haukevb / lucene-serverless

A proof-of-concept serverless full-text search solution built with Apache Lucene and Quarkus framework.

Lucene Serverless

This project demonstrates a proof-of-concept serverless full-text search solution built with Apache Lucene and Quarkus framework.

✔️ No servers

✔️ No fixed costs

✔️ Low (250-300ms) cold starts

⚠️ Better deletion policy is required. Right now old segment files are not deleted as a simple workaround to handle concurrent reads and writes

ℹ️ Cost can be controlled via several factors:

• EFS' Elastic throughput mode cost more $$, but scales better up and down
• Lambda's Provisioned Concurrency provides faster, more consistent first request times but costs more $$
• Lambda memory is kept ~256mb, tune based on index size and observed speeds. Note: vCPU is allocated proportional to memory.

Please note that the project is not ready for production since I haven't tested it under a prolonged load and to be honest interfaces need to be nicer.

Read the blog post about it here

Prerequisites

• Serverless framework >= 1.56.1
• AWS account

Run it

1. Replace region, vpc id and subnets in the serverless.yml file

2. Deploy the stack sls deploy

3. Don't forget to remove it if you're not planning to use it sls remove

Index a document

URL: https://<api-id>.execute-api.<region>.amazonaws.com/dev/index

HTTP method: POST

Example request body:

{
  "indexName": "books",
  "documents": [
    {
      "name": "The Foundation",
      "author": "Isaac Asimov"
    }
  ]
}

Query documents

URL: https://<api-id>.execute-api.<region>.amazonaws.com/dev/query

HTTP method: POST

Example request body:

{
   "indexName": "books",
   "query": "author:isaac"
}

Example response body:

{
  "totalDocuments": "1",
  "documents": [
    {
      "author": "Isaac Asimov",
      "name": "The Foundation"
    }
  ]
}

Build native image

./mvnw clean package