Algolia migration deep dive

Algolia Migration Deep Dive

Overview

Comprehensive guide for migrating from another search engine (Elasticsearch, Typesense, Meilisearch, or custom) to Algolia. Uses the strangler fig pattern: run old and new in parallel, gradually shift traffic, then cut over.

Migration Planning

From	Difficulty	Notes	Duration
Elasticsearch	Medium	query syntax differs significantly	2-4 weeks
Typesense	Low	similar hosted model	1-2 weeks
Meilisearch	Low	similar API concepts	1-2 weeks
Custom SQL LIKE	Low	major upgrade	1-2 weeks
Solr	Medium	config-heavy to API-driven	2-4 weeks

Instructions

Step 1: Assess Current Implementation

# Find all search-related code
grep -rn "elasticsearch\|elastic\|typesense\|meilisearch\|\.search(" \
  --include="*.ts" --include="*.tsx" --include="*.js" src/ | wc -l

# Inventory current search features used
grep -rn "aggregations\|facets\|filters\|sort\|highlight\|suggest" \
  --include="*.ts" --include="*.tsx" src/

// Document current capabilities
interface MigrationAssessment {
  currentEngine: string;
  recordCount: number;
  indexCount: number;
  features: {
    fullTextSearch: boolean;
    faceting: boolean;
    filtering: boolean;
    geoSearch: boolean;
    synonyms: boolean;
    customRanking: boolean;
    analytics: boolean;
    abTesting: boolean;
    recommendations: boolean;
  };
  integrationPoints: string[];  // Files that call the search engine
  queryPatterns: string[];       // Types of queries used
}

Step 2: Create the Adapter Layer

// src/search/adapter.ts
// Abstraction layer — both engines implement the same interface

interface SearchResult<T> {
  hits: T[];
  totalHits: number;
  totalPages: number;
  currentPage: number;
  facets?: Record<string, Record<string, number>>;
  processingTimeMs: number;
}

interface SearchAdapter {
  search<T>(params: {
    index: string;
    query: string;
    filters?: string;
    facets?: string[];
    page?: number;
    hitsPerPage?: number;
  }): Promise<SearchResult<T>>;

  index(params: { index: string; records: Record<string, any>[] }): Promise<void>;
  delete(params: { index: string; ids: string[] }): Promise<void>;
}

Step 3: Implement the Algolia Adapter

// src/search/algolia-adapter.ts
import { algoliasearch, ApiError } from 'algoliasearch';

export class AlgoliaAdapter implements SearchAdapter {
  private client;

  constructor(appId: string, apiKey: string) {
    this.client = algoliasearch(appId, apiKey);
  }

  async search<T>(params: {
    index: string;
    query: string;
    filters?: string;
    facets?: string[];
    page?: number;
    hitsPerPage?: number;
  }): Promise<SearchResult<T>> {
    const result = await this.client.searchSingleIndex<T>({
      indexName: params.index,
      searchParams: {
        query: params.query,
        filters: params.filters,
        facets: params.facets || ['*'],
        page: params.page || 0,
        hitsPerPage: params.hitsPerPage || 20,
      },
    });

    return {
      hits: result.hits,
      totalHits: result.nbHits,
      totalPages: result.nbPages,
      currentPage: result.page,
      facets: result.facets,
      processingTimeMs: result.processingTimeMS,
    };
  }

  async index(params: { index: string; records: Record<string, any>[] }) {
    const { taskID } = await this.client.saveObjects({
      indexName: params.index,
      objects: params.records.map(r => ({
        objectID: r.id || r.objectID,
        ...r,
      })),
    });
    await this.client.waitForTask({ indexName: params.index, taskID });
  }

  async delete(params: { index: string; ids: string[] }) {
    const { taskID } = await this.client.deleteObjects({
      indexName: params.index,
      objectIDs: params.ids,
    });
    await this.client.waitForTask({ indexName: params.index, taskID });
  }
}

Step 4: Query Translation Guide

// Elasticsearch → Algolia query translation

// ES: { "query": { "match": { "title": "laptop" } } }
// Algolia:
await client.searchSingleIndex({ indexName: 'products', searchParams: { query: 'laptop' } });

// ES: { "query": { "bool": { "filter": [{ "term": { "category": "electronics" } }] } } }
// Algolia:
await client.searchSingleIndex({
  indexName: 'products',
  searchParams: { query: '', filters: 'category:electronics' },
});

// ES: { "query": { "range": { "price": { "gte": 50, "lte": 200 } } } }
// Algolia:
await client.searchSingleIndex({
  indexName: 'products',
  searchParams: { query: '', numericFilters: ['price >= 50', 'price <= 200'] },
});

// ES: { "aggs": { "categories": { "terms": { "field": "category" } } } }
// Algolia:
await client.searchSingleIndex({
  indexName: 'products',
  searchParams: { query: '', facets: ['category'] },
});
// facets in response: { category: { electronics: 42, clothing: 18 } }

// ES: { "sort": [{ "price": "asc" }] }
// Algolia: Use a replica index with price sort ranking
await client.searchSingleIndex({ indexName: 'products_price_asc', searchParams: { query: '' } });

// ES: { "highlight": { "fields": { "title": {} } } }
// Algolia: Built-in, use _highlightResult in response

Step 5: Data Migration

// Full data migration with transform
async function migrateData(sourceAdapter: SearchAdapter, targetIndex: string) {
  const client = algoliasearch(process.env.ALGOLIA_APP_ID!, process.env.ALGOLIA_ADMIN_KEY!);

  // Use replaceAllObjects for atomic zero-downtime swap
  // Internally: creates temp index → indexes all records → moves temp → deletes old
  console.log(`Starting migration to ${targetIndex}...`);

  const allRecords: Record<string, any>[] = [];
  let page = 0;
  let hasMore = true;

  // Export from source
  while (hasMore) {
    const result = await sourceAdapter.search({
      index: 'products',
      query: '',
      page,
      hitsPerPage: 1000,
    });
    allRecords.push(...result.hits.map(transformRecord));
    hasMore = page < result.totalPages - 1;
    page++;
    console.log(`Exported ${allRecords.length} records...`);
  }

  // Import to Algolia atomically
  const { taskID } = await client.replaceAllObjects({
    indexName: targetIndex,
    objects: allRecords,
    batchSize: 1000,
  });
  await client.waitForTask({ indexName: targetIndex, taskID });

  console.log(`Migration complete: ${allRecords.length} records in ${targetIndex}`);
}

function transformRecord(record: any): Record<string, any> {
  return {
    objectID: record.id || record._id,
    ...record,
    // Remove Elasticsearch-specific fields
    _id: undefined,
    _source: undefined,
    _score: undefined,
  };
}

Step 6: Traffic Shifting (Strangler Fig)

// Gradually shift traffic from old engine to Algolia
function getSearchAdapter(): SearchAdapter {
  const algoliaPercent = parseInt(process.env.ALGOLIA_TRAFFIC_PERCENT || '0');

  if (Math.random() * 100 < algoliaPercent) {
    return new AlgoliaAdapter(process.env.ALGOLIA_APP_ID!, process.env.ALGOLIA_ADMIN_KEY!);
  }

  return new ElasticsearchAdapter(process.env.ES_URL!);
}

// Deployment steps:
// Week 1: ALGOLIA_TRAFFIC_PERCENT=10   (canary)
// Week 2: ALGOLIA_TRAFFIC_PERCENT=50   (half traffic)
// Week 3: ALGOLIA_TRAFFIC_PERCENT=100  (full cutover)
// Week 4: Remove old adapter code

Step 7: Validation

// Compare results between old and new engines
async function validateMigration(queries: string[]) {
  const old = new ElasticsearchAdapter(process.env.ES_URL!);
  const algolia = new AlgoliaAdapter(process.env.ALGOLIA_APP_ID!, process.env.ALGOLIA_ADMIN_KEY!);

  for (const query of queries) {
    const oldResult = await old.search({ index: 'products', query });
    const algoliaResult = await algolia.search({ index: 'products', query });

    const oldIds = new Set(oldResult.hits.map((h: any) => h.objectID || h.id));
    const algoliaIds = new Set(algoliaResult.hits.map((h: any) => h.objectID));

    const overlap = [...algoliaIds].filter(id => oldIds.has(id)).length;
    const overlapPct = (overlap / Math.max(oldIds.size, 1) * 100).toFixed(0);

    console.log(`"${query}": old=${oldResult.totalHits}, algolia=${algoliaResult.totalHits}, overlap=${overlapPct}%`);
  }
}

Rollback Plan

# Instant rollback: set traffic to 0%
export ALGOLIA_TRAFFIC_PERCENT=0
# Restart services to pick up new env var

# Or if fully cut over: old engine is still running, swap adapter
# Keep old engine data in sync for at least 2 weeks after cutover

Error Handling

Issue	Cause	Solution
Result mismatch	Different ranking algorithms	Tune customRanking and searchableAttributes
Missing records	Transform dropped fields	Add logging to transform, validate counts
Higher latency	Cold Algolia index	Search a few times to warm cache, then benchmark
Filter syntax errors	Elasticsearch query DSL ≠ Algolia filters	Use translation guide above

Resources

• Algolia Migration Guide
• replaceAllObjects
• Strangler Fig Pattern
• Elasticsearch to Algolia

Next Steps

Migration complete. See algolia-prod-checklist for go-live preparation.