Skip to main content
Two tests of the open-source edition: bulk record creation on AWS, and a full read path on a single small server. For Databunker Pro numbers, see Performance.

Test 1: one million records on AWS EKS and RDS

Run on 22 January 2022. Databunker OSS was deployed on AWS with the Terraform configuration and Helm chart from the repository. A single EKS node ran the Databunker container, and a single MySQL RDS instance held the data.

Resource utilization (t3.xlarge)

EKS node utilization

EKS node utilization during the test

RDS MySQL utilization

MySQL RDS db.t3.xlarge utilization during the test

How to reproduce

  1. Deploy the stack with the Terraform configuration for AWS.
  2. Start a separate EC2 instance to run the load script below.
The script runs 100 green threads, and each thread creates 10,000 user records, for one million in total. Each thread uses a requests.Session(), so connections are reused; Databunker supports HTTP keep-alive.
stress-test.py

Test 2: full stack on one 1 GB droplet

Run on 11 April 2021. Every component ran on one DigitalOcean droplet with 1 GB RAM, 1 Intel CPU and a 25 GB NVMe SSD:
  1. Databunker, as a Docker container
  2. MySQL, as a Docker container
  3. A Node.js app server
  4. ApacheBench (ab) generating the load
This measures the whole read path, not Databunker alone. On each request, Databunker loads the user record from MySQL (SELECT), decrypts it, writes an audit event (INSERT), and returns the record to the app server, which prints the user’s email. The Node.js HTTP agent was configured to reuse open sockets. Result: the app server sustained 155 requests per second while loading every response from Databunker, with no failed requests and a 95th-percentile latency of 95 ms.

App server

server.js

Load command and output

Next steps