Schema Design and Field Types

Solr schema defines how fields are indexed, stored, and analyzed. Each field has a type and analysis chain consisting of tokenizer and filters. Field types like text_en apply language-specific stemming, stop-word removal, and lowercasing.

Copy fields aggregate multiple source fields into a single searchable field. Dynamic fields accept field names matching a pattern for schemas that evolve frequently.

Indexing: Full Import and Delta Updates

Solr indexes documents via POSTing XML, JSON, or CSV to /update handler. The DataImportHandler pulls data from relational databases. Full imports rebuild the entire index; delta imports update changed documents.

The commit operation makes documents searchable. Hard commit fsyncs to disk; soft commit makes documents visible without fsyncing.

# Index via JSON:
curl -X POST -H 'Content-Type: application/json' \
  http://localhost:8983/solr/products/update?commit=true \
  -d '[
    {"id": "1", "title": "Wireless Mouse", "price": 29.99},
    {"id": "2", "title": "Mechanical Keyboard", "price": 89.99}
  ]'

Querying: edismax and Filtering

The edismax query parser supports fielded search with boosts, phrase boosting, and minimum-match for fuzzy queries. The qf parameter lists fields with relative boosts. Filter queries (fq) narrow results without affecting scoring.

I use fq for all structured filters — category, price range, in-stock status. Faceting computes counts per field value for navigation.

http://localhost:8983/solr/products/select?
  q=wireless+mouse&
  qf=title^4+description^2+category&
  pf=title^10&
  fq=price:[10+TO+100]&
  fq=in_stock:true&
  facet=true&
  facet.field=category

SolrCloud: Distributed Search and Replication

SolrCloud distributes index data across nodes using shards and replicas. A collection is split into shards, each with multiple replicas for redundancy and read scalability. ZooKeeper manages cluster state, leader election, and shard assignment.

I aim for 5-20 million documents per shard.

# Create a SolrCloud collection:
bin/solr create -c products -shards 4 -replicationFactor 3
# Check status:
http://localhost:8983/solr/admin/collections?action=CLUSTERSTATUS

Analysis and Tokenization Pipeline

Solr analysis chain determines how text is tokenized and filtered. A standard chain includes StandardTokenizer, LowerCaseFilter, StopFilter, and SnowballPorterStemmerFilter. Multi-language content needs different field types per language.

The Analysis page in the Admin UI is invaluable for debugging how text passes through each analyzer step.

Performance Tuning: Caches and Merges

Solr uses three caches: filter cache (fq results), query result cache (complete queries), and document cache (stored fields). The filter cache is most impactful — reuse of fq across requests avoids re-evaluation.

Index merges consolidate segments. TieredMergePolicy is the default and works well for most workloads.