Meta

Deploy and configure a search microservice on Elasticsearch from scratch, which quickly and accurately finds products, supports queries by equipment models and codes, and ranks results according to the formula relevance/availability/popularity. The architecture is modular, ready for further expansion (implementation of extensions as separate tasks).

What needs to be done

Infrastructure

• Deploy Elasticsearch and Kibana in Docker (one docker-compose.yml + README with launch steps).

• Configure resources and storage for a large catalog (memory, volume, logs).

• Basic security and health checks.

Data and Indexing

• Design the index structure for: products, models, brands, type of equipment/parts and relationships between them.

• Support fields uk/ru and basic text normalization (without morphology and without typo handling).

• Perform initial indexing of a test sample of 10,000+ items while preserving the relationships "model ↔ compatible products".

• Implement full and partial reindexing (addition/update/deletion).

Search Logic (Base)

• Search for products by: name, description, brand, internal article, manufacturer codes (uk/ru fields).

• Search by equipment model: when entering a model, show compatible spare parts; if there is no exact match — show similar models.

• Typical scenarios that should work "out of the box":

  • “samsung pump”

  • “DC31-00054A”

  • “spare parts for Samsung WF-J125A”

• Ranking (we fix the rule for acceptance):
  final scoring = relevance 50% + availability 25% + popularity 25%; Coefficients should be set in configs so that they can be changed.
  hidden/removed items are penalized; with equal relevance — available items are ranked higher than unavailable; if the type of part is explicitly specified in the query — corresponding products are ranked higher.

• If there is no exact match — show similar results based on the best combination of factors.

• Not included: morphology uk/ru, autocomplete, typo handling, synonyms (separate tasks).

Observability

• Metrics: health-endpoint, request/error counters, p95 latency, % of empty results.

• Dashboard in Kibana for monitoring indices and performance.

Acceptance Criteria

1. Deployment: the service is launched with one command from docker-compose.yml; in README — environment variables and commands.

2. Indexing: test data loaded; relationships "model ↔ compatible products" work; full and partial reindexing available.

3. Search Scenarios: products, models, and codes are correctly searched uk/ru without morphology and autocomplete.

4. Ranking: the formula 50/25/25 is adhered to; hidden/removed items are penalized; in the absence of exact matches, similar items are shown.

5. Performance (demo stand):

   • p95 for searching by products/codes ≤ 200 ms,

   • p95 for results by model and compatibility ≤ 350 ms
     (in the report — VPS configuration and actual measurements).

6. Documentation: README with a description of indices and fields, ranking scheme, examples of API requests/responses, instructions for launching/reindexing, description of the Kibana dashboard.