Azure Chronicles - Mule

Overview

Continuing from the previous post, I now move to the other end of the spectrum, from Raspberry Pi to the cloud. Microsoft Azure cloud has been making impressive strides in the cloud space, especially among the enterprises.

This is an experiment of running the same Mule run time in an Azure cloud server. The server is created in the East US Azure location.

I created an A series server (Extra Small) in Azure using the ARM templates. An A series Extra Small server has 1 CPU core and 07.GB RAM. On this server, the Mule run time is on a JVM with 256 MB heap size.

Then, I then ran some scripts to perform the following - 
  • Install Oracle JDK 8
  • Configure the Mule run time (version 3.9)
  • Deploy a simple REST based API in Mule

Objective

The purpose of this experiment is to test how much load can a very simple REST API in Mule handle, while running on a very low specification server in Azure. I wanted to see how the Mule run time would perform from a performance perspective.

A simple REST API is created using Anypoint Studio. The response of the REST API is dynamic based on the parameters passed to it. This ensures that the response doesn't get cached in any layer - server/network/client.


Once the REST API is deployed in Mule run time in Azure, a sample request is tested from a browser. The JSON response is as below. This sample API is based on the sample RAML available at Mulesoft documentation site.



As the REST API is publicly exposed, the load tests are setup in Blazemeter.



Before starting the load test from Blazemeter, the resource statistics on the server are normal. Press F5 in htop utility to view the processes in a tree structure.


The Blazemeter tests are now kicked off and the test results are impressive. The 90 percentile response time is under 200 ms. For a "read" operation on a REST API, this response time is pretty good. The average throughput is also looking good at 170 hits/second. This test simulates 20 concurrent virtual users hitting the REST API endpoint. For an A0 server, this performance is not bad at all.




During the above test, the load on the CPU has spiked to 100%. The server stats can be viewed from the Azure Portal.


Now, I setup the New Relic agent on the server, restarted the Mule run time and re-ran the same test. The results are now starkly different. The 90 percentile response time has gone up by 5 times, to above 1 second. The average throughput has dropped by more than 50% to 71 hits/second.


While we can see the application statistics in New Relic, there is a warning as well. This is about a circuit breaker being triggered in the New Relic agent.


The circuit breaker gives more details that the JVM heap is under pressure.

Summary - 

  • When running API's or websites in a production environment, there is usually a server cluster / farm that handles all requests. While a Mule run time can technically run on an A0 server, one should probably use at least a Dv3 or a Ev3 VM type in Azure. For details on more Linux VM types, please refer here.
  • New Relic agent has the "circuit breaker" pattern implemented, which is quite thoughtful and helpful.
  • When sizing the memory for a JVM based application, keep in mind the additional CPU/Memory requirements for an agent such as New Relic.
  • If the API's are publicly exposed, evaluate cloud based tools such as Blazemeter. Also, ensure that the cloud based tool and the server are in different clouds, to simulate traffic across the internet. For instance, here the server is in Azure while Blazemeter test runner is in AWS.
  • Potentially, Blazemeter could be used not just for load testing of API's but also as a synthetic monitoring tool.
  • Mule Community edition is a great way to explore the tool. From an enterprise perspective, as of now, multiple deployment options are available. They are - On Premise, Anypoint Cloudhub (SaaS) and Anypoint Runtime Fabric (PaaS).
Location: Bengaluru, Karnataka, India

Comments

Post a Comment

Popular posts from this blog

Meta Prompting for Success

Image
Meta Prompt  Regardless of the Generative AI model used, sometimes the model may refuse to provide a response for a valid prompt. Often the reasons cited are - "I don't know", "I am not trained for the task", "violation of safety/responsible/ethical guardrails". This is sometimes more acutely felt when using Generative AI models for creating images. Overcoming this behavior can be frustrating and time consuming. An approach is to ask the AI model itself to generate prompts so that the prose and nuances of prompting are aligned with how the AI model has been built. This is called meta-prompting . The theory is that the prompts created by the AI model would have better chance of working. Let's dive in to see if and how this meta-prompting works in Gemini .  How to Create a Meta Prompt As with all good prompts, start with a persona/situation that describes the overall context. Next, describe the task in detail. In this case, it is about creating some ...
Read more

Book Review - The Price of Being Fair

Image
Overview  As the FairPrice group in Singapore completes 50 years, "The Price of Being Fair" book is a timely story of it's birth and evolution. The book has been authored by Sue-Ann Chia and Shing Huei Peh. While being true to history, the book has nuggets of "behind the scenes" information, given the time the authors have spent with various leaders at FairPrice. The book is already a best seller according to Kinokuniya book store. At under 200 pages, it makes for a breezy read. The Usual Suspects Having worked in retail industry for over two decades, the travails of  NTUC FairPrice seem familiar to me. A retail CEO once remarked that retail is a pretty "public" industry. There is not much behind which a retail business can hide. As such, retailers often have to follow the aspirations and concerns of the communities and customers they serve. The journey of FairPrice from being an EDLP (everyday low price) retailer to one having multiple formats, busin...
Read more

How to create a big picture

Image
Typically, you need wide view or fisheye lenses to capture a panoramic picture. An alternate method (though may not give the same quality) is to use software to create a panoramic image. There are several software available which you can use to create a panoramic image. The concept is called - "stitching". Some of the software available are - PTGui, Panavue, PAX-it, AutoStitch and so on. I sometimes use AutoStitch to create the panoramas since I have a 50mm lens for now. Each of these software vary in price and features. The added advantage is that you can take the pictures using any kind of camera - digital, phone or SLR. Steps: 1. Visualize the picture you want to take. 2. Use your camera and "divide" your image into rows and columns. An example being - 2 rows and 5 columns. 3. Take 10 pictures to cover your entire image. 4. Use any of the software to "stitch" them together. Here is an example of a panorama created using AutoStitch. I to...
Read more