error messages like the following:

Query: AQL: Error message received from cluster node 'PRMR-tn50ynem': collection not known to traversal: 'organizations'. please add 'WITH organizations' as the first line in your AQL (while executing)

[anderson900]

I wanted to follow-up on this. I don’t see any contributor guides in the repo’s READMEs so it isn’t clear if you’d appreciate/accept a PR around this issue, but here is my MVP solution to this problem. I tested it by setting --query.require-with true when launching the ArangoDB database for the test suite (docker run --rm -e ARANGO_NO_AUTH=true -p 8529:8529 arangodb/arangodb:3.9.3 --query.require-with true).

Here is my modification to getAQLQuery that adds a WITH <collections> statement for any collection implicitly required by a traversal but not otherwise included in the query. Likely this isn’t the most elegant solution or the right way to solve the problem in the long run, but this is what I was able to come up wit on short notice and not much familiarity with the Cruddl internals. Suggestions for a different approach are welcome, or if you’d like me to open a PR, let me know.

export function getAQLQuery(node: QueryNode): AQLCompoundQuery {
    const context = new QueryContext();

    const compoundQuery = createAQLCompoundQuery(
        node,
        aql.queryResultVariable('result'),
        undefined,
        context,
    );

    const traversalCollections = new Set<string>();

    visitQueryNode(node, {
        leave(object, key: string | undefined) {
            if (object instanceof TraversalQueryNode) {
                (object.relationSegments || []).forEach(seg => {
                    [seg.relationSide.sourceType, seg.relationSide.targetType].forEach(type => {
                        traversalCollections.add(getCollectionForType(type, AccessType.READ, context).toString());
                    })
                })
            } else if (object instanceof FollowEdgeQueryNode) {
                [object.relationSide.sourceType, object.relationSide.targetType].forEach(type => {
                    traversalCollections.add(getCollectionForType(type, AccessType.READ, context).toString());
                });
            } else if (object instanceof RemoveEdgesQueryNode) {
                [object.relation.fromSide.sourceType, object.relation.fromSide.targetType].forEach(type => {
                    traversalCollections.add(getCollectionForType(type, AccessType.READ, context).toString());
                });
            }

            return object;
        }
    });

    if (traversalCollections.size > 0) {
        const addTraversalContext = (originalQuery: AQLCompoundQuery): AQLCompoundQuery => {
            const referencedCollections = new Set([...compoundQuery.readAccessedCollections, ...compoundQuery.writeAccessedCollections]);
            const unknownCollections = [...traversalCollections].filter(c => !referencedCollections.has(c));

            const traversalContext = (unknownCollections.length > 0) ? aql.code(`WITH ${[...unknownCollections].join(", ")}\n`) : aql.code('');

            return new AQLCompoundQuery(
                originalQuery.preExecQueries.map(peq => addTraversalContext(peq)),
                new AQLCompoundFragment([traversalContext, originalQuery.aqlQuery]),
                originalQuery.resultVar,
                originalQuery.resultValidator,
                originalQuery.readAccessedCollections,
                originalQuery.writeAccessedCollections,
            );
        };

        return addTraversalContext(compoundQuery);
    }

    return compoundQuery;
}

[anderson900]

This continues to look good for us in our environments with clustered ArangoDB deployments. Thanks for the quick turn-around!