Skip to content

Latest commit

 

History

History
104 lines (64 loc) · 4.1 KB

README.md

File metadata and controls

104 lines (64 loc) · 4.1 KB
Raw

Meteor performance

Motivation

This repository is the base to measure performance between Meteor 2 and Meteor 3. It serves as a comparison point for various functionalities and helps catch any performance regressions.

This repository can inspire you to adapt the scripts and analyze performance in your applications. We're here to make the process easier, so feel free to share any feedback.

What

This repository includes:

  • Meteor applications for Meteor 2 and 3 to test performance, located in the ./apps folder.
  • Meteor packages for shared and isomorphic code, found in the ./packages folder.
  • Playwright tests to perform actions against the apps, found in ./tests.
  • Artillery configurations for stress testing your server, found in ./artillery.
  • Scripts to monitor performance and log results, found in ./scripts.
  • Logs from local monitoring runs, found in ./logs.
  • Benchmarks from official monitoring runs, found in ./benchmarks.

Requirements

  • Unix System
  • Node 20.x version

How

Running locally

To run a stress test on your machine:

npm install
./scripts/monitor.sh <app> <artillery-script> <log-context>
  • <app>: The app folder name within ./apps to stress test.
  • <artillery-script>: The artillery configuration within ./artillery.
  • <log-context>: A name to prefix the logs generated within ./logs.

The process will take some time, and the logs are updated live at ./logs.

Your machine might struggle with the default artillery configuration, but it should still reveal performance differences between the tests. Adjust the configuration as needed by learning from artillery.io options.

You can analyze performance using a Meteor checkout. This allows you to quickly measure the impact of ongoing changes and ensure consistent or improved performance.

METEOR_CHECKOUT_PATH=<path-to-meteor-checkout> ./scripts/monitor.sh <app> <artillery-script> <log-context>

Running remotely

Currently, only Meteor staff can fully manage the entire process for the tasks app.

To run a stress test on a galaxy server:

  • Prepare a new application container on Galaxy
  • Prepare secrets on .env.prod file
    • MONGO_URL_TASKS_3_X and MONGO_URL_TASKS_2_X. Url to connect to the MongoDB instance for each application.
    • MONGO_VERSION Version of MongoDB used to pre-clean your environment
    • REMOTE_URL_TASKS_3_X REMOTE_URL_TASKS_2_X. Public endpoint to connect to your application.

Run the following monitoring command to start benchmarking:

npm install
./scripts/monitor-remote.sh <app> <artillery-script> <log-context>

This will set up the remote context needed to run the playwright script.

Remote stress testing helps gather extra metrics from MontiAPM for deeper performance analysis.

For each benchmark, we recommend using a new Galaxy container to ensure a clean environment, preventing any interference with your performance results.

Deploy

The apps are deployed to Galaxy to benefit of further analysis with MontiAPM tool.

To deploy each app after changes:

./scripts/deploy.sh <app>

Current apps are accessible at:

Benchmarks

Meteor 2 vs 3

Meteor 3 is in average ~28% faster, uses ~10% less CPU and ~16% less of RAM in a non-reactive scenario.

Meteor 3 is in average ~equal on time, uses ~18% less CPU and ~10% more of RAM in a reactive scenario. However, Meteor 3 handles fewer connections per minute, indicating a regression.

More details on this benchmark can be found at ./benchmarks/meteor2.16-vs-3.0.1, where we observed a performance regression in Meteor 3 related to reactivity.

As part of the Meteor 3.0.3 release, we have found that disabling compression can improve performance. More information is at ./benchmarks/meteor2.16-vs-3.0.3-disable-compression.

We will keep investigating how to address the regression that persists in Meteor 3 with reactivity.