Configure Solr replica placement using PolicyAffinityPlacementFactory in Fusion 5.9+

Goal

Configure Solr replica placement to target specific node types in Fusion 5.9 and above, using the Lucidworks-customized PolicyAffinityPlacementFactory plugin. This is particularly relevant after the deprecation of Solr's native autoscaling policy framework in Solr 9.

Environment

• Fusion 5.9.2 and above
• Solr 9.1 and above
• Kubernetes deployments (EKS, GKE, AKS, or other platforms)

Guide

Understand the Lucidworks replica placement plugin

Fusion 5.9+ includes a customized plugin for Solr replica placement:

com.lucidworks.solr.plugin.placement.PolicyAffinityPlacementFactory

This plugin extends the default AffinityPlacementFactory and allows defining placement policies that map replica types (NRT, TLOG, PULL) to Solr node types, which are identified via the -Dnode_type system property in your Solr container spec (for example, -Dnode_type=search,analytics).

Configure the placement plugin

To enable and configure the plugin, POST the following payload to the Solr /api/cluster/plugin endpoint:

{
  "update": {
    "name": ".placement-plugin",
    "class": "com.lucidworks.solr.plugin.placement.PolicyAffinityPlacementFactory",
    "config": {
      "minimalFreeDiskGB": 20,
      "prioritizedFreeDiskGB": 200,
      "withCollections": {},
      "collectionNodeType": {},
      "policies": {
        "system": {
          "*": "system"
        },
        "search": {
          "nrt": "search",
          "tlog": "search",
          "pull": "analytics,search"
        },
        "analytics": {
          "*": "analytics"
        }
      },
      "strictPolicy": false
    }
  }
}

Example cURL for Fusion.

curl -u "<username>:<password>" -X POST -H "Content-Type: application/json" \
"http://104.196.224.82:6764/api/solrAdmin/api/cluster/plugin" \
-d '{
"update": {
"name": ".placement-plugin",
"class": "com.lucidworks.solr.plugin.placement.PolicyAffinityPlacementFactory",
"config": {
"minimalFreeDiskGB": 20,
"prioritizedFreeDiskGB": 200,
"withCollections": {},
"collectionNodeType": {},
"policies": {
"system": {
"*": "system"
},
"search": {
"nrt": "search",
"tlog": "search",
"pull": "analytics,search"
},
"analytics": {
"*": "analytics"
}
},
"strictPolicy": false
}
}
}'

Key configuration fields

• policies: Each key is a policy name. Collections must reference a policy by setting the property.policy property in their configuration.
• replica types: Use lowercase names nrt, tlog, and pull. The special value "*" applies to all replica types.
• node_type: Each Solr node must have the node_type system property set to match the policy definitions.
• strictPolicy:
  • false (default): Nonexistent or misconfigured policies will be ignored and fallback logic is applied.
  • true: Placement will fail if nodes don't meet policy constraints or if an unknown policy is referenced.

Example: Assign collections to policies

Ensure your collections are tagged with the appropriate policy by setting the property.policy field during creation.

This instructs Solr to use the search policy block defined in the plugin config.

Understand Fusion's default behavior

Fusion applies internal logic when collections are created through its UI or APIs. The following rules apply:

• type=DATA: NRT and PULL replicas are placed in search; PULL replicas may also go to analytics. TLOG and PULL must be explicitly requested via solrParams.
• type=SIGNALS: Replicas are placed in analytics. Default is NRT unless changed.
• type=SIGNALS_AGGREGATION: Replicas default to search, regardless of solrParams. Requires a placement policy override to use analytics.
• type=QUERY_REWRITE: Single-shard NRT replica in search.

Important: Replica placement decisions are only applied at collection creation time. Updating an existing collection's policy does not reassign existing replicas.

Additional guidance

• To ensure configuration changes take effect, restart the Solr deployments and statefulsets.
• If an application is created without the relatedObjects=false parameter, Fusion may auto-create related collections (e.g., signals, signals_aggr), preventing full control over placement configuration.

For full control, create collections explicitly using Solr APIs and assign appropriate properties to match your policy configurations.