Top Two Features of Self Healing Microservices
By Martin Goodwell

Microservices-based environments are more complex than their monolithic counterparts. To operate microservices environments with the same level of convenience that you’ve come expect from operating self-contained monolithic application environments, you need to have the right tools in place and rely on best practices that will keep your microservices healthy.

We’ve noticing an increasing number of microservices environments deployed by our customers, and the trend only seems to be increasing. We recently asked some of these customers about their experiences. Most of their responses were in line with expectations. Two issues however caught us by surprise.

The two issues that surprised us
While increased complexity and the proliferation of services is a challenge that most customers contend with, it’s the orchestration layer that seems to cause the most trouble for customers. The complexity of microservices environments requires automated operation and self-healing capabilities. This means that some sort of orchestration layer is required.

Automated orchestration
The first issue for many customers is that they haven’t even considered using a tool for orchestration. Because they don’t have an orchestration layer in place they’ve been frustrated from the beginning. After a few deployments, maintaining different instances of services simply became too complicated.

This is too bad because plenty of good open-source tools are available to facilitate the operation of microservices. For example, Netflix OSS—one of the toolsets most commonly used by our customers—provides a number of utilities that Netflix has developed over the years.

Of the dozens of available Netflix OSS tools and libraries, Eureka service registry and Hystrix circuit breaker are the most popular and likely also the easiest to use (Spring Cloud is a real easy to getting started with those technologies quickly, by the way). The service registry allows your microservices to register themselves upon startup. Any client that intends to access a microservice can look up the available endpoint addresses via Eureka. This reduces the risk of accessing an unavailable service to near zero.

Netflix OSS Eureka and Hystrix at a glance

Circuit breakers keep your business going
Hystrix circuit breaker makes your service calls more resilient by keeping track of each endpoint’s status. Normally you face expensive request timeouts when an endpoint becomes unavailable. Hystrix saves you from such timeouts by “breaking” the connection to the endpoint (this is why Hystrix is called a “circuit breaker”). It then reports that the service as unavailable so that subsequent requests don’t run into the same timeouts. Hystrix then continues to poll the service in the background to see when it’s available again.

With the risk of accessing unavailable services reduced to near-zero and expensive timeouts reduced to a minimum, your website keeps your content online and your visitors happy. You probably already know how much money you lose with each additional second of wait time, so using a service registry and a circuit breaker is a no-brainer in my opinion.

The second problem reported by our customers really caught us by surprise: a large number of customers report that their orchestration layers cause increased troubleshooting effort when problems are encountered.

Self-healing systems impact troubleshooting?
We were puzzled by this at first because the orchestration layer is supposed to keep an environment healthy and self-healing—quite the opposite of what our customers were reporting. Once we dug into this issue a little deeper however, the cause of the additional troubleshooting effort became obvious. These customers were initially looking for problems only in their services, not the orchestration layer itself. So when the orchestration layer did fail, customers spent considerable time eliminating the possible root-causes in their own code before getting around to investigating the orchestration layer.

These experiences taught us and our customers an important lesson:

Orchestration layers need to be monitored
You need a proper monitoring tool in place to know when your orchestration layer fails. You’re probably used to monitoring your database connection pools, your service queues, and of course your API performance metrics. You need to apply the same monitoring strategy to your orchestration layer.

Eureka and Hystrix are well-crafted pieces of code that allow for monitoring of all the important orchestration-layer related metrics.

Increased service lookup times and failing connections indicate that your orchestration layer need maintenance.

The lessons we gleaned from our customers’ experiences leads us to the conclusion that monitoring of the orchestration layer should be an integral feature of all full-stack monitoring tools. Service registries, circuit breakers, and other orchestration tools should be part of all newly created environments, as should database connection pools and messaging queues.

Short interruptions are handled automatically by the orchestration layer. Ongoing problems indicate that the orchestration layer is in trouble.

Keep control over your environment
Ideally, your orchestration layer is a simple set of tools that hardly ever fails. Unfortunately, things that hardly ever fail are not the initial focus of troubleshooting. So, when your orchestration layer fails, you may have a hard time learning about it quickly.

Orchestration-layer metrics can be a great indicator of looming problems. In cases of bad performance, take a look at your orchestration layer metrics. The orchestration layer may not be your hottest troublespot, but when when there is trouble there, you’ll be glad to know about it.

Do you know your metrics?
Is your environment based on microservices? Do you actively use service registries and circuit breakers? If so, do you know how much time service lookups consume? Do you know how often fallback mechanisms need to jump in? And do you know if your environment is truly self-healing?

If you have Netflix OSS components in use, you can add Netflix OSS monitoring by downloading the plugin file from our GitHub repo at JMX-Extensions/Netflix OSS extensions/plugin.json. Activating it is just a simple matter of uploading the file to your Dynatrace Ruxit environment.

In one of our upcoming releases, we will include this feature to be available out-of-the-box. Ruxit provides immediate value to you out-of-the-box. Haven’t tried Dynatrace Ruxit yet? Take the free trial and see how resilient your environement is. You may be surprised at how Ruxit’s auto-detection and zero-configuration have simplified monitoring setup.

The post Top 2 features of self-healing microservices appeared first on #monitoringlife.