Source Code Reading: SearchService — Unified Abstraction for 6 Search Modes

“

Series: S9 Source Code Reading · Article 8 | Level: Advanced | Reading Time: 25 min

#TL;DR

DefaultSearchService is the full-text search service in Data Layer, built on Quarkus 3.x, leveraging Doris OLAP's full-text indexing and Redis ZSET structures. It supports 6 search modes (BEST_MATCH/EXACT/PREFIX/FUZZY/WILDCARD/REGEX), generates parameterized SQL via SearchQueryBuilder, performs parallel multi-type search with result merging. Core features include: parallel multi-type search (thread pool + CompletableFuture), result highlighting (HighlightProcessor), three Facet aggregation types (Terms/Range/DateRange), Redis-driven recent and popular search tracking, sparse-result fuzzy spell suggestions, and saved search CRUD. This article analyzes the seven-step search flow, thread pool design for parallel search, Redis ZSET recent/popular implementation, three Facet bucketing computations, and automatic spell suggestion fallback strategy.

#Table of Contents

1. Architecture and Five Collaborators
2. Main Search Flow: Seven Steps
3. Parallel Multi-Type Search
4. 6 Search Modes
5. Result Highlighting
6. Facet Aggregation: Terms / Range / DateRange
7. Redis Recent and Popular Search
8. Autocomplete Suggestions
9. Spell Suggestion Fuzzy Fallback
10. Saved Search CRUD
11. Key Takeaways

#1. Architecture and Five Collaborators

@ApplicationScoped
public class DefaultSearchService implements SearchService {
    private final DorisClient dorisClient;
    private final RedisDataSource redisDataSource;
    private final SavedSearchRepository savedSearchRepository;
    private final HighlightProcessor highlightProcessor;
    private final SearchQueryBuilder queryBuilder;

    private final ExecutorService searchExecutor =
            Executors.newFixedThreadPool(Math.min(10, Runtime.getRuntime().availableProcessors() * 2));
}

#2. Main Search Flow: Seven Steps

1. Resolve entity types to search (from filter or all distinct types)
2. Parallel per-type search via CompletableFuture + thread pool
3. Sort by score DESC, truncate to page size
4. Highlight name fields if requested
5. Compute total hits and next page token
6. Spell suggestions when results < 5 (fuzzy fallback)
7. Facet aggregation if facet requests present

#3. Parallel Multi-Type Search

private final ExecutorService searchExecutor =
        Executors.newFixedThreadPool(Math.min(10, Runtime.getRuntime().availableProcessors() * 2));

Thread pool sizing: min(10, CPU * 2) caps max parallelism at 10 on high-core machines, preventing excessive Doris connection pressure. @PreDestroy ensures graceful shutdown with 5-second timeout.

#4. 6 Search Modes

Default is BEST_MATCH when SearchMode is unspecified.

#5. Result Highlighting

if (request.getIncludeHighlights() && query != null && !query.isBlank()) {
    allHits = applyHighlights(allHits, query, preTag, postTag);
}

HighlightProcessor marks matching query terms in the name field using customizable pre/post tags (e.g., <em> / </em>).

#6. Facet Aggregation: Terms / Range / DateRange

#7. Redis Recent and Popular Search

private static final String REDIS_RECENT_PREFIX = "search:recent:";  // per-user ZSET
private static final String REDIS_POPULAR_PREFIX = "search:popular:"; // per-world ZSET
private static final int MAX_RECENT_SIZE = 100;

• Recent: search:recent:{userId} ZSET, score = timestamp, keeps latest 100 entries
• Popular: search:popular:{worldId} ZSET, score = search count

Double-checked locking pattern for lazy Redis SortedSet command handle initialization.

#8. Autocomplete Suggestions

Three suggestion types ranked by score: Instance suggestions (1.0) > Recent keywords (0.9) > Popular keywords (0.7). Instance suggestions from Doris prefix search, keyword suggestions from Redis ZSETs.

#9. Spell Suggestion Fuzzy Fallback

if (allHits.size() < 5 && mode != SearchMode.FUZZY) {
    spellSuggestions = collectSpellSuggestions(worldId, entityTypes, query, filter);
}

When search results are fewer than 5 and current mode is not FUZZY, automatically performs a FUZZY search fallback, extracting top 3 names as spell suggestions -- similar to Google's "Did you mean..." feature.

#10. Saved Search CRUD

Saved search functionality allows users to name and save search queries for later reuse, persisted via SavedSearchRepository. Supports list (with pagination), delete, and get-recent operations.

#11. Key Takeaways

1. Parallel multi-type search: Thread pool + CompletableFuture queries multiple entity types in parallel, merged and sorted by score
2. 6 search modes: From exact match to regex, covering all search scenarios
3. Redis ZSET tracking: Recent (per-user) and popular (per-world) searches via efficient ZSET operations
4. Three Facet types: Terms/Range/DateRange cover categorical, numeric, and temporal aggregation
5. Automatic spell suggestions: Sparse results trigger FUZZY fallback for "Did you mean..." experience
6. Result highlighting: Customizable pre/post tag keyword highlighting

#Next Article

S9-09: MetricRegistryService -- Metric Computation Strategy Routing. We will dive into the metric registry service, examining metric registration, validation, strategy recommendation, and automatic materialized view integration.

Tags: #coomia-dip #source-code-reading #data-Layer #search #full-text #doris #redis #facets #autocomplete #spell-suggest