Solr is the open source search engine at the core of Fusion. When you index your data, it is stored in a Solr collection.

​
Collections

Your data is organized into collections. When you create an app, Fusion automatically creates a collection with the same name. You can create additional collections in any app. A primary collection contains the data that your users will search. Every primary collection is associated with a set of auxiliary collections that contain related data, such as signals, aggregations, and more. Under the hood, a Fusion collection is a distributed index in Solr, defined by a named configuration stored in ZooKeeper, with these properties:
  • Number of shards
    Documents are distributed across this number of partitions.
  • Document routing strategy
    How documents are assigned to shards.
  • Replication factor
    How many copies of each document in the collection.
  • Replica placement strategy
    Where to place replicas in the cluster.
If your data is already stored in a Solr instance or cluster, you can manage this collection in Fusion by creating a Fusion collection that imports the existing Solr collection.
Collection names are case-insensitive, but Fusion preserves case when displaying collection names.

​
Automated Solr backups

You can schedule backups of Solr collections, store the backups for a configurable period of time, and restore these backups into a specified Fusion cluster when needed. The following guide uses Google Kubernetes Engine (GKE) for examples and assumes you used the setup scripts in the fusion-cloud-native repository to install Fusion.
The standard approach of using a provider-specific Persistent Volume Claim (PVC) for storing collection backups ensures consistency in configuration. However, this method does not leverage the unique storage features offered by each cloud provider. For instance, Google Cloud provides Google Cloud Storage, which includes additional features such as access control management, various storage tiers, and other capabilities that are not available when using a PVC. To take advantage of these features, Solr instances running within Fusion require additional provider-specific information.Refer to the Solr documentation for detailed information on the repositories available for configuring collection backups. Each repository type comes with specific configuration options and features. Generally, you will need to integrate the provider-specific configuration into the solr.xml configuration file and ensure that the appropriate library or module for the provider is included in the Solr classpath. This step is necessary for the repository implementation to resolve correctly at runtime.For example, when configuring GCSBackupRepository to store backups in Google Cloud Storage (GCS), it is essential to include the corresponding library for the provider in the Solr classpath. Additionally, you will need to add a section to the solr.xml file similar to the XML example below to specify the target bucket where backups will be stored:
<backup>
  <repository name="gcs_backup" class="org.apache.solr.gcs.GCSBackupRepository" default="false">
    <str name="gcsBucket">solrBackups</str>
    <str name="gcsCredentialPath">/local/path/to/credential/file</str>
    <str name="location">/default/gcs/backup/location</str>
    <int name="gcsClientMaxRetries">5</int>
    <int name="gcsClientHttpInitialRetryDelayMillis">1500</int>
    <double name="gcsClientHttpRetryDelayMultiplier">1.5</double>
    <int name="gcsClientHttpMaxRetryDelayMillis">10000</int>
  </repository>
</backup>
After configuring the backup provider, you can utilize the standard Solr backup and restore APIs to create new backups or restore from existing ones. Instead of writing to a PVC, backups will be stored in the storage solution specific to the provider.
Backups are taken using the Solr collection BACKUP command. This requires that each Solr node has access to a shared volume or a ReadWriteMany volume in Kubernetes. Most cloud providers offer a simple way of creating a shared filestore and exposing it as a PersistentVolumeClaim within Kubernetes to mount into the Solr pods. An option is added to the setup_f5_PROVIDER.sh scripts in the fusion-cloud-native repository to provision these.The backup action of the script is invoked by a Kubernetes CronJob to run the backup schedule. The backups are saved to a configurable directory with an automatically generated name: <collection_name>-<timestamp_in_some_format>.A separate CronJob is responsible for cleanup and retention of backups. Cleanup can be disabled if not needed. Setting a series of retention periods can automatically remove backups as they become outdated.For example, a cluster that backs up a collection every 3 hours could specify a retention policy that:
  • Keeps all backups for a single day.
  • Keeps a single daily backup for a week.
  • Keeps a single weekly backup for a month.
  • Keeps a single monthly backup for 6 months.
  • Deletes all backups that are older than 6 months.
All times are configurable as part of the configmap for this service.The process for restoring a collection is a manual step involving kubectl run to invoke the Solr RESTORE action pointing to the collection and the name of the backup being restored.
These instructions are for GKE only. For other platforms, backup and restoration involves copying the collection to the cloud and using Parallel Bulk Loader.

​
Install using a PVC with GKE

The solr-backup-runner requires that a ReadWriteMany volume is mounted onto all solr and backup-runner pods so they all back up to the same filesystem.The easiest way to install on GKE is by using a GCP Filestore as the ReadWriteMany volume.
  1. Create the Filestore. 
    gcloud --project "${GCLOUD_PROJECT}" filestore instances create "${NFS_NAME}"  --tier=STANDARD --file-share=name="solrbackups,capacity=${SOLR_BACKUP_NFS_GB}GB" --zone="${GCLOUD_ZONE}" --network=name="${network_name}"
  2. Fetch the IP of the Filestore. 
    NFS_IP="$(gcloud filestore instances describe "${NFS_NAME}" --project="${GCLOUD_PROJECT}" --zone="${GCLOUD_ZONE}" --format="value(networks.ipAddresses[0])")"
  3. Create a Persistent Volume in Kubernetes that is backed by this volume. 
    cat <<EOF | kubectl -n "${NAMESPACE}" apply -f -
apiVersion: v1
kind: PersistentVolume
metadata:
 name: ${NAMESPACE}-solr-backups
 annotations:
   pv.beta.kubernetes.io/gid: "8983"
spec:
 capacity:
   storage: ${SOLR_BACKUP_NFS_GB}G
 accessModes:
   - ReadWriteMany
 nfs:
   path: /solrbackups
   server: ${NFS_IP}
EOF
  4. Create a Persistent Volume Claim in the namespace that Solr is running in. 
    cat <<EOF | kubectl -n "${NAMESPACE}" apply -f -
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
 name: fusion-solr-backup-claim
spec:
 volumeName: ${NAMESPACE}-solr-backups
 accessModes:
   - ReadWriteMany
 storageClassName: ""
 resources:
   requests:
     storage: ${SOLR_BACKUP_NFS_GB}G
EOF
  5. Add the following values to your existing (or a new) Helm values file. 
    solr-backup-runner:
 enabled: true
 sharedPersistentVolumeName: fusion-solr-backup-claim
solr:
 additionalInitContainers:
   - name: chown-backup-directory
     securityContext:
       runAsUser: 0
     image: busybox:latest
     command: ['/bin/sh', '-c', "owner=$(stat -c '%u' /mnt/solr-backups);  if [ ! \"${owner}\" = \"8983\" ]; then chown -R 8983:8983 /mnt/solr-backups; fi "]
     volumeMounts:
       - mountPath: /mnt/solr-backups
         name: solr-backups
 additionalVolumes:
   - name: solr-backups
     persistentVolumeClaim:
       claimName: fusion-solr-backup-claim
 additionalVolumeMounts:
   - name: solr-backups
     mountPath: "/mnt/solr-backups"
  6. Upgrade the release. Solr backups are now enabled.

​
Auxiliary Collections

Every primary collection is associated with a set of auxiliary collections that contain related data, such as signals, aggregations, and more. Some auxiliary collections are created for every primary collection. Others are created only for the app’s default collection, one per app. Auxiliary collections are described below:
APP_NAME_job_reportsOutput from Fusion experiments, Ranking Metrics jobs, and Head/Tail Analysis jobs.1 per app
APP_NAME_query_rewrite

A collection of documents to use for rewriting queries, optimized for high‑volume traffic. These documents originate from the COLLECTION_NAME_query_rewrite_staging collection. Certain Fusion query pipeline stages read from this collection:
• Text Tagger
• Apply Rules
• Modify Response with Rules

1 per app
APP_NAME_query_rewrite_staging

A collection of documents created by the Rules Editor or by certain Fusion jobs, not optimized for production traffic. Documents move from this collection to the COLLECTION_NAME_query_rewrite collection as follows:
• Job output documents with high confidence contain a review=auto field and are moved to the COLLECTION_NAME_query_rewrite collection automatically.
• Job output documents with low confidence contain a review=pending field. When these are approved by a Fusion user, Fusion copies them to the COLLECTION_NAME_query_rewrite collection.

1 per app
COLLECTION_NAME_signalsA search query logs and signals collection.1 per collection
COLLECTION_NAME_signals_aggrA collection for aggregated signals.1 per collection
APP_NAME_user_prefsA collection of data to support App Studio’s social features, such as user‑generated tags, bookmarks, comments, ratings, and so on.1 per app
Do not create primary collections with names that end in the suffixes above; these are reserved for Fusion auxiliary collections, which are created and managed by Fusion directly.
Fusion maintains a set of Solr collections that store Fusion’s own log files and other internal information. These are called System Collections, described below.
Do not create primary collections named “logs” or beginning with “system_”. These names are reserved for Fusion system collections.
Fusion uses ZooKeeper to register information about all collections, and the Fusion components and services related to a collection. The Fusion components associated with a collection include:
  • Datasources
  • Pipelines
  • Profiles
  • Signals and aggregations
  • Analytics dashboards

​
System Collections

Fusion automatically creates some collections that are used for internal purposes and shared across all apps:
  • system_autocomplete store the content that the Fusion UI displays when you use the search bar.
  • system_blobs stores blobs in Solr. This is used to store model files for the NLP components and other binary files used by Fusion components.
  • system_history keeps a record of configuration changes, start and stop times for services and experiments, and more.
  • system_jobs_history keeps a record of Fusion jobs, including start/stop times and status.
  • system_messages is used by Fusion’s messaging services.

​
Collection Configuration Properties

Collections have three properties that you can configure only when you are creating a collection using the Collections API.
PropertyDescriptionDefault behavior
signals*The signals property determines whether to create auxiliary collections with suffixes _signals and _signals_aggr.When you create a collection in the Fusion UI, signals defaults to true. When you create a collection using the Fusion API, this property defaults to false.
searchLogsThe searchLogs property determines whether to create an auxiliary search query logs collection with suffix _logs.When you create a collection in the Fusion UI, this property defaults to true. When you create a collection using the Fusion API, this property defaults to false.
*Signals are events with timestamps that can be used to improve search results. For more information about signals in Fusion, see Signals in the Fusion documentation. **In schemaless mode, if a document contains a field not currently in the Solr schema, Solr processes the field value to determine what the field type should be defined as, and then adds a new field to the schema with the field name and field type. This behavior can be convenient during preliminary application development, but it is rarely appropriate in a production environment.

​
Using profiles to associate collections with pipelines

Index pipelines and query pipelines are not connected to a specific collection by default. Index profiles and query profiles are configurations that create consistent endpoints for indexing and querying, each with a specific pipeline and collection.

​
Field Editor UI

The Fusion UI includes a space under Collections to edit Fields. The Fusion UI includes a space under Collections to edit Fields. Descriptions for these fields can be found in the Field Type Definitions section of the Solr Reference Guide associated with your Fusion release. Field options displayed in the UI include:
  • Dynamic checkbox (cannot change via UI)
  • Field Name (cannot change via UI)
  • Field Type (a preset value is shown that can be changed using edit mode)
  • Checkboxes for Indexed, Stored, Multivalued, Required
  • Text field to enter a Default Value
  • Copy Fields uses the plus sign to add rows (static can copy to raw_content or text; dynamic can copy to any raw_content/text or any other dynamic field)
  • Advanced toggles checkboxes for Doc Values, Omit Norms, Omit Positions, Omit term freq and positions, Term Vectors, Term Positions, Term Offsets