The Aging Developer

The Aging Developer

for growing old in the software development community
Richard Klein
August 14, 2020

Default HTTP Config Considered Dangerous

Gotchas using http clients with default configuration
israel palacio
Image by israel palacio on Unsplash
3 min read
493 words

An alternatively headline might be: know how your service communicates with it's dependencies. Ideally, a microservice could work in isolation without having to communicate with any other service or dependency. This is not likely the case and HTTP is the most common way that I've seen these services communicating. I've witnessed several problems that have been caused by default configurations being used by these clients. In this article I will explore a couple of these issues and how you can remediate them.

Java DNS Caching

When the JVM resolves hostnames to ip addresses it caches those lookups. Depending on the configuration this might not be refreshed until the JVM is restarted. This can especially be an issue if you are using AWS services where the IP Address may change for a service. I've seen this occurs with aurora failovers and other scenarios. You can prevent this by setting networkaddress.cache.ttl in the $JAVA_HOME/jre/lib/security/ file to some low number of seconds. Amazon recommends no more than 60 seconds.


Maximum Connections Per Route

Creating a tcp connection between services is a time consuming process and consumes lots of resources. This is especially true when TLS is involved. Because of this popular http clients try and reuse the same connection if they are going to the same route. The OkHttp library keeps only a single connection. The Apache HttpClient defaults to two. The Go http package also defaults to two.

This is fine when you are communicating over HTTP/2 because that protocol allows for multiple concurrent requests over a single connection.

Http/1.1 however, can only have a single request over the connection at a time. This can create a Head-of-line (HOL) blocking issue if you have to make multiple concurrent requests to the same dependency. The result of which may cause your service's requests to back up and cause a cascading failure. The most basic way to mitigated this is by increasing the maximum connections per route.

HttpClientBuilder builder = HttpClients.custom()

Request Timeout

A lot of clients by default use the timeout set by the OS. This is true for clients in both Java and Go. This is another possible way to cause cascading failures. The best advice is to configure your client with something other than the default. The example below does this globally for the client, but you can also set these on a per request basis.

RequestConfig requestConfig = RequestConfig.custom()
SocketConfig socketConfig = SocketConfig.custom()
HttpClientBuilder builder = HttpClients.custom()

Other Best Practices

By following the practices above, you can help keep your service healthy. A few other best practices that I'm not going to go into, but can keep your service resilient include: using circuit breakers to prevent cascading failures, using retry designs for those temporary intermittent failures, and using thread pools or executor services for bulk heading and fault isolation.

© 2024 The Aging Developer