Slow join query in clustered setup

[vzamashka]

My Environment

• ArangoDB Version: 3.4.6
• Storage Engine: RocksDB
• Deployment Mode: Cluster
• Deployment Strategy: ArangoDB Starter
• Infrastructure: own local network
• Operating System: MacOS
• Total RAM in your machine: 2 machines wiht 8 Gb of RAM, 1 machine with 16 Gb
• Disks in use: SSD

Component, Query & Data

AQL query (if applicable):
FOR pobj IN PlatformObj let user = (for user, link in outbound pobj ParentChildLink RETURN {vertex: user, edge: link}) Return {r: pobj, c: user[0].vertex}

AQL explain (if applicable):
Execution plan:
Id NodeType Site Est. Comment
1 SingletonNode DBS 1 * ROOT
2 EnumerateCollectionNode DBS 100001 - FOR pobj IN PlatformObj /* full collection scan, 3 shard(s) /
13 RemoteNode COOR 100001 - REMOTE
14 GatherNode COOR 100001 - GATHER
7 SubqueryNode COOR 100001 - LET user = ... / subquery /
3 SingletonNode COOR 1 * ROOT
4 TraversalNode COOR 1 - FOR user / vertex /, link / edge / IN 1..1 / min..maxPathDepth / OUTBOUND pobj / startnode / ParentChildLink
5 CalculationNode COOR 1 - LET #8 = { "vertex" : user, "edge" : link } / simple expression /
10 LimitNode COOR 1 - LIMIT 0, 1
6 ReturnNode COOR 1 - RETURN #8
8 CalculationNode COOR 100001 - LET #10 = { "r" : pobj, "c" : user[0].vertex } / simple expression / / collections used: pobj : PlatformObj */
9 ReturnNode COOR 100001 - RETURN #10

Indexes used:
By Type Collection Unique Sparse Selectivity Fields Ranges
4 edge ParentChildLink false false n/a [ _from ] base OUTBOUND

Traversals on graphs:
Id Depth Vertex collections Edge collections Options Filter / Prune Conditions
4 1..1 ParentChildLink uniqueVertices: none, uniqueEdges: path

Optimization rules applied:
Id RuleName
1 optimize-subqueries
2 scatter-in-cluster
3 remove-unnecessary-remote-scatter

Dataset:
2 collections
PlatformObj - document, 100k rows
ParentChildLink - edge, 100k rows

Replication Factor & Number of Shards (Cluster only):
3 shards

Problem:
When executing join query on standalone setup, it takes 3-5 seconds
on clustered setup - 4-5 minutes
Can you help, please?